Refactor position precision handling to honor explicit channel settings and prevent location leaks

Add Ethernet configuration to platformio.ini for ThinkNode variants
Develop protos should be on develop
2026-10-04 03:18:10 +10:00 · 2026-05-20 08:39:49 -05:00 · 2026-05-20 07:12:43 -05:00 · 2026-05-19 06:57:05 -05:00 · 2026-05-19 05:37:01 -05:00 · 2026-05-19 05:23:01 -05:00
609 changed files with 44378 additions and 4713 deletions
@@ -24,7 +24,12 @@ Call the meshtastic MCP tool bundle and format a structured health report for on
   - `mcp__meshtastic__get_config(section="lora", port=<p>)` — region, preset, channel_num, tx_power, hop_limit.
   - Optionally, if the device seems unhappy (fails to connect, `num_nodes==1` when ≥2 are plugged in, missing firmware*version), open a short firmware log window: `mcp__meshtastic__serial_open(port=<p>, env=<inferred-env>)`, wait 3s, `serial_read(session_id=<s>, max_lines=100)`, `serial_close(session_id=<s>)`. The env should be inferred from the VID map in `mcp-server/run-tests.sh` (nrf52 → rak4631, esp32s3 → heltec-v3) unless `MESHTASTIC_MCP_ENV*<ROLE>` is set.

-4. **Render per-device report** as:
+4. **Hub health** (call once, not per-device): `mcp__meshtastic__uhubctl_list()` — enumerates every USB hub the host can see. Note which hubs advertise `ppps=true` and which hub hosts each Meshtastic device (cross-reference by VID). Flag it in the report if:
+   - No hub advertises PPPS → `tests/recovery/` can't run on this setup; hard-recovery via `uhubctl_cycle` isn't available.
+   - A Meshtastic device is on a non-PPPS hub → note it; operator may want to move the device to a PPPS hub to unlock auto-recovery.
+   - `uhubctl_list` raises `ConfigError: uhubctl not found` → just say `uhubctl not installed` in the report; don't treat as a fault.
+
+5. **Render per-device report** as:

   ```text
   [nrf52 @ /dev/cu.usbmodem1101]      fw=2.7.23.bce2825, hw=RAK4631
@@ -33,20 +38,28 @@ Call the meshtastic MCP tool bundle and format a structured health report for on
     tx_power    : 30 dBm, hop_limit=3
     peers       : 1 (esp32s3 0x433c2428, pubkey ✓, SNR 6.0 / RSSI -24 dBm)
     primary ch  : McpTest
+     hub         : 1-1.3 port 2 (PPPS, uhubctl-controllable)
     firmware    : no panics in last 3s; NodeInfoModule emitted 2 broadcasts
   ```

-   Keep it scannable. If a field is missing or abnormal (no pubkey for a known peer, region=UNSET, num_nodes inconsistent with the hub), flag it inline with a short `⚠︎ <one-line reason>`.
+   Keep it scannable. If a field is missing or abnormal (no pubkey for a known peer, region=UNSET, num_nodes inconsistent with the hub, device on non-PPPS hub), flag it inline with a short `⚠︎ <one-line reason>`.

-5. **Cross-device correlation** (only when >1 device is inspected):
+6. **Cross-device correlation** (only when >1 device is inspected):
   - Do both sides see each other in `nodesByNum`? If one does and the other doesn't, that's asymmetric NodeInfo — flag it.
   - Do the LoRa configs match? (region, channel_num, modem_preset should all agree; mismatch = no mesh)
   - Do the primary channel NAMES match? Mismatch = different PSK = no decode.

-6. **Suggest next actions only for specific, recognisable failure modes**:
+7. **Recorder slice (cheap, always available).** The mcp-server runs an autouse log recorder that's been collecting from every connected device. Pull two short slices to surface anything weird that's already happened:
+   - `mcp__meshtastic__logs_window(start="-2m", level="WARN|ERROR|CRIT", max_lines=20)` — recent firmware errors. If empty, say "no recent errors"; don't manufacture concern.
+   - `mcp__meshtastic__telemetry_timeline(window="1h", field="free_heap", max_points=60)` — heap trend. If `slope_per_min < -50`, flag it and recommend `/leakhunt window=6h` for a deeper read; otherwise just note the current free heap.
+   - If `recorder_status` shows `running:false` or `files.telemetry.last_ts` is null, note "recorder has no telemetry yet — enable `set_debug_log_api(True)` to populate" and skip this step gracefully.
+
+8. **Suggest next actions only for specific, recognisable failure modes**:
   - Stale PKI pubkey one-way → "run `/test tests/mesh/test_direct_with_ack.py` — the retry + nodeinfo-ping heals this in the test path."
   - Region mismatch → "re-bake one side via `./mcp-server/run-tests.sh --force-bake`."
-   - Device unreachable → point at touch_1200bps + the CP2102-wedged-driver note in run-tests.sh.
+   - Device unreachable, reachable via DFU → `touch_1200bps(port=...)` + `pio_flash`. If not even DFU responds AND the device is on a PPPS hub, escalate to `uhubctl_cycle(role=..., confirm=True)`.
+   - CP2102-wedged-driver on macOS → see the note in `run-tests.sh`.
+   - Heap slope strongly negative → "run `/leakhunt window=6h` for a full timeline + classification."

 ## What NOT to do

@@ -0,0 +1,103 @@
+---
+description: Hunt for memory leaks (and other slow degradations) by reading the persistent recorder's heap timeline + log slice over a window
+argument-hint: [window=1h] [field=free_heap] [variant=local]
+---
+
+<!-- markdownlint-disable MD029 -->
+
+# `/leakhunt` — read the recorder, classify a memory leak
+
+Use the always-on recorder (`mcp-server/.mtlog/`) to read a heap timeline plus the matching log slice and produce a one-page verdict: **steady / slow leak / fragmentation / OOM-imminent**. No firmware changes, no special build flags — the LocalStats telemetry packet that the firmware already broadcasts every ~60 s carries `heap_free_bytes` and `heap_total_bytes`.
+
+## Two signal paths — pick the right one
+
+| Path                  | Build flag       | Cadence        | Per-thread attribution | Cost                      |
+| --------------------- | ---------------- | -------------- | ---------------------- | ------------------------- |
+| LocalStats packet     | (default)        | ~60 s          | No                     | Free — always on          |
+| `[heap N]` log prefix | `-DDEBUG_HEAP=1` | every log line | Yes (Thread X leaked)  | Bigger flash + log volume |
+
+Both feed the same `telemetry_timeline(field="free_heap")` query — when DEBUG_HEAP is on, the recorder synthesizes telemetry rows from log prefixes (tagged `source: debug_heap`), so a single timeline call gets whichever signal is available. **For a slow leak diagnosis, the default path is plenty** (60 s cadence over 6 h = 360 points; linear regression over that nails sub-100-byte/min slopes). **DEBUG_HEAP is for attribution** — when the slope is real and you need to know which thread is leaking.
+
+## What to do
+
+1. **Parse `$ARGUMENTS`**: optional `window` (default `1h`, accepts `30m`/`6h`/`-3d`/etc.), optional `field` (default `free_heap`; alternates: `total_heap`, `battery_level`, anything in the LocalStats variant), optional `variant` (default `local`; alternates: `device`, `environment`, `power`, `airQuality`, `health`).
+
+2. **Verify the recorder is alive** — call `mcp__meshtastic__recorder_status`. Check:
+   - `running == True`
+   - `files.telemetry.lines > 0` (at least one telemetry packet recorded — if zero, the device hasn't broadcast LocalStats yet OR `set_debug_log_api` has never been on; tell the operator to run `mcp__meshtastic__set_debug_log_api(enabled=True)` and wait one device-update interval)
+   - `files.telemetry.last_ts` within the last 5 minutes (if older, the device is silent — log that, not "leak detected")
+
+3. **Detect whether DEBUG_HEAP is active** — `mcp__meshtastic__logs_window(start="-2m", grep=r"\\[heap \\d+\\]", max_lines=3)`. If any line matches, the firmware has the prefix → DEBUG_HEAP is on, expect higher-cadence data and `heap_event` rows. If zero matches over the last 2 minutes, you're on the LocalStats-only path.
+
+4. **Pull the timeline** — `mcp__meshtastic__telemetry_timeline(window=$window, variant=$variant, field=$field, max_points=200)`. Read:
+   - `samples` — how many raw points contributed
+   - `min`, `max` — total swing
+   - `slope_per_min` — units per minute (linear regression over the whole window)
+
+5. **Pull the log context for the same window** — `mcp__meshtastic__logs_window(start="-${window}", grep="Heap status|leaked heap|freed heap|out of memory|Alloc an err|panic|abort", max_lines=200)`. These are the strings the firmware emits when something memory-related happens (`DEBUG_HEAP` builds emit `"Heap status:"` and `"leaked heap"` lines; production builds emit `"Alloc an err"` on failure and `"out of memory"` on OOM).
+
+6. **Pull marker events** so we know if the operator labeled phases — `mcp__meshtastic__events_window(start="-${window}", kind="mark|connection_lost|connection_established")`. If a `connection_lost` overlaps a sharp drop, that's not a leak; that's a reboot.
+
+6a. **(DEBUG_HEAP only) Per-thread attribution** — `mcp__meshtastic__logs_window(start="-${window}", grep="leaked heap", max_lines=200)`. Each row has a structured `heap_event` field with `{kind, thread, before, after, delta}`. Aggregate by thread: sum the `delta` over the window per thread name. The thread with the largest cumulative negative delta is your suspect. Note the count too — a thread with 50× small leaks is different from 1× big leak.
+
+7. **Classify** based on what the data says, NOT on what you wish it said. Use these rules in order:
+   - **Insufficient data** (< 5 samples): say so. Suggest a longer window or longer wait. Stop.
+   - **Reboot mid-window**: if any `connection_lost` event is present AND `free_heap` jumped UP at that timestamp, the device rebooted. Note it; pre-reboot trend may be a leak but you only have part of the curve.
+   - **OOM-imminent**: any `Alloc an err=` or `out of memory` line in the log slice. This trumps everything; flag urgently.
+   - **Slow leak**: `slope_per_min < -50` AND `max - min > 1000` AND no reboot. The heap is monotonically (or near-monotonically) declining. Estimate time-to-zero: `min / -slope_per_min` minutes. Surface it.
+   - **Fragmentation suspect**: `slope_per_min` close to zero (|x| < 50) BUT min trends down across the window AND the log slice shows `Alloc an err` warnings WITHOUT total OOM. Means free total is OK but largest contiguous block is shrinking. Recommend a `DEBUG_HEAP` build to confirm.
+   - **Steady**: |slope_per_min| < 50, no error lines. Heap is fine.
+   - **Recovery curve**: slope is POSITIVE — heap recovered. Either a workload completed or GC fired. Note it; not a leak.
+
+8. **Report**:
+
+   ```text
+   /leakhunt window=6h field=free_heap variant=local
+   ────────────────────────────────────────────────────
+   recorder        : running, telem last_ts 8s ago
+   build           : DEBUG_HEAP=ON (per-line prefix detected)
+   samples         : 14,200 over 6h (cadence ~1.5s, log-line synth)
+   free_heap       : min 92,344 / max 124,008 / range 31,664
+   slope           : -82 bytes/min (negative — heap declining)
+   reboots         : none in window
+   OOM events      : none
+   error lines     : 3× "Alloc an err=ESP_ERR_NO_MEM" at +4h12m, +5h08m, +5h44m
+   thread leaks    : (DEBUG_HEAP) MeshPacket -3,124 B over 18 events
+                                  Router       -1,408 B over 4 events
+                                  others       -240 B
+   verdict         : SLOW LEAK — primary suspect MeshPacket thread
+   est. time-to-OOM: ~1,127 min (~18.8 h) at current slope
+   evidence        : (3 log line citations with uptimes)
+   ```
+
+   Then: **what to do next.**
+   - SLOW LEAK, **DEBUG_HEAP off** → recommend rebuilding with the flag and re-running this skill. Concrete one-liner the operator can copy:
+     ```text
+     mcp__meshtastic__build(env="<env>", build_flags={"DEBUG_HEAP": 1})
+     mcp__meshtastic__pio_flash(env="<env>", port="<port>", confirm=True)
+     ```
+     After flash, set debug_log_api back on and wait one window; re-run `/leakhunt`.
+   - SLOW LEAK, **DEBUG_HEAP on** → cite the top-leaking thread name from step 6a. Point at the corresponding source file (`grep -rn "ThreadName(\"<name>\")" src/`); the operator decides what to fix.
+   - FRAGMENTATION SUSPECT → propose pre-allocating any per-packet buffers; or rebuilding with `CONFIG_HEAP_TASK_TRACKING=y` on ESP32 to see who's holding the largest blocks.
+   - OOM-IMMINENT → flag for immediate attention; don't wait for the next telemetry interval.
+   - STEADY → say so; stop. Don't invent problems.
+
+## What NOT to do
+
+- Don't assume a leak from a single dip. LocalStats fires every ~60 s and the firmware naturally allocates+frees on each broadcast cycle; one packet sees the trough. Look at the slope, not the deltas.
+- Don't recommend code changes. This skill diagnoses; the operator decides what to fix.
+- Don't enable `set_debug_log_api` automatically — if it's off, telemetry isn't reaching pubsub anyway, and the recorder will be empty. Tell the operator to flip it on and wait, then re-run.
+- Don't run heavy workloads to "trigger the leak." The recorder is passive; we read what's there.
+
+## Companion: `mark_event` for stress runs
+
+If the operator wants to test under stimulus (e.g. blast 50 broadcasts and see what the heap does), they can frame the experiment with markers:
+
+```text
+mark_event("burst-start")
+… run the workload …
+mark_event("burst-end")
+/leakhunt window=15m
+```
+
+The markers land in both `events.jsonl` and `logs.jsonl`, so the report can show "free_heap dipped 8 KB during the burst window, recovered to baseline within 2 LocalStats cycles" → not a leak.
@@ -3,6 +3,8 @@ description: Re-run a specific test N times in isolation to triage flakes, diff
 argument-hint: <test-node-id> [count=5]
 ---

+<!-- markdownlint-disable MD029 -->
+
 # `/repro` — flakiness triage for one test

 Re-run a single pytest node ID N times in isolation, track pass rate, and surface what's _different_ in the firmware logs between the passing attempts and the failing ones. Turns "it's flaky, I guess" into "it fails when X, passes when Y."
@@ -40,11 +42,14 @@ Re-run a single pytest node ID N times in isolation, track pass rate, and surfac

   Surface the top 3 differences as a "passes when / fails when" table. Don't dump full logs — pull specific lines with uptime timestamps.

+5a. **Archive recorder slices per attempt** (no extra device interaction; the recorder runs autouse). Right after each attempt finishes, capture its `(start_ts, end_ts)` and call `mcp__meshtastic__recorder_export(start=<start>, end=<end>, dest_dir="mcp-server/tests/repro_artifacts/<safe-test-id>/attempt_<n>/")`. This drops a `logs.jsonl`, `telemetry.jsonl`, `packets.jsonl`, and `events.jsonl` snapshot scoped to the attempt window. Use these for cross-attempt diffs in step 5: `jq '.line' logs.jsonl` is faster than re-running the test, and the telemetry slice lets you compare heap behavior across attempts.
+
 6. **Classify the flake** into one of:
   - **LoRa airtime collision** → pass rate improves with fewer concurrent transmitters; propose a `time.sleep` gap or retry bump in the test body.
   - **PKI key staleness** → fails on first attempt, passes after self-heal; existing retry loop in `test_direct_with_ack.py` handles this.
   - **NodeInfo cooldown** → `Skip send NodeInfo since we sent it <600s ago` in fail-only logs; needs `broadcast_nodeinfo_ping()` warmup.
-   - **Hardware-specific** (one direction fails, other passes; one device's firmware is older; driver wedged) → specific recovery pointer.
+   - **Hardware-specific** (one direction fails, other passes; one device's firmware is older; driver wedged) → specific recovery pointer. For a device that's wedged past `touch_1200bps`, the next escalation is `uhubctl_cycle(role=..., confirm=True)` to hard-power-cycle its hub port (requires `uhubctl` installed).
+   - **Device went dark mid-run** → fails from some attempt onward, never recovers, firmware log stops arriving. Almost always hardware: a Guru crash + frozen CDC. Hard-power-cycle via `uhubctl_cycle(role=..., confirm=True)` before the next iteration; if that also fails, escalate to replug.
   - **Genuinely unknown** → say so; don't invent a root cause.

 7. **Report back** with:
@@ -19,16 +19,21 @@ Run `mcp-server/run-tests.sh` and make sense of the output so the operator doesn

 2. **Read the pre-flight header.** First ~6 lines print the detected hub (role → port → env). If that line reads `detected hub : (none)`, the wrapper will narrow to `tests/unit` only — say so explicitly in your summary so the operator knows hardware tiers were skipped.

-3. **On pass**: one-line summary of the form `N passed, M skipped in <duration>`. Don't enumerate the 52 test names — the user can read those. Do mention if any test was SKIPPED for a NON-placeholder reason (e.g. "role not present on hub" is worth flagging).
+3. **On pass**: one-line summary of the form `N passed, M skipped in <duration>`. Don't enumerate the test names — the user can read those. Do mention any SKIPPED tests and name the cause:
+   - `"role not present on hub"` → device unplugged; operator knows to reconnect.
+   - `"firmware not baked with USERPREFS_UI_TEST_LOG"` → tests/ui skipped because the macro isn't in firmware yet; suggest `--force-bake`.
+   - `"uhubctl not installed"` → tests/recovery + peer-offline skipped; suggest `brew install uhubctl` / `apt install uhubctl`.
+   - `"no PPPS-capable hubs detected"` → tests/recovery skipped because the hub doesn't support per-port power; the tier will never run on that setup.
+   - `"opencv-python-headless is not installed"` → tests/ui auto-deselected by run-tests.sh; suggest `pip install -e 'mcp-server/.[ui]'`.

-4. **On failure**: for every FAILED test, open `mcp-server/tests/report.html` and extract the `Meshtastic debug` section for that test. pytest-html embeds the firmware log stream + device state dump there; the 200-line firmware log tail is usually enough to explain the failure. Summarise: which test, one-line assertion message, the firmware log lines that matter (things like `PKI_UNKNOWN_PUBKEY`, `Skip send NodeInfo`, `Error=`, `Guru Meditation`, `assertion failed`).
+4. **On failure**: for every FAILED test, open `mcp-server/tests/report.html` and extract the `Meshtastic debug` section for that test. pytest-html embeds the firmware log stream + device state dump there; the 200-line firmware log tail is usually enough to explain the failure. Summarise: which test, one-line assertion message, the firmware log lines that matter (things like `PKI_UNKNOWN_PUBKEY`, `Skip send NodeInfo`, `Error=`, `Guru Meditation`, `assertion failed`). For UI-tier failures also glance at `mcp-server/tests/ui_captures/<session>/<test>/transcript.md` — it records each step's frame + OCR.

 5. **Classify the failure** as one of:
   - **Transient/flake**: LoRa collision, timing-sensitive assertion, first-attempt NAK + successful retry pattern. Propose `/repro <test_node_id>` to confirm.
-   - **Environmental**: device unreachable, port busy, CP2102 driver wedged. Suggest the specific recovery (replug USB, `touch_1200bps`, check `git status userPrefs.jsonc`).
+   - **Environmental**: device unreachable, port busy, CP2102 driver wedged. Suggest the specific recovery in escalation order: (a) replug USB, (b) `touch_1200bps(port=...)` + `pio_flash` for nRF52 DFU, (c) `uhubctl_cycle(role="nrf52", confirm=True)` when a device is fully wedged past DFU (needs `uhubctl` installed — `baked_single`'s auto-recovery hook does this once automatically). Also check `git status userPrefs.jsonc`.
   - **Regression**: same assertion fails repeatedly, firmware log shows a new/unusual error. Surface the diff between expected and observed, identify the module likely responsible.

-6. **Never run destructive recovery automatically.** If a failure looks like it needs a reflash, factory*reset, or USB replug, \_describe what to do* — don't execute. The operator decides.
+6. **Never run destructive recovery automatically.** If a failure looks like it needs a reflash, factory*reset, `uhubctl_cycle`, or USB replug, \_describe what to do* — don't execute. The operator decides.

 ## Arguments handling

@@ -8,17 +8,21 @@
  "features": {
    "ghcr.io/devcontainers/features/python:1": {
      "installTools": true,
-      "version": "3.14"
+      "version": "3.13"
    }
  },
  "customizations": {
    "vscode": {
      "extensions": [
        "ms-vscode.cpptools",
-        "platformio.platformio-ide",
+        "Jason2866.esp-decoder",
+        "pioarduino.pioarduino-ide",
        "Trunk.io"
      ],
-      "unwantedRecommendations": ["ms-azuretools.vscode-docker"],
+      "unwantedRecommendations": [
+        "ms-azuretools.vscode-docker",
+        "platformio.platformio-ide"
+      ],
      "settings": {
        "extensions.ignoreRecommendations": true
      }
@@ -75,11 +75,11 @@ body:
  - type: checkboxes
    id: mui
    attributes:
-      label: Is this bug report about any UI component firmware like InkHUD or Meshtatic UI (MUI)?
+      label: Is this bug report about any UI (https://meshtastic.org/docs/configuration/device-uis/) component firmware?
      options:
-        - label: Meshtastic UI aka MUI colorTFT
-        - label: InkHUD ePaper
-        - label: OLED slide UI on any display
+        - label: Meshtastic UI aka MUI
+        - label: InkHUD
+        - label: BaseUI

  - type: input
    id: version
@@ -13,6 +13,7 @@ Meshtastic is an open-source LoRa mesh networking project for long-range, low-po
 - **RP2040/RP2350** - Raspberry Pi Pico variants
 - **STM32WL** - STM32 with integrated LoRa
 - **Linux/Portduino** - Native Linux builds (Raspberry Pi, etc.)
+- **macOS native** - Headless `meshtasticd` on Apple Silicon / x86_64; see `variants/native/portduino/platformio.ini` for Homebrew prereqs + CH341 LoRa setup

 ### Supported Radio Chips

@@ -70,6 +71,163 @@ PKI (Public Key Infrastructure) messages have special handling:
 - Accepted on a special "PKI" channel
 - Allow encrypted DMs between nodes that discovered each other on downlink-enabled channels

+## Encryption & Key Management
+
+Meshtastic packets on the air are typically encrypted one of two ways: the **per-channel symmetric** layer (AES-CTR with a shared PSK) for broadcasts and channel traffic, and the **per-peer PKI** layer (X25519 ECDH → AES-256-CCM) for direct messages and remote admin. A channel with a 0-byte PSK (or Ham mode, which wipes PSKs) transmits cleartext — see the size table below. Both are implemented in `src/mesh/CryptoEngine.cpp`; the send/receive dispatch lives in `src/mesh/Router.cpp`; admin authorization lives in `src/modules/AdminModule.cpp`.
+
+### High-level model
+
+- **Channels** are symmetric rooms: anyone with the PSK can read any message on the channel. Channel 0 is the "primary" channel and ships with the short-form default PSK on factory devices, forming the public mesh most users join. (The LoRa modem preset `LONG_FAST` lives on `config.lora.modem_preset` and is an independent field — don't conflate "channel 0 default PSK" with the modem preset name.)
+- **DMs** addressed to a single node require PKI so that other holders of the channel PSK can't read them. Outside Ham mode, Meshtastic does not fall back to channel-symmetric encryption when the destination public key is unknown.
+- **Remote admin** is a DM carrying an `AdminMessage`. The receiver only acts on it if the sender's public key is on its allowlist (`config.security.admin_key[0..2]`).
+- **Ham mode** (`owner.is_licensed=true`, where `owner` is the local `meshtastic_User` record) disables PKI entirely and sends cleartext — FCC Part 97 prohibits encryption on amateur bands.
+- **No ratchet, no session.** Every packet is encrypted from scratch — a stateless design that matches the high-loss, store-and-forward nature of LoRa.
+
+### Symmetric channel encryption (AES-CTR)
+
+`CryptoEngine::encryptPacket` / `decrypt` / `encryptAESCtr` in `src/mesh/CryptoEngine.cpp`.
+
+- **Cipher**: AES-CTR, AES-128 or AES-256 depending on key length. Same routine in both directions (CTR is a stream cipher, so encrypt == decrypt).
+- **Key**: `ChannelSettings.psk` bytes. Size semantics:
+  - **0 bytes** → no encryption, cleartext on the air
+  - **1 byte** → short-form index into the well-known `defaultpsk[]` in `src/mesh/Channels.h`. Index 0 = cleartext; 1 = defaultpsk unchanged; 2..255 = defaultpsk with its last byte incremented by (index − 1). This is what the CLI's `--ch-set psk default` produces.
+  - **16 bytes** → raw AES-128 key
+  - **32 bytes** → raw AES-256 key
+  - **2..15 bytes** → zero-padded to 16 and used as AES-128 (with a warn log); **17..31 bytes** → zero-padded to 32 and used as AES-256 (with a warn log). Defensive fallback for malformed PSK input, not something to rely on.
+- **Nonce (128 bit)**: `packet_id` (u64 LE) ‖ `from_node` (u32 LE) ‖ `block_counter` (u32, starts at 0). Built in `CryptoEngine::initNonce`.
+- **No AEAD**: channel packets carry no MAC, so the channel-hash byte is not an integrity or authenticity check. `Channels::getHash` is a 1-byte XOR-derived hint over the channel name bytes and PSK bytes that helps receivers pick a candidate channel/PSK for decryption. Because it is only a small hint and collisions are easy to find, it should be described purely as a PSK-selection aid, not as a security filter an attacker cannot bypass.
+- **Channel 0 is special in one way only**: it's the channel the Router attempts PKI decryption on before falling through to AES-CTR. Non-zero channels always go straight to AES-CTR.
+
+### PKI encryption for DMs (X25519 ECDH + AES-256-CCM)
+
+`CryptoEngine::encryptCurve25519` / `decryptCurve25519` in `src/mesh/CryptoEngine.cpp`.
+
+- **Keypair**: Curve25519 (aka X25519), 32-byte public + 32-byte private. Stored in `config.security.public_key` / `private_key`; the public half is mirrored into `owner.public_key` so it rides along in NodeInfo broadcasts and propagates through the mesh like any other identity field.
+- **Key generation** (`generateKeyPair`): stirs `HardwareRNG::fill()` (64 B from platform TRNG when available), the 16-byte `myNodeInfo.device_id`, and a call to `random()` into the rweather/Crypto library's software RNG, then `Curve25519::dh1`. `regeneratePublicKey` recomputes the public half from a known private (used when restoring from backup).
+- **Keygen entry points**: at boot, `NodeDB` calls `generateKeyPair` (or `regeneratePublicKey` when a stored private key is present and passes a low-entropy check) **directly** when `!owner.is_licensed` and `config.lora.region != UNSET`. `ensurePkiKeys` wraps the same logic for runtime/admin flows — it's the path `AdminModule::handleSetConfig` runs when first assigning a valid region or when security config is written; **do not assume it's the universal boot-time gate**, because the NodeDB path bypasses it.
+- **Handshake**: `Curve25519::dh2(local_private, remote_public) → 32-byte shared secret → SHA-256 → 32-byte AES-256 key`. Recomputed per packet. The SHA-256 step is effectively a KDF over the raw ECDH output.
+- **Cipher**: AES-256-CCM via `aes_ccm_ae` / `aes_ccm_ad` (`src/mesh/aes-ccm.cpp`). MAC length (the `M` parameter) is **8 bytes**. No AAD — the MAC covers ciphertext only.
+- **Nonce (13 bytes / 104 bit)**: `aes_ccm_ae`/`aes_ccm_ad` use a 13-byte CCM nonce (`L = 2` is hardcoded in `src/mesh/aes-ccm.cpp`), not a 16-byte nonce. For PKI packets, `CryptoEngine::initNonce(fromNode, packetNum, extraNonce)` starts from the usual packet-derived nonce material, then overwrites nonce bytes `4..7` with a fresh 32-bit `extraNonce = random()`. The effective nonce bytes are therefore: bytes `0..3` = `packet_id`, bytes `4..7` = transmitted `extraNonce`, bytes `8..11` = `from_node`, byte `12` = `0x00`. The receiver reconstructs the same 13-byte nonce from the packet metadata plus the appended `extraNonce`.
+- **Wire overhead**: 12 bytes appended to the ciphertext = 8-byte MAC ‖ 4-byte extraNonce. Defined as `MESHTASTIC_PKC_OVERHEAD = 12` in `src/mesh/RadioInterface.h`. Only the 4-byte `extraNonce` is sent; the rest of the 13-byte CCM nonce is reconstructed from packet fields as described above. The Router's send path checks this overhead against `MAX_LORA_PAYLOAD_LEN` before committing to PKI.
+- **Send selection** (`Router::send`): the sender enters the PKI path when **all** hold — we're the originator AND not Ham mode AND not Portduino simradio AND not on the `serial`/`gpio` channels (unless the packet is already marked `pki_encrypted`) AND `config.security.private_key.size == 32` AND destination is a single node (not broadcast) AND the portnum isn't infrastructure. `TRACEROUTE_APP`, `NODEINFO_APP`, `ROUTING_APP`, and `POSITION_APP` are routed through channel encryption even when DMed (these need to be readable by relaying peers). Once on the PKI path, if the destination's public key isn't in our NodeDB the send **fails** with `PKI_SEND_FAIL_PUBLIC_KEY` — it does not silently fall back to channel encryption. If the client explicitly set `pki_encrypted=true` and any condition blocks PKI, the send fails with `PKI_FAILED`.
+- **Receive selection** (`Router::perhapsDecode`): try PKI decrypt first when `channel == 0` AND `isToUs(p)` AND not broadcast AND both peers have public keys in NodeDB AND `rawSize > MESHTASTIC_PKC_OVERHEAD`. On success the packet gets `pki_encrypted=true` stamped and the sender's public key copied into `p->public_key` for downstream authorization.
+
+### Remote admin authorization
+
+Implemented in `src/modules/AdminModule.cpp` → `handleReceivedProtobuf`. The authorization check runs in this order:
+
+1. **Response messages** — if `messageIsResponse(r)` is true (the payload is a response to one of our earlier admin requests), it's accepted without any further check. The in-file comment flags this as a known-untightened gap: a stricter implementation would remember which `public_key` we last queried and reject responses that don't match.
+2. **Local admin** — `mp.from == 0` (phone app over BLE, serial CLI, internal module); never travels over the air. **Rejected** if `config.security.is_managed` is true, because managed devices expect admin to arrive over the air through an authorized remote path.
+3. **Legacy admin channel (deprecated)** — the packet arrived on a channel named literally `"admin"`. Gated by `config.security.admin_channel_enabled`; returns `NOT_AUTHORIZED` if the flag is false. Kept for backward compatibility; new deployments should use PKI admin.
+4. **PKI admin (preferred for remote)** — `mp.pki_encrypted == true` AND `mp.public_key` matches one of `config.security.admin_key[0..2]` (up to three authorized 32-byte Curve25519 public keys, typically copied from the admin node's own `user.public_key`).
+5. **Fallthrough** → `NOT_AUTHORIZED`.
+
+On top of authorization, any remote admin message that **mutates** state (not a request, not a response) also has to pass a session-key check (`checkPassKey`): the client must first pull a fresh 8-byte `session_passkey` via `get_admin_session_key_request`, then echo that passkey back in the mutating message. The device rotates the passkey after 150 s and rejects values older than 300 s — a narrow anti-replay window on top of the PKI layer.
+
+`config.security.is_managed = true` disables **local** admin writes (`mp.from == 0` is rejected). It does not by itself force every admin action through PKI — the legacy `"admin"` channel still authorizes remote admin when `config.security.admin_channel_enabled == true`. The AdminModule refuses to persist `is_managed=true` unless at least one `admin_key` is populated — a deliberate guard against operators locking themselves out.
+
+### Key-rotation hazards (actions that invalidate peers)
+
+- **`factory_reset_device`** (the "full" variant, calls `NodeDB::factoryReset(eraseBleBonds=true)`) → **wipes** the X25519 private key; a fresh keypair is generated on the next region-set. Every existing peer holds the old public key, so DMs to this node silently fail PKI decrypt until every peer re-exchanges NodeInfo.
+- **`factory_reset_config`** (the "partial" variant, calls `NodeDB::factoryReset()` with `eraseBleBonds=false`) → **preserves** the X25519 private key in `installDefaultConfig(preserveKey=true)`; the public key is zeroed and gets rebuilt from the preserved private key on the next boot via the NodeDB path's `regeneratePublicKey` call. Identity is preserved and the mesh does not need to re-exchange keys.
+- **`region=UNSET → valid region`** → `ensurePkiKeys` runs inside the same `handleSetConfig` path; missing keys get generated at that moment.
+- **Ham mode transitions** — entering Ham mode (`user.is_licensed=true`) runs `Channels::ensureLicensedOperation`, which **wipes every channel PSK** (all traffic becomes cleartext) and disables the legacy admin channel. The X25519 private key is preserved on the device but not used because `Router::send` skips PKI when `owner.is_licensed` is true. Leaving Ham mode re-enables PKI with the preserved keypair but does not restore the wiped channel PSKs — the operator has to re-set them.
+- **Channel 0 PSK change** → every peer must re-learn the channel hash; cached NodeInfo becomes temporarily unreachable until the next broadcast.
+- **`security.private_key` blanked via admin** → regenerates both halves (unless in Ham mode) and propagates the new public key via NodeInfo.
+
+## NodeDB Layout (v25)
+
+`DEVICESTATE_CUR_VER = 25`, `DEVICESTATE_MIN_VER = 24`. The on-device NodeDB was split in v25 into a slim header table plus four optional satellite stores. Older v24 saves auto-migrate at boot. Old training-data instincts (`node->user.long_name`, `node->position.latitude_i`, `node->is_favorite`, `node->device_metrics.battery_level`) are wrong now — the fields aren't there. Read this section before touching anything that walks `nodeDB->meshNodes`.
+
+### Slim `NodeInfoLite`
+
+`UserLite` is flattened onto `NodeInfoLite` (no nested sub-message); `position` and `device_metrics` are removed entirely (tags reserved). MAC address is dropped. Long names are capped at 25 chars (`max_size:25` in `deviceonly.options`); `hw_model` and `role` are `int_size:8`. Encoded size dropped from ~166 B → ~105 B per node.
+
+Booleans are bit-packed into `NodeInfoLite.bitfield`. **Do not read or write the bits directly** — use the inline helpers in `src/mesh/NodeDB.h`:
+
+```cpp
+nodeInfoLiteHasUser(n)                  // bit 5 — user fields populated
+nodeInfoLiteIsFavorite(n)               // bit 3
+nodeInfoLiteIsIgnored(n)                // bit 4
+nodeInfoLiteIsMuted(n)                  // bit 1
+nodeInfoLiteIsLicensed(n)               // bit 6 — Ham mode peer
+nodeInfoLiteIsKeyManuallyVerified(n)    // bit 0
+nodeInfoLiteHasIsUnmessagable(n)        // bit 8 — "is_unmessagable was sent"
+nodeInfoLiteIsUnmessagable(n)           // bit 7
+// via_mqtt is bit 2 (mask exposed; predicate uses the mask directly)
+
+nodeInfoLiteSetBit(n, NODEINFO_BITFIELD_IS_FAVORITE_MASK, true);  // setter
+```
+
+### Satellite stores
+
+Four `std::unordered_map<NodeNum, …>` members on `NodeDB`, each gated by its own build flag:
+
+| Map               | Value type                      | Build flag                         |
+| ----------------- | ------------------------------- | ---------------------------------- |
+| `nodePositions`   | `meshtastic_PositionLite`       | `MESHTASTIC_EXCLUDE_POSITIONDB`    |
+| `nodeTelemetry`   | `meshtastic_DeviceMetrics`      | `MESHTASTIC_EXCLUDE_TELEMETRYDB`   |
+| `nodeEnvironment` | `meshtastic_EnvironmentMetrics` | `MESHTASTIC_EXCLUDE_ENVIRONMENTDB` |
+| `nodeStatus`      | `meshtastic_StatusMessage`      | `MESHTASTIC_EXCLUDE_STATUSDB`      |
+
+Defaults are ON (i.e., maps **excluded**) for STM32WL only — see `src/mesh/mesh-pb-constants.h`. On every other arch all four maps are present. When excluded, the map member is absent and the corresponding accessors return `false`.
+
+All four maps are guarded by **`mutable concurrency::Lock satelliteMutex`** — concurrent access from receive threads, the phone API state machine, and the renderer is the rule, not the exception.
+
+### Accessor convention
+
+**Never hand out pointers into the maps.** Use the copy-out accessors on `NodeDB`:
+
+```cpp
+bool copyNodePosition(NodeNum, meshtastic_PositionLite &out)       const;
+bool copyNodeTelemetry(NodeNum, meshtastic_DeviceMetrics &out)     const;
+bool copyNodeEnvironment(NodeNum, meshtastic_EnvironmentMetrics &out) const;
+bool copyNodeStatus(NodeNum, meshtastic_StatusMessage &out)        const;
+```
+
+Each takes the lock, copies the value if present, returns `false` if the entry is absent or the DB is excluded. Pass-by-out-param is deliberate — pointer-style accessors would invite UAF and lock-leak bugs across the renderer. The "has any X" convenience predicates (`hasValidPosition` etc.) are implemented in terms of these.
+
+Writers go through `setNodeStatus`, `updatePosition`, `updateTelemetry` (which dispatches on `which_variant` for device vs environment metrics) — these own the lock and the eviction hooks.
+
+### Eviction
+
+Every code path that drops a node from the header table must also evict the satellites. The single chokepoint is `eraseNodeSatellites(NodeNum)`; it's already called from `getOrCreateMeshNode`'s oldest-boring eviction, `removeNodeByNum`, both branches of `resetNodes`, `cleanupMeshDB`, `addFromContact`'s ignored-branch, and `AdminModule`'s `set_ignored_node`. Add new eviction sites here, not by calling `.erase()` directly.
+
+### Sync flow: thin NodeInfo + post-COMPLETE_ID replay (no opt-in)
+
+There is no capability flag and no special "gradient" nonce. The **default** sync flow is:
+
+1. Config / module-config / channel / metadata segments (same as before).
+2. `STATE_SEND_OWN_NODEINFO` — **our own** NodeInfo, still bundled with our position and device_metrics (because the replay snapshot excludes our own NodeNum). Emitted via `ConvertToNodeInfo(lite)`.
+3. `STATE_SEND_OTHER_NODEINFOS` — every other peer's NodeInfo, **always thin** (no `position`, no `device_metrics`). Emitted via `ConvertToNodeInfoThin(lite)`.
+4. `STATE_SEND_FILEMANIFEST` → `STATE_SEND_COMPLETE_ID` — the phone sees `config_complete_id` and treats sync as done.
+5. `STATE_SEND_PACKETS` — live mesh packets, with a trailing replay drain interleaved. The replay drain walks four cached satellite stores in order (positions → telemetry → environment → status) and emits each cached entry as an ordinary `MeshPacket` on the matching portnum (`POSITION_APP`, `TELEMETRY_APP` device + environment variants, `NODE_STATUS_APP`). These are indistinguishable on the wire from live mesh traffic, so clients need no special handling — any code that already updates UI on `POSITION_APP` etc. works.
+
+`PhoneAPI::sendConfigComplete()` arms `replayPhase = REPLAY_PHASE_POSITIONS` for default/full sync and `SPECIAL_NONCE_ONLY_NODES`, while `SPECIAL_NONCE_ONLY_CONFIG` skips replay. The drain runs inside `STATE_SEND_PACKETS` via `popReplayPacket()`, lower priority than live traffic. When all four phases drain, `replayPhase` flips back to `REPLAY_PHASE_IDLE` and the snapshot vectors get `shrink_to_fit`ed.
+
+STM32WL and any other build with all four `MESHTASTIC_EXCLUDE_*DB` flags set produces zero replay packets — `popReplayPacket` advances through each phase in microseconds without emitting anything.
+
+Special nonces that still mean something:
+
+- `SPECIAL_NONCE_ONLY_CONFIG` (69420) — skip node sync entirely, just config.
+- `SPECIAL_NONCE_ONLY_NODES` (69421) — skip config segments, jump straight to `STATE_SEND_OWN_NODEINFO`. Still gets the post-COMPLETE_ID replay drain.
+
+There are no other reserved nonces; everything else is a fresh random `want_config_id` from the client.
+
+### v24 → v25 migration
+
+The legacy migration code lives in **`src/mesh/NodeDBLegacyMigration.cpp`**, not in `NodeDB.cpp`. It owns the `meshtastic_NodeDatabase_Legacy` callback and `NodeDB::migrateLegacyNodeDatabase()`. The legacy proto descriptor is `protobufs/meshtastic/deviceonly_legacy.proto` (only included by the migration TU). The boot path peeks the file's leading version tag, runs the migration if `version < 25`, then re-saves in v25 layout. The legacy descriptor is scheduled for removal once `DEVICESTATE_MIN_VER` is bumped.
+
+### Read-site rules of thumb
+
+- Never `node->position.X` / `node->device_metrics.X` — those fields no longer exist. Pull from the satellite map via `copyNodePosition` / `copyNodeTelemetry`.
+- Never `node->user.long_name` — `long_name`, `short_name`, `public_key`, `hw_model`, `role`, `macaddr` (gone), `is_licensed`, `is_unmessagable` are flat on `NodeInfoLite`.
+- Never `node->is_favorite` / `node->is_ignored` / `node->via_mqtt` / `node->is_key_manually_verified` — use the bitfield helpers.
+- Never assume `nodeDB->getMeshNode(num)->position.time` — call `copyNodePosition` and check the return.
+- Don't lock `satelliteMutex` yourself in renderer code; the copy-out accessors already do.
+
+Unit tests for the conversion layer live in `test/test_type_conversions/test_main.cpp` (Unity) — bitfield round-trips, `long_name` truncation, thin-vs-full conversions. Add cases there when extending the schema.
+
 ## Project Structure

 ```
@@ -80,7 +238,7 @@ firmware/
 │   │   ├── NodeDB.*       # Node database management
 │   │   ├── Router.*       # Packet routing
 │   │   ├── Channels.*     # Channel management
-│   │   ├── CryptoEngine.* # AES-CCM encryption
+│   │   ├── CryptoEngine.* # AES-CTR (channels) + X25519 ECDH→AES-256-CCM (PKI for DMs/admin)
 │   │   ├── *Interface.*   # Radio interface implementations
 │   │   ├── api/           # WiFi/Ethernet server APIs (ServerAPI, PacketAPI)
 │   │   ├── http/          # HTTP server (WebServer, ContentHandler)
@@ -131,6 +289,8 @@ firmware/
 - Prefer `LOG_DEBUG`, `LOG_INFO`, `LOG_WARN`, `LOG_ERROR` for logging
 - Use `assert()` for invariants that should never fail
 - C++17 features are available (`std::optional`, structured bindings, `if constexpr`, etc.)
+- **Keep code comments minimal — one or two lines, max.** Comment only when the _why_ isn't obvious from the code; never restate what the next line does. No multi-paragraph block comments explaining straightforward changes. The diff and commit message carry the rationale; the code carries the behavior.
+- **Use `Throttle` for time-based rate limiting, not raw `millis()` math.** `src/mesh/Throttle.h` provides `Throttle::isWithinTimespanMs(lastMs, intervalMs)` (returns true while inside the cooldown) and `Throttle::execute(&lastMs, intervalMs, func)` (function-pointer form that updates the timestamp on fire). Use these for any "did N ms pass since X" check — raw `millis() > lastMs + N` is rollover-unsafe (breaks after ~49.7 days) and inconsistent with the rest of the codebase. The helpers compute `now - lastMs` with unsigned subtraction, which wraps correctly.

 ### Naming Conventions

@@ -296,6 +456,23 @@ Key defines in variant.h:

 ## Build System

+## Agent Tooling Baseline
+
+Mirror counterpart: `AGENTS.md` under **Agent Tooling Baseline**.
+
+To reduce avoidable agent mistakes, assume these tools are available (or install them before significant repo work):
+
+- **Required CLI basics**: `bash`, `git`, `find`, `grep`, `sed`, `awk`, `xargs`
+- **Strongly recommended**: `rg` (ripgrep) for fast file/text search, `jq` for JSON processing
+- **Build/test tools**: `python3`, `pip`, virtualenv (`python3 -m venv`), `platformio` (`pio`)
+- **Containerized native testing**: `docker` (fallback for non-Linux hosts; macOS can also build natively via `pio run -e native-macos`)
+
+Fallback expectations for agents:
+
+- If `rg` is unavailable, use `find` + `grep` instead of failing.
+- For native tests on hosts without Linux deps, prefer `./bin/test-native-docker.sh`.
+- The simulator helper script is `./bin/test-simulator.sh`.
+
 Uses **PlatformIO** with custom scripts:

 - `bin/platformio-pre.py` - Pre-build script
@@ -307,6 +484,7 @@ Build commands:
 pio run -e tbeam              # Build specific target
 pio run -e tbeam -t upload    # Build and upload
 pio run -e native             # Build native/Linux version
+pio run -e native-macos       # Build headless macOS meshtasticd (Homebrew prereqs in variants/native/portduino/platformio.ini)
 ```

 ### Build Manifest
@@ -448,6 +626,8 @@ Run with: `pio test -e native`

 Simulation testing: `bin/test-simulator.sh`

+Quick entry point for new test modules: `test/README.md` (native unit-test authoring guide, skeleton, pitfalls, and setup checklist).
+
 ### Hardware-in-the-loop tests (`mcp-server/tests/`)

 Separate pytest suite that exercises real USB-connected Meshtastic devices. See the **MCP Server & Hardware Test Harness** section below for invocation, tier layout, and agent usage rules.
@@ -474,7 +654,7 @@ The repo registers the server via `.mcp.json` at the repo root — Claude Code p

 **One MCP call per port at a time.** `SerialInterface` holds an exclusive OS-level lock on the serial port for its lifetime. If a `serial_*` session is open on `/dev/cu.usbmodem101`, calling `device_info` on the same port will fail fast pointing at the active session. Sequence calls: open → read/mutate → close, then next device. Never parallelize tool calls on the same port.

-### MCP tool surface (~32 tools)
+### MCP tool surface (43 tools)

 Grouped by purpose. Full argument shapes in `mcp-server/README.md`; a few high-value signatures are called out here.

@@ -482,11 +662,15 @@ Grouped by purpose. Full argument shapes in `mcp-server/README.md`; a few high-v
 - **Build & flash**: `build`, `clean`, `pio_flash`, `erase_and_flash` (ESP32 only), `update_flash` (ESP32 OTA), `touch_1200bps`
 - **Serial sessions** (long-running, 10k-line ring buffer): `serial_open`, `serial_read`, `serial_list`, `serial_close`
 - **Device reads**: `device_info`, `list_nodes`
- **Device writes** (all require `confirm=True`): `set_owner`, `get_config`, `set_config`, `get_channel_url`, `set_channel_url`, `send_text`, `reboot`, `shutdown`, `factory_reset`, `set_debug_log_api`
+- **Device writes**: `set_owner`, `get_config`, `set_config`, `get_channel_url`, `set_channel_url`, `send_text`, `send_input_event` (inject a button/key press via the firmware's InputBroker), `set_debug_log_api`; destructive/power-state writes require `confirm=True`: `reboot`, `shutdown`, `factory_reset`
 - **userPrefs admin** (build-time constants, not runtime config): `userprefs_get`, `userprefs_set`, `userprefs_reset`, `userprefs_manifest`, `userprefs_testing_profile`
 - **Vendor escape hatches**: `esptool_chip_info`, `esptool_erase_flash`, `esptool_raw`, `nrfutil_dfu`, `nrfutil_raw`, `picotool_info`, `picotool_load`, `picotool_raw`
+- **USB power control** (via `uhubctl`, per-port PPPS toggle): `uhubctl_list` (read-only), `uhubctl_power(action='on'|'off', confirm=True)`, `uhubctl_cycle(delay_s, confirm=True)`. Target by raw `(location, port)` or by `role` (`"nrf52"`, `"esp32s3"`); role lookup checks `MESHTASTIC_UHUBCTL_LOCATION_<ROLE>` + `_PORT_<ROLE>` env vars first, falls back to VID auto-detection.
+- **Observability** (UI tier + operator ad-hoc): `capture_screen(role, ocr=True)` — grabs a USB-webcam frame of the device OLED and optionally OCRs it. Requires `mcp-server[ui]` extras (`opencv-python-headless`, `easyocr`) and `MESHTASTIC_UI_CAMERA_DEVICE_<ROLE>` env var; falls through to a 1×1 black PNG `NullBackend` when unconfigured.

-`confirm=True` is a tool-level gate on top of whatever permission prompt your MCP host shows. **Don't bypass it** by asking the host to auto-approve — it exists specifically because MCP hosts sometimes remember "always allow this tool" and that's dangerous for `factory_reset` and `erase_and_flash`.
+`confirm=True` is a tool-level gate on top of whatever permission prompt your MCP host shows. **Don't bypass it** by asking the host to auto-approve — it exists specifically because MCP hosts sometimes remember "always allow this tool" and that's dangerous for `factory_reset`, `erase_and_flash`, `uhubctl_power(action='off')`, and `uhubctl_cycle`.
+
+**TCP / native-host nodes.** Setting `MESHTASTIC_MCP_TCP_HOST=<host[:port]>` makes `list_devices` surface a `meshtasticd` daemon (e.g. the `native-macos` build) as a synthetic `tcp://host:port` entry, and `connect()` routes through `meshtastic.tcp_interface.TCPInterface` instead of `SerialInterface`. Every read/write/admin tool that flows through `connect()` works against the daemon transparently. USB-only tools (`pio_flash`, `erase_and_flash`, `update_flash`, `touch_1200bps`, `serial_open`, `esptool_*`, `nrfutil_*`, `picotool_*`) raise a clear `ConnectionError` when handed a `tcp://` port; `pio_flash` against a `native*` env raises a `FlashError` (no upload step — use `build` and run the binary directly). The pytest harness still assumes USB-attached devices per role; TCP-aware fixtures are deferred. See `mcp-server/README.md` § "TCP / native-host nodes".

 ### Hardware test suite (`mcp-server/run-tests.sh`)

@@ -494,14 +678,16 @@ The wrapper auto-detects connected devices (VID → role map: `0x239A` → `nrf5

 Suite tiers (collected + run in this order via `pytest_collection_modifyitems`):

-1. `tests/unit/` — pure Python (boards parse, pio wrapper, userPrefs parse, testing profile). No hardware.
+1. `tests/unit/` — pure Python (boards parse, pio wrapper, userPrefs parse, testing profile, uhubctl parser). No hardware.
 2. `tests/test_00_bake.py` — flashes each detected device with current `userPrefs.jsonc` merged with the session's test profile. Has its own skip-if-already-baked check comparing region + primary channel to the session profile; skips cheaply on warm devices.
-3. `tests/mesh/` — multi-device mesh: bidirectional send, broadcast delivery, direct-with-ACK, mesh formation within 60s. Parametrized `[nrf52->esp32s3]` and `[esp32s3->nrf52]`.
+3. `tests/mesh/` — multi-device mesh: bidirectional send, broadcast delivery, direct-with-ACK, mesh formation within 60s. Parametrized `[nrf52->esp32s3]` and `[esp32s3->nrf52]`. Includes `test_peer_offline_recovery` which uses uhubctl to physically power off one peer mid-conversation (requires uhubctl; skips without).
 4. `tests/telemetry/` — `DEVICE_METRICS_APP` broadcast timing.
 5. `tests/monitor/` — boot-log panic check.
-6. `tests/fleet/` — PSK seed session isolation.
-7. `tests/admin/` — channel URL roundtrip, owner persistence across reboot.
-8. `tests/provisioning/` — region + modem + slot bake, admin key presence, `UNSET` region blocks TX, userPrefs survive factory reset.
+6. `tests/recovery/` — `uhubctl` power-cycle round-trip + NVS persistence across hard reset. Requires `uhubctl` installed and a PPPS-capable hub; entire tier auto-skips otherwise.
+7. `tests/ui/` — input-broker-driven screen navigation with camera + OCR evidence.
+8. `tests/fleet/` — PSK seed session isolation.
+9. `tests/admin/` — channel URL roundtrip, owner persistence across reboot.
+10. `tests/provisioning/` — region + modem + slot bake, admin key presence, `UNSET` region blocks TX, userPrefs survive factory reset.

 Invocation patterns:

@@ -586,15 +772,19 @@ If you're modifying `StreamAPI`, `PhoneAPI`, `NodeInfoModule`, or `userPrefs` fl

 ### Recovery playbooks

-| Symptom                                                    | First check                                                   | Fix                                                                                                                                                                        |
-| ---------------------------------------------------------- | ------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
-| `userPrefs.jsonc` dirty after test run                     | `git status --porcelain userPrefs.jsonc`                      | If non-empty, re-run `./mcp-server/run-tests.sh` once — the pre-flight self-heal restores from sidecar. If still dirty, `git checkout userPrefs.jsonc`.                    |
-| Port busy / wedged CP2102 on macOS                         | `lsof /dev/cu.usbserial-0001`                                 | Kill the holder. USB replug if the kernel still reports busy. Often a stale `pio device monitor` or zombie `meshtastic_mcp` process.                                       |
-| nRF52 appears unresponsive                                 | `list_devices` shows VID `0x239A` but `device_info` times out | `touch_1200bps(port=...)` drops it into the DFU bootloader → `pio_flash` re-installs.                                                                                      |
-| Multiple MCP server processes                              | `ps aux \| grep meshtastic_mcp` shows >1                      | Kill all but the one your MCP host spawned. Zombies hold ports and break tests.                                                                                            |
-| Mesh formation fails, one side sees peer but other doesn't | `/diagnose` (or `list_nodes` on both sides)                   | Asymmetric NodeInfo. `test_direct_with_ack` has a heal path; `/repro` it a few times. If persistent, both devices' clocks may be out of sync with their NodeInfo cooldown. |
-| "role not present on hub" in skip reasons                  | `list_devices`                                                | Expected if a device is unplugged. Reconnect before re-running the tier.                                                                                                   |
-| Tests fail only on first attempt then pass on rerun        | —                                                             | State leak from a prior session. Run with `--force-bake` to reset to a known state.                                                                                        |
+| Symptom                                                                           | First check                                                   | Fix                                                                                                                                                                                                                              |
+| --------------------------------------------------------------------------------- | ------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `userPrefs.jsonc` dirty after test run                                            | `git status --porcelain userPrefs.jsonc`                      | If non-empty, re-run `./mcp-server/run-tests.sh` once — the pre-flight self-heal restores from sidecar. If still dirty, `git checkout userPrefs.jsonc`.                                                                          |
+| Port busy / wedged CP2102 on macOS                                                | `lsof /dev/cu.usbserial-0001`                                 | Kill the holder. USB replug if the kernel still reports busy. Often a stale `pio device monitor` or zombie `meshtastic_mcp` process.                                                                                             |
+| nRF52 appears unresponsive                                                        | `list_devices` shows VID `0x239A` but `device_info` times out | `touch_1200bps(port=...)` drops it into the DFU bootloader → `pio_flash` re-installs.                                                                                                                                            |
+| Device fully wedged (Guru Meditation, frozen CDC, no DFU)                         | `list_devices` shows the VID but every admin call times out   | `uhubctl_cycle(role="nrf52", confirm=True)` hard-power-cycles the port via USB hub PPPS. `baked_single`'s auto-recovery hook does this once automatically if uhubctl is installed. Falls back to physical replug if no PPPS hub. |
+| Multiple MCP server processes                                                     | `ps aux \| grep meshtastic_mcp` shows >1                      | Kill all but the one your MCP host spawned. Zombies hold ports and break tests.                                                                                                                                                  |
+| Mesh formation fails, one side sees peer but other doesn't                        | `/diagnose` (or `list_nodes` on both sides)                   | Asymmetric NodeInfo. `test_direct_with_ack` has a heal path; `/repro` it a few times. If persistent, both devices' clocks may be out of sync with their NodeInfo cooldown.                                                       |
+| "role not present on hub" in skip reasons                                         | `list_devices`                                                | Expected if a device is unplugged. Reconnect before re-running the tier.                                                                                                                                                         |
+| Entire `tests/recovery/` tier skipped                                             | `command -v uhubctl`                                          | Expected if `uhubctl` isn't on PATH. Install via `brew install uhubctl` (macOS) or `apt install uhubctl` (Debian/Ubuntu). Also skips if no hub advertises PPPS.                                                                  |
+| Entire `tests/ui/` tier skipped ("firmware not baked with USERPREFS_UI_TEST_LOG") | reportlog.jsonl for the skip reason                           | Re-run with `--force-bake` so the UI-log macro gets compiled into the fresh firmware. First run after the Round-3 landing always re-bakes.                                                                                       |
+| `tests/ui/` runs but captures are all 1×1 black PNGs                              | `MESHTASTIC_UI_CAMERA_DEVICE_ESP32S3`                         | Env var not set → `NullBackend`. Point a USB webcam at the heltec-v3 OLED and set the device index; `.venv/bin/python -c "import cv2; [print(i, cv2.VideoCapture(i).read()[0]) for i in range(5)]"` discovers it.                |
+| Tests fail only on first attempt then pass on rerun                               | —                                                             | State leak from a prior session. Run with `--force-bake` to reset to a known state.                                                                                                                                              |

 ### Never do these without asking

@@ -26,7 +26,12 @@ This prompt assumes the meshtastic MCP server is registered with your VS Code Co
   - `get_config(section="lora", port=<p>)` → region, preset, channel_num, tx_power, hop_limit
   - If anything looks off (can't connect, `num_nodes` wrong, missing `firmware_version`), open a short firmware-log window: `serial_open(port=<p>, env=<inferred>)`, wait 3 seconds, `serial_read(session_id, max_lines=100)`, `serial_close(session_id)`. Infer env from VID (0x239a → `rak4631`, 0x303a/0x10c4 → `heltec-v3`) unless an `MESHTASTIC_MCP_ENV_<ROLE>` env var overrides it.

-4. **Render per-device report** as a compact block:
+4. **Hub health** (call once, not per-device): `uhubctl_list()` — enumerates every USB hub the host sees. Cross-reference each Meshtastic device's VID to find which hub + port it's on. Flag in the report if:
+   - No hub advertises `ppps=true` → `tests/recovery/` can't run; hard-recovery via `uhubctl_cycle` isn't available.
+   - A Meshtastic device is on a non-PPPS hub → note it; moving to a PPPS hub unlocks auto-recovery.
+   - `uhubctl_list` raises `ConfigError: uhubctl not found` → report as "uhubctl not installed"; don't treat as a device fault.
+
+5. **Render per-device report** as a compact block:

   ```text
   [nrf52 @ /dev/cu.usbmodem1101]      fw=2.7.23.bce2825, hw=RAK4631
@@ -35,20 +40,22 @@ This prompt assumes the meshtastic MCP server is registered with your VS Code Co
     tx_power    : 30 dBm, hop_limit=3
     peers       : 1 (esp32s3 0x433c2428, pubkey ✓, SNR 6.0 / RSSI -24 dBm)
     primary ch  : McpTest
+     hub         : 1-1.3 port 2 (PPPS, uhubctl-controllable)
     firmware    : no panics in last 3s
   ```

-   Flag abnormalities inline with `⚠︎ <short reason>` — missing pubkey on a known peer, region UNSET, mismatched channel name, etc.
+   Flag abnormalities inline with `⚠︎ <short reason>` — missing pubkey on a known peer, region UNSET, mismatched channel name, device on non-PPPS hub, etc.

-5. **Cross-device correlation** (when >1 device selected):
+6. **Cross-device correlation** (when >1 device selected):
   - Do both see each other in `nodesByNum`?
   - Do `region`, `channel_num`, `modem_preset` match across devices?
   - Do the primary channel names match? (Different name → different PSK → no decode.)

-6. **Suggest next steps only for recognizable failure modes**, never speculatively:
+7. **Suggest next steps only for recognizable failure modes**, never speculatively:
   - Stale PKI one-way → "`/mcp-test tests/mesh/test_direct_with_ack.py` — the test's retry+nodeinfo-ping heals this."
   - Region mismatch → "re-bake one side via `./mcp-server/run-tests.sh --force-bake`."
-   - Device unreachable → refer operator to the touch_1200bps + CP2102-wedged-driver notes in `run-tests.sh`.
+   - Device unreachable, DFU reachable → `touch_1200bps(port=...)` + `pio_flash`. If not even DFU responds and the device is on a PPPS hub, escalate to `uhubctl_cycle(role=..., confirm=True)`.
+   - CP2102-wedged-driver on macOS → see `run-tests.sh` notes.

 ## Hard constraints

@@ -46,7 +46,8 @@ Equivalent of `.claude/commands/repro.md`. Use when the operator says "that one
   - **LoRa airtime collision** — pass rate improves with fewer concurrent transmitters. Suggest a `time.sleep` gap or retry bump in the test body.
   - **PKI key staleness** — first attempt fails, subsequent ones pass; existing retry-loop pattern in `test_direct_with_ack.py` is the fix.
   - **NodeInfo cooldown** — `Skip send NodeInfo since we sent it <600s ago` in fail-only logs; needs a `broadcast_nodeinfo_ping()` warmup.
-   - **Hardware-specific** — one direction consistently fails, firmware versions differ, CP2102 driver wedged, etc.
+   - **Hardware-specific** — one direction consistently fails, firmware versions differ, CP2102 driver wedged, etc. For a device wedged past `touch_1200bps`, recommend `uhubctl_cycle(role=..., confirm=True)` to hard-power-cycle its hub port (requires `uhubctl` installed).
+   - **Device went dark mid-run** — fails from some iteration onward and never recovers; firmware log stops arriving. Almost always a Guru crash with frozen CDC. Recommend `uhubctl_cycle` before the next iteration; escalate to replug if that also fails.
   - **Unknown** — say so. Don't invent a root cause.

 7. **Report back** with:
@@ -21,19 +21,25 @@ Equivalent of the Claude Code `/test` slash command in `.claude/commands/test.md

 2. **Read the pre-flight header** (first few lines of wrapper output). The `detected hub :` line lists role → port → env mappings. If it reads `(none)`, the wrapper narrowed to `tests/unit` only — call that out explicitly so the operator knows hardware tiers were skipped.

-3. **On pass**: one-line summary like `N passed, M skipped in <duration>`. Don't enumerate test names. DO mention any non-placeholder SKIPs (things like "role not present on hub") because they indicate missing hardware or setup issues.
+3. **On pass**: one-line summary like `N passed, M skipped in <duration>`. Don't enumerate test names. DO mention any non-placeholder SKIPs and name the cause:
+   - `"role not present on hub"` → device unplugged; operator should reconnect.
+   - `"firmware not baked with USERPREFS_UI_TEST_LOG"` → tests/ui skipped; the UI-log compile macro isn't in the baked firmware. Suggest `--force-bake`.
+   - `"uhubctl not installed"` → tests/recovery + `test_peer_offline_recovery` skipped. Suggest `brew install uhubctl` / `apt install uhubctl`.
+   - `"no PPPS-capable hubs detected"` → tests/recovery skipped because the attached hub doesn't support per-port power switching; won't run on that setup.
+   - `"opencv-python-headless is not installed"` → tests/ui auto-deselected by `run-tests.sh`. Suggest `pip install -e 'mcp-server/.[ui]'`.

 4. **On failure**: open `mcp-server/tests/report.html` (pytest-html output, self-contained) and extract the `Meshtastic debug` section for each failed test. That section includes a firmware log stream (last 200 lines) and device state dump. For each failure, summarise:
   - test name
   - one-line assertion message
   - the specific firmware log lines that explain why (look for `PKI_UNKNOWN_PUBKEY`, `Skip send NodeInfo`, `Error=`, `Guru Meditation`, `assertion failed`, `No suitable channel`)
+   - for UI-tier failures also check `mcp-server/tests/ui_captures/<session>/<test>/transcript.md` (per-step frame + OCR)

 5. **Classify each failure** as one of:
   - **Transient flake** — LoRa collision, first-attempt NAK with self-heal pattern, timing-sensitive assertion. Suggest `/mcp-repro <test-id>` to confirm.
-   - **Environmental** — device unreachable, port busy, CP2102 driver wedged on macOS. Suggest specific recovery (USB replug, `touch_1200bps`, `git status userPrefs.jsonc`).
+   - **Environmental** — device unreachable, port busy, CP2102 driver wedged on macOS. Suggest recovery in escalation order: (a) replug USB, (b) `touch_1200bps` + `pio_flash` for nRF52 DFU, (c) `uhubctl_cycle(role=..., confirm=True)` for a device wedged past DFU (needs `uhubctl` installed; `baked_single` does this once automatically when available). Also check `git status userPrefs.jsonc`.
   - **Regression** — same assertion fails repeatedly on re-runs, firmware log shows novel errors. Identify the firmware module likely responsible.

-6. **Do NOT run destructive recovery automatically**. If a failure looks like it needs a reflash, factory*reset, or replug — \_describe the steps* and let the operator decide. Never burn airtime or flash cycles without approval.
+6. **Do NOT run destructive recovery automatically**. If a failure looks like it needs a reflash, factory*reset, `uhubctl_cycle`, or replug — \_describe the steps* and let the operator decide. Never burn airtime or flash cycles without approval.

 ## Arguments convention

@@ -118,7 +118,7 @@ CallbackObserver<MyModule, const meshtastic::Status *> statusObserver =

 Add test suite in `test/test_mymodule/`:

-```
+```text
 test/
 └── test_mymodule/
    └── test_main.cpp
@@ -6,7 +6,7 @@ Guide for adding a new Meshtastic hardware variant to the firmware.

 Create under `variants/<arch>/<name>/`:

-```
+```text
 variants/
 ├── esp32/          # ESP32
 ├── esp32s3/        # ESP32-S3
@@ -32,10 +32,15 @@ jobs:
        shell: bash
        working-directory: meshtasticd
        run: |
+          # Build-tools (notably platformio) come from the Meshtastic project
+          # on the OpenSUSE Build Service:
+          # https://build.opensuse.org/project/show/network:Meshtastic:build-tools
+          echo 'deb http://download.opensuse.org/repositories/network:/Meshtastic:/build-tools/xUbuntu_24.04/ /' \
+            | sudo tee /etc/apt/sources.list.d/network:Meshtastic:build-tools.list
+          curl -fsSL https://download.opensuse.org/repositories/network:Meshtastic:build-tools/xUbuntu_24.04/Release.key \
+            | gpg --dearmor | sudo tee /etc/apt/trusted.gpg.d/network_Meshtastic_build-tools.gpg >/dev/null
          sudo apt-get update -y --fix-missing
-          sudo apt-get install -y software-properties-common build-essential devscripts equivs
-          sudo add-apt-repository ppa:meshtastic/build-tools -y
-          sudo apt-get update -y --fix-missing
+          sudo apt-get install -y build-essential devscripts equivs
          sudo mk-build-deps --install --remove --tool='apt-get -o Debug::pkgProblemResolver=yes --no-install-recommends --yes' debian/control

      - name: Import GPG key
@@ -0,0 +1,51 @@
+name: Build MacOS Binary
+
+on:
+  workflow_call:
+    inputs:
+      macos_ver:
+        required: false
+        default: "26" # ARM64
+        type: string
+
+permissions:
+  contents: read
+
+jobs:
+  build-MacOS:
+    runs-on: macos-${{ inputs.macos_ver }}
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v6
+        with:
+          submodules: recursive
+
+      - name: Install deps
+        shell: bash
+        run: |
+          brew update
+          brew install platformio yaml-cpp libuv openssl@3 libusb argp-standalone pkg-config ulfius
+
+      - name: Get release version string
+        run: |
+          echo "long=$(./bin/buildinfo.py long)" >> $GITHUB_OUTPUT
+        id: version
+
+      - name: Build for MacOS
+        run: |
+          platformio run -e native-macos
+        env:
+          PKG_VERSION: ${{ steps.version.outputs.long }}
+        # Errors in this step should not fail the entire workflow while MacOS support is in development.
+        continue-on-error: true
+
+      - name: List output files
+        run: ls -lah .pio/build/native-macos/
+
+      - name: Store binaries as an artifact
+        uses: actions/upload-artifact@v7
+        with:
+          name: firmware-macos-${{ inputs.macos_ver }}-${{ steps.version.outputs.long }}
+          overwrite: true
+          path: |
+            .pio/build/native-macos/meshtasticd
@@ -4,9 +4,14 @@ on:
  workflow_dispatch:
    inputs:
      # trunk-ignore(checkov/CKV_GHA_7)
+      target:
+        type: string
+        required: false
+        description: Choose the target board, e.g. nrf52_promicro_diy_tcxo. If blank, will find available targets.
      arch:
        type: choice
        options:
+          - all
          - esp32
          - esp32s3
          - esp32c3
@@ -15,32 +20,18 @@ on:
          - rp2040
          - rp2350
          - stm32
-      target:
-        type: string
-        required: false
-        description: Choose the target board, e.g. nrf52_promicro_diy_tcxo. If blank, will find available targets.
-      # find-target:
-      #   type: boolean
-      #   default: true
-      #   description: 'Find the available targets'
+        description: Choose an arch to limit the search, or 'all' to search all architectures.
+        default: all

 permissions: read-all

 jobs:
  find-targets:
-    if: ${{ inputs.target == '' }}
    strategy:
      fail-fast: false
      matrix:
        arch:
-          - esp32
-          - esp32s3
-          - esp32c3
-          - esp32c6
-          - nrf52840
-          - rp2040
-          - rp2350
-          - stm32
+          - all
    runs-on: ubuntu-24.04
    steps:
      - uses: actions/checkout@v6
@@ -51,14 +42,37 @@ jobs:
      - run: pip install -U platformio
      - name: Generate matrix
        id: jsonStep
+        env:
+          BUILDTARGET: ${{ inputs.target }}
+          MATRIXARCH: ${{ inputs.arch }}
        run: |
          TARGETS=$(./bin/generate_ci_matrix.py ${{matrix.arch}} --level extra)
-          echo "Name: $GITHUB_REF_NAME" >> $GITHUB_STEP_SUMMARY
-          echo "Base: $GITHUB_BASE_REF" >> $GITHUB_STEP_SUMMARY
-          echo "Arch: ${{matrix.arch}}" >> $GITHUB_STEP_SUMMARY
-          echo "Ref: $GITHUB_REF" >> $GITHUB_STEP_SUMMARY
-          echo "Targets:" >> $GITHUB_STEP_SUMMARY
-          echo $TARGETS | jq -r 'sort_by(.board) |.[] | "- " + .board' >> $GITHUB_STEP_SUMMARY
+          if [ "$BUILDTARGET" = "" ]; then
+            echo "Name: $GITHUB_REF_NAME" >> $GITHUB_STEP_SUMMARY
+            echo "Base: $GITHUB_BASE_REF" >> $GITHUB_STEP_SUMMARY
+            echo "Arch: $MATRIXARCH" >> $GITHUB_STEP_SUMMARY
+            echo "Ref: $GITHUB_REF" >> $GITHUB_STEP_SUMMARY
+            echo "## 🎯 The following target boards are available to build:" >> $GITHUB_STEP_SUMMARY
+            echo "| Platform | Board |" >> $GITHUB_STEP_SUMMARY
+            echo "| -------- | ----- |" >> $GITHUB_STEP_SUMMARY
+            echo $TARGETS | jq -r 'sort_by(.board) | sort_by(.platform) |.[] | "| " + .platform + " | " + .board + " |" ' >> $GITHUB_STEP_SUMMARY
+          else
+            echo "We build this one:" >> $GITHUB_STEP_SUMMARY
+            ARCH=$(echo "$TARGETS" | jq --arg BUILDTARGET "$BUILDTARGET" -r '.[] | select(.board==$BUILDTARGET) | .platform')
+            echo "| Platform | Board |" >> $GITHUB_STEP_SUMMARY
+            echo "| -------- | ----- |" >> $GITHUB_STEP_SUMMARY
+            echo "| $ARCH | "$BUILDTARGET" |" >> $GITHUB_STEP_SUMMARY
+            echo "" >> $GITHUB_STEP_SUMMARY
+            if [[ "$ARCH" == "" ]]; then
+              echo "## ❌ Error: Target "$BUILDTARGET" not found!" >> $GITHUB_STEP_SUMMARY
+            else
+              echo "## ✅ Target "$BUILDTARGET" found, proceeding to build." >> $GITHUB_STEP_SUMMARY
+            fi
+            echo "You may need to refresh this page to make the built firmware appear below." >> $GITHUB_STEP_SUMMARY
+            echo "arch=$ARCH" >> $GITHUB_OUTPUT
+          fi
+    outputs:
+      arch: ${{ steps.jsonStep.outputs.arch }}

  version:
    if: ${{ inputs.target != '' }}
@@ -78,12 +92,12 @@ jobs:

  build:
    if: ${{ inputs.target != '' && inputs.arch != 'native' }}
-    needs: [version]
+    needs: [version, find-targets]
    uses: ./.github/workflows/build_firmware.yml
    with:
      version: ${{ needs.version.outputs.long }}
      pio_env: ${{ inputs.target }}
-      platform: ${{ inputs.arch }}
+      platform: ${{ needs.find-targets.outputs.arch }}

  gather-artifacts:
    permissions:
@@ -73,7 +73,9 @@ jobs:
      - name: Sanitize platform string
        id: sanitize_platform
        # Replace slashes with underscores
-        run: echo "cleaned_platform=${{ inputs.platform }}" | sed 's/\//_/g' >> $GITHUB_OUTPUT
+        env:
+          plat: ${{ inputs.platform }}
+        run: echo "cleaned_platform=${plat}" | sed 's/\//_/g' >> $GITHUB_OUTPUT

      - name: Docker login
        if: ${{ inputs.push }}
@@ -43,6 +43,15 @@ jobs:
      push: true
    secrets: inherit

+  docker-debian-riscv64:
+    uses: ./.github/workflows/docker_build.yml
+    with:
+      distro: debian
+      platform: linux/riscv64
+      runs-on: ubuntu-24.04-arm
+      push: true
+    secrets: inherit
+
  docker-alpine-amd64:
    uses: ./.github/workflows/docker_build.yml
    with:
@@ -70,16 +79,27 @@ jobs:
      push: true
    secrets: inherit

+  docker-alpine-riscv64:
+    uses: ./.github/workflows/docker_build.yml
+    with:
+      distro: alpine
+      platform: linux/riscv64
+      runs-on: ubuntu-24.04-arm
+      push: true
+    secrets: inherit
+
  docker-manifest:
    needs:
      # Debian
      - docker-debian-amd64
      - docker-debian-arm64
      - docker-debian-armv7
+      - docker-debian-riscv64
      # Alpine
      - docker-alpine-amd64
      - docker-alpine-arm64
      - docker-alpine-armv7
+      - docker-alpine-riscv64
    runs-on: ubuntu-24.04
    steps:
      - name: Checkout code
@@ -162,6 +182,7 @@ jobs:
            meshtastic/meshtasticd@${{ needs.docker-debian-amd64.outputs.digest }}
            meshtastic/meshtasticd@${{ needs.docker-debian-arm64.outputs.digest }}
            meshtastic/meshtasticd@${{ needs.docker-debian-armv7.outputs.digest }}
+            meshtastic/meshtasticd@${{ needs.docker-debian-riscv64.outputs.digest }}

      - name: Docker meta (Alpine)
        id: meta_alpine
@@ -182,3 +203,4 @@ jobs:
            meshtastic/meshtasticd@${{ needs.docker-alpine-amd64.outputs.digest }}
            meshtastic/meshtasticd@${{ needs.docker-alpine-arm64.outputs.digest }}
            meshtastic/meshtasticd@${{ needs.docker-alpine-armv7.outputs.digest }}
+            meshtastic/meshtasticd@${{ needs.docker-alpine-riscv64.outputs.digest }}
@@ -116,6 +116,20 @@ jobs:
      build_location: local
    secrets: inherit

+  MacOS:
+    strategy:
+      fail-fast: false
+      matrix:
+        macos_ver:
+          - "26" # ARM64
+          # - '26-intel' # x86_64
+          - "15" # ARM64
+          # - '15-intel' # x86_64
+    uses: ./.github/workflows/build_macos_bin.yml
+    with:
+      macos_ver: ${{ matrix.macos_ver }}
+    # secrets: inherit
+
  package-pio-deps-native-tft:
    if: ${{ github.repository == 'meshtastic/firmware' && github.event_name == 'workflow_dispatch' }}
    uses: ./.github/workflows/package_pio_deps.yml
@@ -286,6 +300,7 @@ jobs:
      - gather-artifacts
      - build-debian-src
      - package-pio-deps-native-tft
+      # - MacOS
    steps:
      - name: Checkout
        uses: actions/checkout@v6
@@ -86,7 +86,13 @@ jobs:
        run: sed -i 's/-DBUILD_EPOCH=$UNIX_TIME/#-DBUILD_EPOCH=$UNIX_TIME/' platformio.ini

      - name: PlatformIO Tests
-        run: platformio test -e coverage -v --junit-output-path testreport.xml
+        run: |
+          set -o pipefail
+          # Filter out SKIPPED summary rows for hardware variants that can't run on the
+          # native host. They flood the log and make it harder to spot real failures.
+          # The JUnit XML is written directly to testreport.xml before the pipe, so
+          # the test artifact is unaffected.
+          platformio test -e coverage -v --junit-output-path testreport.xml 2>&1 | grep -v "[[:space:]]SKIPPED$"

      - name: Save test results
        if: always() # run this step even if previous step failed
@@ -47,6 +47,10 @@ data/boot/logo.*
 managed_components/*
 arduino-lib-builder*
 dependencies.lock
+
+# JLink / RTT debug artifacts (nRF SoCs)
+flash.jlink
+rtt_*.txt
 idf_component.yml
 CMakeLists.txt
 /sdkconfig.*
@@ -56,3 +60,9 @@ CMakeLists.txt
 .python3
 .claude/scheduled_tasks.lock
 userPrefs.jsonc.mcp-session-bak
+
+# Fake-NodeDB fixture pipeline (bin/regen-fake-nodedbs.sh)
+# JSONL seeds are committed (test/fixtures/nodedb/seed_v25_*.jsonl);
+# compiled .proto outputs are ephemeral build artifacts.
+build/fixtures/
+bin/_generated/
@@ -4,29 +4,29 @@ cli:
 plugins:
  sources:
    - id: trunk
-      ref: v1.7.6
+      ref: v1.10.0
      uri: https://github.com/trunk-io/plugins
 lint:
  enabled:
-    - checkov@3.2.517
-    - renovate@43.110.9
-    - prettier@3.8.1
-    - trufflehog@3.94.3
+    - checkov@3.2.529
+    - renovate@43.150.0
+    - prettier@3.8.3
+    - trufflehog@3.95.3
    - yamllint@1.38.0
    - bandit@1.9.4
-    - trivy@0.69.3
+    - trivy@0.70.0
    - taplo@0.10.0
-    - ruff@0.15.9
+    - ruff@0.15.13
    - isort@8.0.1
    - markdownlint@0.48.0
-    - oxipng@10.1.0
+    - oxipng@10.1.1
    - svgo@4.0.1
    - actionlint@1.7.12
    - flake8@7.3.0
    - hadolint@2.14.0
    - shfmt@3.6.0
    - shellcheck@0.11.0
-    - black@26.3.1
+    - black@26.5.1
    - git-diff-check
    - gitleaks@8.30.1
    - clang-format@16.0.3
@@ -34,9 +34,16 @@ lint:
    - linters: [ALL]
      paths:
        - bin/**
+    # Fake-NodeDB fixture JSONL files contain deterministic synthetic
+    # public_key_hex (64-char hex) values that gitleaks misidentifies as
+    # generic-api-key. These are not secrets — they're test fixtures
+    # produced by bin/gen-fake-nodedb-seed.py with a fixed RNG seed.
+    - linters: [gitleaks]
+      paths:
+        - test/fixtures/nodedb/seed_v25_*.jsonl
 runtimes:
  enabled:
-    - python@3.10.8
+    - python@3.14.4
    - go@1.21.0
    - node@22.16.0
 actions:
@@ -1,10 +1,10 @@
 {
-    // See http://go.microsoft.com/fwlink/?LinkId=827846
-    // for the documentation about the extensions.json format
    "recommendations": [
-        "platformio.platformio-ide"
+        "Jason2866.esp-decoder",
+        "pioarduino.pioarduino-ide"
    ],
    "unwantedRecommendations": [
-        "ms-vscode.cpptools-extension-pack"
+        "ms-vscode.cpptools-extension-pack",
+        "platformio.platformio-ide"
    ]
 }
@@ -10,17 +10,18 @@ This file (`AGENTS.md`) is a short pointer + quick reference for agents that don

 ## Quick command reference

-| Action                           | Command                                                                             |
-| -------------------------------- | ----------------------------------------------------------------------------------- |
-| Build a firmware variant         | `pio run -e <env>` (e.g. `pio run -e rak4631`, `pio run -e heltec-v3`)              |
-| Clean + rebuild                  | `pio run -e <env> -t clean && pio run -e <env>`                                     |
-| Flash a device                   | `pio run -e <env> -t upload --upload-port <port>` (or use the `pio_flash` MCP tool) |
-| Run firmware unit tests (native) | `pio test -e native`                                                                |
-| Run MCP hardware tests           | `./mcp-server/run-tests.sh`                                                         |
-| Live TUI test runner             | `mcp-server/.venv/bin/meshtastic-mcp-test-tui`                                      |
-| Format before commit             | `trunk fmt`                                                                         |
-| Regenerate protobuf bindings     | `bin/regen-protos.sh`                                                               |
-| Generate CI matrix               | `./bin/generate_ci_matrix.py all [--level pr]`                                      |
+| Action                           | Command                                                                                                       |
+| -------------------------------- | ------------------------------------------------------------------------------------------------------------- |
+| Build a firmware variant         | `pio run -e <env>` (e.g. `pio run -e rak4631`, `pio run -e heltec-v3`)                                        |
+| Build native macOS host binary   | `pio run -e native-macos` (Homebrew prereqs + CH341 LoRa setup in `variants/native/portduino/platformio.ini`) |
+| Clean + rebuild                  | `pio run -e <env> -t clean && pio run -e <env>`                                                               |
+| Flash a device                   | `pio run -e <env> -t upload --upload-port <port>` (or use the `pio_flash` MCP tool)                           |
+| Run firmware unit tests (native) | `pio test -e native`                                                                                          |
+| Run MCP hardware tests           | `./mcp-server/run-tests.sh`                                                                                   |
+| Live TUI test runner             | `mcp-server/.venv/bin/meshtastic-mcp-test-tui`                                                                |
+| Format before commit             | `trunk fmt`                                                                                                   |
+| Regenerate protobuf bindings     | `bin/regen-protos.sh`                                                                                         |
+| Generate CI matrix               | `./bin/generate_ci_matrix.py all [--level pr]`                                                                |

 ## MCP server (device + test automation)

@@ -48,6 +49,15 @@ Three test-and-diagnose workflows exist as slash commands:

 Bodies live in `.claude/commands/` and `.github/prompts/` respectively. `.claude/commands/README.md` is the index.

+## Encryption at a glance
+
+Two layers, both in `src/mesh/CryptoEngine.cpp`:
+
+- **Channel (symmetric)** — **AES-CTR** with a channel-wide PSK (AES-128 or AES-256). Nonce = packet_id ‖ from_node ‖ block_counter. No AEAD; integrity is soft (channel-hash filter). The well-known default PSK lives in `src/mesh/Channels.h`; a 1-byte PSK is a short-form index into it.
+- **Per-peer PKI** — **X25519 ECDH** (Curve25519, 32-byte keys) → SHA-256 → **AES-256-CCM** with an 8-byte MAC. Fresh 32-bit `extraNonce` per packet, sent in the clear alongside the MAC. 12-byte wire overhead (`MESHTASTIC_PKC_OVERHEAD`). Used for DMs. Also used for remote admin (`src/modules/AdminModule.cpp`), where AdminMessage authorization is gated by `config.security.admin_key[0..2]`. Disabled entirely in Ham mode (`user.is_licensed=true`).
+
+Key rotation to never trigger casually: only the **full** factory reset (`factory_reset_device`, `eraseBleBonds=true`) wipes `security.private_key` and regenerates the keypair — every peer holds the old public key, so DMs silently fail PKI decrypt until NodeInfo re-exchanges. The **partial** config reset (`factory_reset_config`) preserves the private key and doesn't invalidate peer relationships. Explicitly blanking `security.private_key` via admin also triggers regen. See the **Encryption & Key Management** section of `.github/copilot-instructions.md` for the full spec (nonce layout, send/receive selection logic including infrastructure-portnum exceptions, admin-key + session-passkey authorization, `is_managed` scope, key-rotation hazards).
+
 ## House rules

 - **No destructive device operations without operator approval.** `factory_reset`, `erase_and_flash`, `reboot`, `shutdown`, history-rewriting git ops — describe the action and stop. Operator authorizes.
@@ -56,6 +66,8 @@ Bodies live in `.claude/commands/` and `.github/prompts/` respectively. `.claude
 - **Don't speculate about firmware root causes.** When evidence doesn't support a classification, say "unknown" and list what would disambiguate.
 - **Run `trunk fmt` before proposing a commit.** The `trunk_check` CI gate will reject unformatted code.
 - **`confirm=True` on destructive MCP tools is a real gate, not a formality.** Don't bypass it via auto-approve settings.
+- **Keep code comments minimal — one or two lines, max.** Comment only when the _why_ isn't obvious from the code; never restate what the next line does. No multi-paragraph block comments explaining straightforward changes. The diff and commit message carry the rationale; the code carries the behavior.
+- **Use `Throttle` for time-based rate limiting, not raw `millis()` math.** `src/mesh/Throttle.h` provides `Throttle::isWithinTimespanMs(lastMs, intervalMs)` (returns true while inside the cooldown) and `Throttle::execute(&lastMs, intervalMs, func)` (function-pointer form that updates the timestamp on fire). Use these for any "did N ms pass since X" check — raw `millis() > lastMs + N` is rollover-unsafe (breaks after ~49.7 days) and inconsistent with the rest of the codebase. The helpers compute `now - lastMs` with unsigned subtraction, which wraps correctly.

 ## Typical agent workflows

@@ -89,25 +101,44 @@ Sequence these; don't parallelize on the same port.

 ## Where to look

-| Path                              | What's there                                                                                         |
-| --------------------------------- | ---------------------------------------------------------------------------------------------------- |
-| `src/`                            | Firmware C++ source (`mesh/`, `modules/`, `platform/`, `graphics/`, `gps/`, `motion/`, `mqtt/`, …)   |
-| `src/mesh/`                       | Core: NodeDB, Router, Channels, CryptoEngine, radio interfaces, StreamAPI, PhoneAPI                  |
-| `src/modules/`                    | Feature modules; `Telemetry/Sensor/` has 50+ I2C sensor drivers                                      |
-| `variants/`                       | 200+ hardware variant definitions (`variant.h` + `platformio.ini` per board)                         |
-| `protobufs/`                      | `.proto` definitions; regenerate with `bin/regen-protos.sh`                                          |
-| `test/`                           | Firmware unit tests (12 suites; `pio test -e native`)                                                |
-| `mcp-server/`                     | Python MCP server + pytest hardware integration tests                                                |
-| `mcp-server/tests/`               | Tiered pytest suite: `unit/`, `mesh/`, `telemetry/`, `monitor/`, `fleet/`, `admin/`, `provisioning/` |
-| `.claude/commands/`               | Claude Code slash command bodies                                                                     |
-| `.github/prompts/`                | Copilot prompt bodies (mirrors of the Claude Code ones)                                              |
-| `.github/copilot-instructions.md` | **Primary agent instructions — read this**                                                           |
-| `.github/workflows/`              | CI pipelines                                                                                         |
-| `.mcp.json`                       | MCP server registration for Claude Code                                                              |
+| Path                              | What's there                                                                                                             |
+| --------------------------------- | ------------------------------------------------------------------------------------------------------------------------ |
+| `src/`                            | Firmware C++ source (`mesh/`, `modules/`, `platform/`, `graphics/`, `gps/`, `motion/`, `mqtt/`, …)                       |
+| `src/mesh/`                       | Core: NodeDB, Router, Channels, CryptoEngine, radio interfaces, StreamAPI, PhoneAPI                                      |
+| `src/modules/`                    | Feature modules; `Telemetry/Sensor/` has 50+ I2C sensor drivers                                                          |
+| `variants/`                       | 200+ hardware variant definitions (`variant.h` + `platformio.ini` per board)                                             |
+| `protobufs/`                      | `.proto` definitions; regenerate with `bin/regen-protos.sh`                                                              |
+| `test/`                           | Firmware unit tests (12 suites; `pio test -e native`)                                                                    |
+| `mcp-server/`                     | Python MCP server + pytest hardware integration tests                                                                    |
+| `mcp-server/tests/`               | Tiered pytest suite: `unit/`, `mesh/`, `telemetry/`, `monitor/`, `recovery/`, `ui/`, `fleet/`, `admin/`, `provisioning/` |
+| `.claude/commands/`               | Claude Code slash command bodies                                                                                         |
+| `.github/prompts/`                | Copilot prompt bodies (mirrors of the Claude Code ones)                                                                  |
+| `.github/copilot-instructions.md` | **Primary agent instructions — read this**                                                                               |
+| `.github/workflows/`              | CI pipelines                                                                                                             |
+| `.mcp.json`                       | MCP server registration for Claude Code                                                                                  |

 ## Recovery one-liners

 - **`userPrefs.jsonc` dirty after a test run?** Re-run `./mcp-server/run-tests.sh` once (pre-flight self-heals from the sidecar). If still dirty: `git checkout userPrefs.jsonc`.
 - **nRF52 not responding?** `mcp__meshtastic__touch_1200bps(port=...)` drops it into the DFU bootloader, then `pio_flash` re-installs.
+- **Device fully wedged (no DFU)?** `mcp__meshtastic__uhubctl_cycle(role="nrf52", confirm=True)` hard-power-cycles it via USB hub PPPS. Needs `uhubctl` installed (`brew install uhubctl` / `apt install uhubctl`); on Linux without udev rules, permission errors fail fast, so use `sudo uhubctl` yourself or configure udev access.
 - **Port busy?** `lsof <port>` to find the holder. Usually a stale `pio device monitor` or zombie `meshtastic_mcp` process. Kill it.
 - **Multiple MCP servers running?** `ps aux | grep meshtastic_mcp` — zombies hold ports. Kill all but the one your host spawned.
+- **macOS: `LIBUSB_ERROR_BUSY` on a CH341 LoRa adapter?** A third-party WCH `CH34xVCPDriver` is claiming interface 0. Find the bundle ID with `ioreg -p IOUSB -l -w 0 | grep -B2 -A30 0x5512`, then `sudo kmutil unload -b <bundleID>`. Apple's bundled CH34x kext targets the CH340 UART (PID 0x7523), not the SPI bridge — it's never the culprit.
+
+## Environment variables (test harness)
+
+| Var                                  | Purpose                                                                                                                                                                                                    |
+| ------------------------------------ | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `MESHTASTIC_MCP_ENV_<ROLE>`          | Override PlatformIO env for a role (e.g. `MESHTASTIC_MCP_ENV_NRF52=rak4631-dap`). Default map: `nrf52→rak4631`, `esp32s3→heltec-v3`.                                                                       |
+| `MESHTASTIC_MCP_SEED`                | PSK seed for the session test profile. Defaults to `mcp-<user>-<host>`.                                                                                                                                    |
+| `MESHTASTIC_MCP_FLASH_LOG`           | File path to tee pio/esptool/nrfutil/picotool output. `run-tests.sh` sets this to `tests/flash.log` so the TUI can stream live flash progress.                                                             |
+| `MESHTASTIC_MCP_TCP_HOST`            | `host` or `host:port` of a `meshtasticd` daemon (e.g. the `native-macos` build). Surfaces it in `list_devices` as `tcp://host:port` so `connect()`-based tools target it transparently. Default port 4403. |
+| `MESHTASTIC_UHUBCTL_BIN`             | Absolute path to `uhubctl` binary. Default: PATH lookup.                                                                                                                                                   |
+| `MESHTASTIC_UHUBCTL_LOCATION_<ROLE>` | Pin a role to a specific uhubctl hub location (e.g. `1-1.3`). Wins over VID auto-detection — use when multiple devices share a VID.                                                                        |
+| `MESHTASTIC_UHUBCTL_PORT_<ROLE>`     | Pin a role to a specific hub port number. Required alongside `LOCATION_<ROLE>`.                                                                                                                            |
+| `MESHTASTIC_UI_CAMERA_BACKEND`       | Camera backend for UI tier + `capture_screen` tool: `opencv` / `ffmpeg` / `null` / `auto` (default).                                                                                                       |
+| `MESHTASTIC_UI_CAMERA_DEVICE`        | Generic camera device (index or path). Used by the UI tier when no per-role var is set.                                                                                                                    |
+| `MESHTASTIC_UI_CAMERA_DEVICE_<ROLE>` | Per-role camera pinning (e.g. `MESHTASTIC_UI_CAMERA_DEVICE_ESP32S3=0` for the OLED-bearing heltec-v3).                                                                                                     |
+| `MESHTASTIC_UI_OCR_BACKEND`          | OCR engine selection: `easyocr` / `pytesseract` / `null` / `auto` (default).                                                                                                                               |
+| `MESHTASTIC_UI_TUI_CAMERA`           | Set to `1` to mount the live camera-feed panel in `meshtastic-mcp-test-tui`.                                                                                                                               |
@@ -3,15 +3,17 @@
 # trunk-ignore-all(hadolint/DL3008): Do not pin apt package versions
 # trunk-ignore-all(hadolint/DL3013): Do not pin pip package versions

-FROM python:3.14-slim-trixie AS builder
+FROM debian:trixie AS builder
 ARG PIO_ENV=native
 ENV DEBIAN_FRONTEND=noninteractive
 ENV TZ=Etc/UTC

 # Install Dependencies
 ENV PIP_ROOT_USER_ACTION=ignore
+ENV PIP_BREAK_SYSTEM_PACKAGES=1
 RUN apt-get update && apt-get install --no-install-recommends -y \
        curl wget g++ zip git ca-certificates pkg-config \
+        python3-pip python3-grpc-tools \
        libgpiod-dev libyaml-cpp-dev libbluetooth-dev libi2c-dev libuv1-dev \
        libusb-1.0-0-dev libulfius-dev liborcania-dev libssl-dev \
        libx11-dev libinput-dev libxkbcommon-x11-dev libsqlite3-dev libsdl2-dev \
@@ -3,12 +3,19 @@
 # trunk-ignore-all(hadolint/DL3018): Do not pin apk package versions
 # trunk-ignore-all(hadolint/DL3013): Do not pin pip package versions

-FROM python:3.14-alpine3.22 AS builder
+# Ensure the Alpine version is updated in both stages of the container!
+FROM alpine:3.23 AS builder
 ARG PIO_ENV=native
-ENV PIP_ROOT_USER_ACTION=ignore

+# Enable Alpine community repository (for 'py3-grpcio-tools')
+RUN echo "https://dl-cdn.alpinelinux.org/alpine/v$(cut -d. -f1,2 /etc/alpine-release)/community" >> /etc/apk/repositories
+
+# Install Dependencies
+ENV PIP_ROOT_USER_ACTION=ignore
+ENV PIP_BREAK_SYSTEM_PACKAGES=1
 RUN apk --no-cache add \
        bash g++ libstdc++-dev linux-headers zip git ca-certificates libbsd-dev \
+        py3-pip py3-grpcio-tools \
        libgpiod-dev yaml-cpp-dev bluez-dev \
        libusb-dev i2c-tools-dev libuv-dev openssl-dev pkgconf argp-standalone \
        libx11-dev libinput-dev libxkbcommon-dev sqlite-dev sdl2-dev \
@@ -60,4 +67,4 @@ EXPOSE 4403

 CMD [ "sh",  "-cx", "meshtasticd --fsdir=/var/lib/meshtasticd" ]

-HEALTHCHECK NONE
+HEALTHCHECK NONE
@@ -0,0 +1,64 @@
+#!/usr/bin/env python3
+"""Post-process protoc-generated Python files to live under a local namespace.
+
+Called by bin/regen-py-protos.sh. Walks the generated *_pb2.py files in the
+target directory and rewrites every `meshtastic` reference (imports, dotted
+attribute access) to use the new namespace (e.g., `meshtastic_v25`).
+
+Why: the .proto files declare `package meshtastic;`, so protoc emits
+`from meshtastic import mesh_pb2 as ...` lines. That would shadow the PyPI
+`meshtastic` package which other parts of the mcp-server depend on. Renaming
+to a local namespace keeps both available.
+
+Usage:
+    _rewrite_proto_namespace.py <generated_dir> <new_namespace>
+"""
+
+from __future__ import annotations
+
+import pathlib
+import re
+import sys
+
+
+def rewrite(dir_path: pathlib.Path, new_ns: str) -> int:
+    # Standard protoc import forms:
+    #   from meshtastic.X_pb2 import ...     (rare, for direct symbol pulls)
+    #   from meshtastic import X_pb2 as ...  (common, the cross-file ref)
+    #   import meshtastic.X_pb2              (also possible)
+    pattern_dotted_from = re.compile(r"^from meshtastic\.", re.MULTILINE)
+    pattern_bare_from = re.compile(r"^from meshtastic import ", re.MULTILINE)
+    pattern_dotted_import = re.compile(r"^import meshtastic\.", re.MULTILINE)
+
+    count = 0
+    for p in dir_path.glob("*.py"):
+        text = p.read_text(encoding="utf-8")
+        new = pattern_dotted_from.sub(f"from {new_ns}.", text)
+        new = pattern_bare_from.sub(f"from {new_ns} import ", new)
+        new = pattern_dotted_import.sub(f"import {new_ns}.", new)
+        # NOTE: we deliberately leave `meshtastic/X.proto` source-filename
+        # references inside descriptor strings alone. The descriptor pool is
+        # keyed by source filename (independent of Python package layout), so
+        # those don't collide with the PyPI package's descriptors.
+        if new != text:
+            p.write_text(new, encoding="utf-8")
+            count += 1
+    return count
+
+
+def main(argv: list[str]) -> int:
+    if len(argv) != 2:
+        print("usage: _rewrite_proto_namespace.py <generated_dir> <new_namespace>", file=sys.stderr)
+        return 2
+    dir_path = pathlib.Path(argv[0])
+    new_ns = argv[1]
+    if not dir_path.is_dir():
+        print(f"directory not found: {dir_path}", file=sys.stderr)
+        return 2
+    n = rewrite(dir_path, new_ns)
+    print(f"rewrote {n} file(s) in {dir_path} → namespace {new_ns}", file=sys.stderr)
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main(sys.argv[1:]))
@@ -38,4 +38,4 @@ cp bin/device-install.* $OUTDIR/
 cp bin/device-update.* $OUTDIR/

 echo "Copying manifest"
-cp $BUILDDIR/$basename.mt.json $OUTDIR/$basename.mt.json || true
+cp $BUILDDIR/$basename.mt.json $OUTDIR/$basename.mt.json
@@ -31,5 +31,6 @@ basename=meshtasticd-$1-$VERSION
 pio pkg install --environment "$PIO_ENV" || platformioFailed
 pio run --environment "$PIO_ENV" || platformioFailed

-cp "$BUILDDIR/meshtasticd" "$OUTDIR/meshtasticd_linux_$(uname -m)"
+os_name=$(uname -s | tr '[:upper:]' '[:lower:]')
+cp "$BUILDDIR/meshtasticd" "$OUTDIR/meshtasticd_${os_name}_$(uname -m)"
 cp bin/native-install.* $OUTDIR/
@@ -0,0 +1,30 @@
+---
+Lora:
+## Ebyte E80-900M22S
+## This is a bit experimental
+## 
+##
+  Module: lr1121
+  gpiochip: 1 # subtract 32 from the gpio numbers
+  DIO3_TCXO_VOLTAGE: 1.8
+  CS: 16 #pin6 / GPIO48 1C0
+  IRQ: 23  #pin17 / GPIO55 1C7
+  Busy: 22 #pin16 / GPIO54 1C6
+  Reset: 25 #pin13 / GPIO57 1D1
+
+
+  spidev: spidev0.0 #pins are (CS=16, CLK=17, MOSI=18, MISO=19)
+  spiSpeed: 2000000
+  
+rfswitch_table:
+  pins: [DIO5, DIO6, DIO7]
+  MODE_STBY: [LOW, LOW, LOW]
+  MODE_RX: [LOW, HIGH, LOW]
+  MODE_TX: [HIGH, HIGH, LOW]
+  MODE_TX_HP: [HIGH, LOW, LOW]
+  MODE_TX_HF: [LOW, LOW, LOW]
+  MODE_GNSS: [LOW, LOW, HIGH]
+  MODE_WIFI: [LOW, LOW, LOW]
+
+General:
+  MACAddressSource: eth0
@@ -0,0 +1,46 @@
+---
+Lora:
+## Ebyte E80-900M22S
+## This is a bit experimental
+## 
+##
+  Module: lr1121
+  gpiochip: 1 # subtract 32 from the gpio numbers
+  DIO3_TCXO_VOLTAGE: 1.8
+  CS: 16 #pin6 / GPIO48 1C0
+  IRQ: 23  #pin17 / GPIO55 1C7
+  Busy: 22 #pin16 / GPIO54 1C6
+  Reset: 25 #pin13 / GPIO57 1D1
+
+
+  spidev: spidev0.0 #pins are (CS=16, CLK=17, MOSI=18, MISO=19)
+  spiSpeed: 2000000
+  
+rfswitch_table:
+  pins:
+    - DIO5
+    - DIO6
+  MODE_STBY:
+    - LOW
+    - LOW
+  MODE_RX:
+    - HIGH
+    - LOW
+  MODE_TX:
+    - HIGH
+    - HIGH
+  MODE_TX_HP:
+    - LOW
+    - HIGH
+  MODE_TX_HF:
+    - LOW
+    - LOW
+  MODE_GNSS:
+    - LOW
+    - LOW
+  MODE_WIFI:
+    - LOW
+    - LOW
+
+General:
+  MACAddressSource: eth0
@@ -0,0 +1,30 @@
+---
+Lora:
+## Ebyte E80-900M22S
+## This is a bit experimental
+## 
+##
+  Module: lr1121
+  gpiochip: 1 # subtract 32 from the gpio numbers
+  DIO3_TCXO_VOLTAGE: 1.8
+  CS: 16 #pin6 / GPIO48 1C0
+  IRQ: 23  #pin17 / GPIO55 1C7
+  Busy: 22 #pin16 / GPIO54 1C6
+  Reset: 25 #pin13 / GPIO57 1D1
+
+
+  spidev: spidev0.0 #pins are (CS=16, CLK=17, MOSI=18, MISO=19)
+  spiSpeed: 2000000
+  
+rfswitch_table:
+  pins: [DIO5, DIO6, DIO7]
+  MODE_STBY: [LOW, LOW, LOW]
+  MODE_RX: [LOW, LOW, LOW]
+  MODE_TX: [LOW, HIGH, LOW]
+  MODE_TX_HP: [HIGH, LOW, LOW]
+  # MODE_TX_HF: []
+  # MODE_GNSS: []
+  MODE_WIFI: [LOW, LOW, LOW]
+
+General:
+  MACAddressSource: eth0
@@ -0,0 +1,31 @@
+# For use with Armbian luckfox-pico-max
+# Waveshare LoRa HAT for Raspberry Pi Pico
+# https://www.waveshare.com/wiki/Pico-LoRa-SX1262
+
+Meta:
+  name: luckfox-pico-max-ws-raspberry-pi-pico-hat
+  support: community
+  compatible:
+    - luckfox-pico-max # Armbian
+
+Lora:
+  Module: sx1262
+  DIO2_AS_RF_SWITCH: true
+  DIO3_TCXO_VOLTAGE: true
+  spidev: spidev0.0
+  Busy: # GPIO1_C7 / GP2
+    pin: 55
+    gpiochip: 1
+    line: 23
+  CS: # GPIO1_C6 / GP3
+    pin: 54
+    gpiochip: 1
+    line: 22
+  Reset: # GPIO1_D1 / GP15
+    pin: 57
+    gpiochip: 1
+    line: 25
+  IRQ: # GPIO2_A2 / GP20
+    pin: 66
+    gpiochip: 2
+    line: 2
@@ -0,0 +1,18 @@
+# Station G3 motherboard with a Raspberry Pi Zero 2W as the MCU daughterboard.
+# Verify spidev / I2C device paths for your OS — they may differ.
+Meta:
+  name: Station G3
+  support: community
+  compatible:
+    - raspberry-pi
+
+Lora:
+  Module: sx1262
+  IRQ: 22                       # BCM pin — wiki spec
+  Reset: 16                     # BCM pin — wiki spec
+  Busy: 24                      # BCM pin — wiki spec
+  # CS: 8                       # BCM 8 = SPI0 CE0 (default); uncomment only to override
+  DIO2_AS_RF_SWITCH: true
+  DIO3_TCXO_VOLTAGE: true
+  spidev: spidev0.0
+  # SX126X_MAX_POWER: 19        # matches Station G2 firmware cap; raise carefully per PA jumper mode
@@ -0,0 +1,439 @@
+#!/usr/bin/env python3
+"""Deterministic seed-data generator for the fake NodeDB fixture pipeline.
+
+Writes a JSONL file describing N fake-but-realistic Meshtastic peers.
+The output is hand-editable and committed; a sibling compile step
+(bin/seed-json-to-proto.py) turns it into a binary `meshtastic_NodeDatabase`
+v25 protobuf with fresh "now-relative" timestamps.
+
+Determinism contract:
+  Same --seed -> byte-identical JSONL output, regardless of wall clock.
+  All timestamps are stored as `*_offset_sec` (seconds before "now"); the
+  compile step resolves them to absolute epochs at compile time.
+
+Structural fields covered:
+  * NodeInfoLite header: num, long_name, short_name, hw_model, role,
+    public_key, snr, channel, hops_away, next_hop, bitfield flags
+  * PositionLite: lat/long Gaussian around --centroid, altitude, source
+  * DeviceMetrics: battery/voltage/util/uptime
+  * EnvironmentMetrics: temp/humidity/pressure/iaq
+  * StatusMessage: error_code (usually zero)
+
+Active-board allow-list:
+  hw_model values are restricted to the intersection of
+    (a) variants with `custom_meshtastic_support_level = 1` in
+        variants/*/*/platformio.ini, AND
+    (b) values present in the `HardwareModel` enum in mesh.proto.
+  See HW_MODEL_WEIGHTS below. Deprecated boards (legacy TLORA / Heltec V1-2 /
+  classic TBEAM / TBEAM_V0P7 / Nano G1 / etc.) and fuzzer-only sentinels
+  (PORTDUINO, ANDROID_SIM, DIY_V1, ...) are excluded.
+
+Active-role allow-list:
+  Excludes ROUTER_CLIENT (deprecated v2.3.15) and REPEATER (deprecated v2.7.11).
+"""
+
+from __future__ import annotations
+
+import argparse
+import datetime as _dt
+import json
+import math
+import pathlib
+import random
+import sys
+
+# --------------------------------------------------------------------------
+# Active-board allow-list (intersection of tier-1 variants + HardwareModel enum).
+# Refresh by running:
+#   for f in $(find variants -name 'platformio.ini' | xargs grep -lE 'custom_meshtastic_support_level = 1'); do
+#       grep custom_meshtastic_hw_model_slug $f | awk -F= '{print $2}' | tr -d ' ';
+#   done | sort -u | comm -12 - <(python3 -c "from meshtastic.protobuf.mesh_pb2 import HardwareModel; print('\\n'.join(HardwareModel.keys()))" | sort)
+# --------------------------------------------------------------------------
+HW_MODEL_WEIGHTS: dict[str, float] = {
+    "HELTEC_V3": 14.0,
+    "T_DECK": 9.0,
+    "HELTEC_V4": 8.0,
+    "RAK4631": 8.0,
+    "HELTEC_MESH_POCKET": 6.0,
+    "TRACKER_T1000_E": 5.0,
+    "HELTEC_MESH_NODE_T114": 5.0,
+    "T_DECK_PRO": 5.0,
+    "LILYGO_TBEAM_S3_CORE": 4.0,
+    "HELTEC_WIRELESS_PAPER": 4.0,
+    "HELTEC_WSL_V3": 3.0,
+    "T_ECHO": 3.0,
+    "HELTEC_WIRELESS_TRACKER": 3.0,
+    "HELTEC_WIRELESS_TRACKER_V2": 2.0,
+    "HELTEC_VISION_MASTER_E290": 2.0,
+    "HELTEC_MESH_SOLAR": 2.0,
+    "SEEED_WIO_TRACKER_L1": 2.0,
+    "T_LORA_PAGER": 1.5,
+    "HELTEC_VISION_MASTER_E213": 1.5,
+    "T_ECHO_PLUS": 1.0,
+    "MUZI_BASE": 1.0,
+    "WISMESH_TAP_V2": 1.0,
+    "THINKNODE_M2": 1.0,
+    "THINKNODE_M5": 1.0,
+    "TLORA_T3_S3": 1.0,
+    # Long tail (uniform low weight across remaining tier-1 boards):
+    "HELTEC_V4_R8": 0.3,
+    "HELTEC_VISION_MASTER_T190": 0.3,
+    "HELTEC_HT62": 0.3,
+    "HELTEC_MESH_NODE_T096": 0.3,
+    "M5STACK_C6L": 0.3,
+    "MINI_EPAPER_S3": 0.3,
+    "MUZI_R1_NEO": 0.3,
+    "NOMADSTAR_METEOR_PRO": 0.3,
+    "RAK3312": 0.3,
+    "RAK3401": 0.3,
+    "SEEED_SOLAR_NODE": 0.3,
+    "SEEED_WIO_TRACKER_L1_EINK": 0.3,
+    "SENSECAP_INDICATOR": 0.3,
+    "TBEAM_1_WATT": 0.3,
+    "THINKNODE_M1": 0.3,
+    "THINKNODE_M3": 0.3,
+    "THINKNODE_M6": 0.3,
+    "T_ECHO_LITE": 0.3,
+    "WISMESH_TAG": 0.3,
+    "WISMESH_TAP": 0.3,
+    "XIAO_NRF52_KIT": 0.3,
+    "CROWPANEL": 0.3,
+}
+
+# Non-deprecated roles only.
+ROLE_WEIGHTS: dict[str, float] = {
+    "CLIENT": 75.0,
+    "CLIENT_MUTE": 5.0,
+    "ROUTER": 7.0,
+    "TRACKER": 3.0,
+    "SENSOR": 2.0,
+    "CLIENT_HIDDEN": 2.0,
+    "ROUTER_LATE": 2.0,
+    "CLIENT_BASE": 2.0,
+    "TAK": 1.0,
+    "TAK_TRACKER": 0.5,
+    "LOST_AND_FOUND": 0.5,
+}
+
+# Name pools — 60 firsts × 60 lasts = 3600 combinations.
+FIRSTS = [
+    "Quick", "Brave", "Silent", "Wild", "Lone", "Bright", "Red", "Blue",
+    "Green", "Black", "White", "Iron", "Steel", "Copper", "Silver", "Gold",
+    "Stone", "River", "Forest", "Mountain", "Canyon", "Desert", "Storm", "Sky",
+    "Solar", "Lunar", "Dawn", "Dusk", "Misty", "Frosty", "Sunny", "Shady",
+    "Happy", "Sleepy", "Drowsy", "Sneaky", "Sharp", "Smooth", "Rough", "Loud",
+    "Soft", "Slow", "Fast", "Tall", "Short", "Old", "New", "Tiny",
+    "Giant", "Hidden", "Lost", "Found", "Wandering", "Roving", "Drifting", "Floating",
+    "Burning", "Frozen", "Whispering", "Howling",
+]
+LASTS = [
+    "Phoenix", "Lion", "Bear", "Wolf", "Hawk", "Eagle", "Fox", "Lynx",
+    "Cougar", "Coyote", "Raven", "Owl", "Crow", "Falcon", "Heron", "Crane",
+    "Otter", "Badger", "Bison", "Elk", "Moose", "Stag", "Doe", "Hare",
+    "Marmot", "Mole", "Beaver", "Squirrel", "Mustang", "Bronco", "Pony", "Colt",
+    "Cobra", "Viper", "Mamba", "Adder", "Gecko", "Iguana", "Tortoise", "Turtle",
+    "Salmon", "Trout", "Bass", "Pike", "Shark", "Whale", "Dolphin", "Seal",
+    "Cactus", "Yucca", "Sage", "Juniper", "Pine", "Cedar", "Aspen", "Oak",
+    "Bluff", "Mesa", "Arroyo", "Ridge",
+]
+
+# Brief callsign pool for licensed-looking suffixes.
+CALLSIGN_PREFIXES = ["KX", "WD", "N5", "KE", "AB", "W5", "K1", "KQ", "AE", "NM"]
+
+# Only emojis that fit in 4 UTF-8 bytes (no variation selectors). short_name's
+# nanopb max_size:5 (incl. NUL) limits content to 4 bytes. ❄️ / ☀️ would be
+# 6 bytes due to U+FE0F variation selector — explicitly excluded.
+EMOJI_SHORTNAMES = ["🦊", "🐺", "🦅", "🐢", "🌵", "🔥", "🌙",
+                    "🌊", "🗻", "🌲", "🦌", "🐝", "🦂", "🦉",
+                    "🦇", "🦋"]
+
+# --------------------------------------------------------------------------
+# Helpers
+# --------------------------------------------------------------------------
+
+NUM_RESERVED = 4         # firmware reserves 0..3 (per NodeDB constants)
+NUM_MAX_EXCLUSIVE = 0x80000000  # restrict to positive int32 range for readability
+
+
+def _weighted_choice(rng: random.Random, weights: dict[str, float]) -> str:
+    """Deterministic weighted pick. Uses sorted keys so dict order is fixed."""
+    keys = sorted(weights.keys())
+    totals = [weights[k] for k in keys]
+    return rng.choices(keys, weights=totals, k=1)[0]
+
+
+def _gen_long_name(rng: random.Random, is_licensed: bool) -> str:
+    base = f"{rng.choice(FIRSTS)} {rng.choice(LASTS)}"
+    if is_licensed:
+        prefix = rng.choice(CALLSIGN_PREFIXES)
+        # Two trailing alpha chars after the digit; keep within 25 - len(base) - 1
+        suffix = f" {prefix}{rng.randint(0,9)}{rng.choice('ABCDEFGHIJKLMNOPQRSTUVWXYZ')}{rng.choice('ABCDEFGHIJKLMNOPQRSTUVWXYZ')}"
+        # nanopb max_size:25 means C string fits 24 bytes + NUL.
+        if len(base) + len(suffix) <= 24:
+            base = base + suffix
+    # Hard cap to 24 chars (nanopb max_size:25 minus NUL).
+    return base[:24]
+
+
+def _gen_short_name(rng: random.Random, long_name: str) -> str:
+    # 10% emoji-only short_name
+    if rng.random() < 0.10:
+        return rng.choice(EMOJI_SHORTNAMES)
+    first_char = long_name[0].upper() if long_name else "X"
+    alphanums = "ABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789"
+    return first_char + "".join(rng.choices(alphanums, k=3))
+
+
+def _gen_hops_away(rng: random.Random) -> int:
+    # Geometric-ish: 0→55%, 1→25%, 2→12%, 3→5%, 4→2%, 5+→1%
+    r = rng.random()
+    if r < 0.55:
+        return 0
+    if r < 0.80:
+        return 1
+    if r < 0.92:
+        return 2
+    if r < 0.97:
+        return 3
+    if r < 0.99:
+        return 4
+    return rng.randint(5, 7)
+
+
+def _gen_position(
+    rng: random.Random,
+    centroid_lat: float,
+    centroid_lon: float,
+    spread_km: float,
+    last_heard_offset_sec: int,
+) -> dict:
+    # 1 deg ≈ 111 km at the equator; we use this as a flat approximation.
+    lat = centroid_lat + rng.gauss(0.0, spread_km / 111.0)
+    lon = centroid_lon + rng.gauss(0.0, spread_km / 111.0)
+    altitude = max(0, round(rng.gauss(1376.0, 250.0)))  # T or C valley floor + relief
+    # Position was reported up to 300s before last_heard.
+    time_offset_sec = last_heard_offset_sec + rng.randint(0, 300)
+    return {
+        "latitude": round(lat, 6),
+        "longitude": round(lon, 6),
+        "altitude": altitude,
+        "time_offset_sec": time_offset_sec,
+        "location_source": "LOC_INTERNAL",
+    }
+
+
+def _gen_telemetry(rng: random.Random) -> dict:
+    # 5% plugged-in (battery_level == 101); rest uniform [10..100].
+    if rng.random() < 0.05:
+        battery_level = 101
+        voltage = 4.20
+    else:
+        battery_level = rng.randint(10, 100)
+        voltage = round(3.3 + (battery_level / 100.0) * 0.9, 3)
+    # Beta distributions for low/right-skewed metrics; randomly draw via gammavariate.
+    def _beta(a: float, b: float) -> float:
+        x = rng.gammavariate(a, 1.0)
+        y = rng.gammavariate(b, 1.0)
+        return x / (x + y)
+    channel_utilization = round(_beta(2.0, 15.0) * 100.0, 2)
+    air_util_tx = round(_beta(1.5, 20.0) * 10.0, 3)
+    uptime_seconds = int(rng.expovariate(1.0 / 86400.0))
+    return {
+        "battery_level": battery_level,
+        "voltage": voltage,
+        "channel_utilization": channel_utilization,
+        "air_util_tx": air_util_tx,
+        "uptime_seconds": uptime_seconds,
+    }
+
+
+def _gen_environment(rng: random.Random) -> dict:
+    return {
+        "temperature": round(rng.gauss(22.0, 8.0), 2),
+        "relative_humidity": round(min(100.0, max(0.0, rng.gauss(55.0, 20.0))), 2),
+        "barometric_pressure": round(rng.gauss(1013.0, 8.0), 2),
+        "iaq": int(min(500, max(0, round(rng.gauss(50.0, 30.0))))),
+    }
+
+
+def _gen_status(rng: random.Random) -> dict:
+    # `StatusMessage` (mesh.proto:1445) has a single free-form `string status`.
+    # Most peers report a healthy short status; occasional alert string.
+    healthy = ["OK", "online", "active", "running", "ready", "nominal"]
+    alert = ["low-batt", "no-gps", "weak-signal", "rebooted", "offline-soon"]
+    if rng.random() < 0.92:
+        return {"status": rng.choice(healthy)}
+    return {"status": rng.choice(alert)}
+
+
+def _gen_node(
+    rng: random.Random,
+    num: int,
+    centroid_lat: float,
+    centroid_lon: float,
+    spread_km: float,
+    coverage: dict[str, float],
+    last_heard_mean_sec: int,
+    last_heard_max_sec: int,
+) -> dict:
+    is_licensed = rng.random() < 0.05
+    long_name = _gen_long_name(rng, is_licensed)
+    short_name = _gen_short_name(rng, long_name)
+    hw_model = _weighted_choice(rng, HW_MODEL_WEIGHTS)
+    role = _weighted_choice(rng, ROLE_WEIGHTS)
+    has_public_key = rng.random() < 0.92
+    public_key_hex = (
+        "".join(f"{rng.randint(0,255):02x}" for _ in range(32)) if has_public_key else ""
+    )
+    snr = round(max(-20.0, min(12.0, rng.gauss(6.0, 4.0))), 2)
+    channel = 0 if rng.random() < 0.90 else rng.randint(1, 7)
+    hops_away = _gen_hops_away(rng)
+    next_hop = rng.randint(0, 255) if hops_away > 0 else 0
+    last_heard_offset_sec = int(min(rng.expovariate(1.0 / last_heard_mean_sec), last_heard_max_sec))
+
+    bitfield = {
+        "has_user": True,
+        "is_favorite": rng.random() < 0.08,
+        "is_muted": rng.random() < 0.03,
+        "via_mqtt": rng.random() < 0.12,
+        "is_ignored": rng.random() < 0.01,
+        "is_licensed": is_licensed,
+        "has_is_unmessagable": True,
+        "is_unmessagable": rng.random() < 0.02,
+        "is_key_manually_verified": rng.random() < 0.04,
+    }
+
+    node: dict = {
+        "num": f"0x{num:08x}",
+        "long_name": long_name,
+        "short_name": short_name,
+        "hw_model": hw_model,
+        "role": role,
+        "public_key_hex": public_key_hex,
+        "snr": snr,
+        "channel": channel,
+        "hops_away": hops_away,
+        "next_hop": next_hop,
+        "last_heard_offset_sec": last_heard_offset_sec,
+        "bitfield": bitfield,
+        "position": (
+            _gen_position(rng, centroid_lat, centroid_lon, spread_km, last_heard_offset_sec)
+            if rng.random() < coverage["position"]
+            else None
+        ),
+        "telemetry": _gen_telemetry(rng) if rng.random() < coverage["telemetry"] else None,
+        "environment": _gen_environment(rng) if rng.random() < coverage["environment"] else None,
+        "status": _gen_status(rng) if rng.random() < coverage["status"] else None,
+    }
+    return node
+
+
+def _parse_my_node_num(s: str | None) -> int | None:
+    if s is None:
+        return None
+    s = s.strip()
+    if s.startswith("0x") or s.startswith("0X"):
+        return int(s, 16)
+    return int(s)
+
+
+def main(argv: list[str]) -> int:
+    p = argparse.ArgumentParser(
+        description="Deterministic JSONL seed for the fake NodeDB fixture.",
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter,
+    )
+    p.add_argument("--count", type=int, required=True, help="Number of fake nodes to emit.")
+    p.add_argument("--seed", type=int, required=True, help="Deterministic seed.")
+    p.add_argument("--out", required=True, help="Output JSONL path.")
+    p.add_argument(
+        "--centroid",
+        default="33.1284,-107.2528",
+        help="LAT,LON centroid (default: Truth or Consequences, NM).",
+    )
+    p.add_argument("--spread-km", type=float, default=60.0, help="Gaussian std-dev in km.")
+    p.add_argument("--position-coverage", type=float, default=0.85)
+    p.add_argument("--telemetry-coverage", type=float, default=0.70)
+    p.add_argument("--environment-coverage", type=float, default=0.25)
+    p.add_argument("--status-coverage", type=float, default=0.40)
+    p.add_argument("--my-node-num", default=None, help="Exclude this NodeNum from generated set (hex or dec).")
+    p.add_argument("--last-heard-mean-sec", type=int, default=3600)
+    p.add_argument("--last-heard-max-sec", type=int, default=7 * 86400)
+    args = p.parse_args(argv)
+
+    if args.count <= 0:
+        print("--count must be positive", file=sys.stderr)
+        return 2
+
+    try:
+        centroid_lat, centroid_lon = (float(s) for s in args.centroid.split(","))
+    except ValueError:
+        print(f"--centroid must be LAT,LON; got {args.centroid!r}", file=sys.stderr)
+        return 2
+
+    my_node_num = _parse_my_node_num(args.my_node_num)
+
+    rng = random.Random(args.seed)
+
+    # 1) Generate a unique deterministic set of NodeNums.
+    nums: set[int] = set()
+    while len(nums) < args.count:
+        n = rng.randrange(NUM_RESERVED, NUM_MAX_EXCLUSIVE)
+        if my_node_num is not None and n == my_node_num:
+            continue
+        nums.add(n)
+    ordered_nums = sorted(nums)  # sort to fix output order independent of set hash
+
+    # 2) Per-node generation (in num order, single RNG continues).
+    coverage = {
+        "position": args.position_coverage,
+        "telemetry": args.telemetry_coverage,
+        "environment": args.environment_coverage,
+        "status": args.status_coverage,
+    }
+    nodes = [
+        _gen_node(
+            rng,
+            n,
+            centroid_lat,
+            centroid_lon,
+            args.spread_km,
+            coverage,
+            args.last_heard_mean_sec,
+            args.last_heard_max_sec,
+        )
+        for n in ordered_nums
+    ]
+
+    # 3) Write JSONL.
+    out_path = pathlib.Path(args.out)
+    out_path.parent.mkdir(parents=True, exist_ok=True)
+    # `generated_at_iso` is informational; it does NOT affect determinism because
+    # we derive it from the seed, not from wall clock. (Same seed -> same string.)
+    generated_at = _dt.datetime.fromtimestamp(args.seed, tz=_dt.timezone.utc).isoformat().replace("+00:00", "Z")
+    meta = {
+        "_meta": {
+            "version": 25,
+            "seed": args.seed,
+            "count": args.count,
+            "centroid": [centroid_lat, centroid_lon],
+            "spread_km": args.spread_km,
+            "generated_at_iso": generated_at,
+            "my_node_num_excluded": (None if my_node_num is None else f"0x{my_node_num:08x}"),
+            "coverage": coverage,
+            "last_heard_mean_sec": args.last_heard_mean_sec,
+            "last_heard_max_sec": args.last_heard_max_sec,
+        }
+    }
+    with out_path.open("w", encoding="utf-8") as f:
+        # `ensure_ascii=False` so emoji short_names survive. `sort_keys=True` for
+        # determinism (insertion order varies by Python version otherwise).
+        f.write(json.dumps(meta, ensure_ascii=False, sort_keys=True) + "\n")
+        for node in nodes:
+            f.write(json.dumps(node, ensure_ascii=False, sort_keys=True) + "\n")
+
+    print(f"wrote {args.count} nodes to {out_path} ({out_path.stat().st_size} bytes)", file=sys.stderr)
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main(sys.argv[1:]))
@@ -87,6 +87,9 @@
  </screenshots>

  <releases>
+    <release version="2.7.24" date="2026-05-08">
+      <url type="details">https://github.com/meshtastic/firmware/releases?q=tag%3Av2.7.24</url>
+    </release>
    <release version="2.7.23" date="2026-04-14">
      <url type="details">https://github.com/meshtastic/firmware/releases?q=tag%3Av2.7.23</url>
    </release>
@@ -293,9 +293,12 @@ if ("HAS_TFT", 1) in env.get("CPPDEFINES", []):
 board_arch = infer_architecture(env.BoardConfig())
 should_skip_manifest = board_arch is None

-# For host/native envs, avoid depending on 'buildprog' (some targets don't define it)
-mtjson_deps = [] if should_skip_manifest else ["buildprog"]
-if not should_skip_manifest and platform.name == "espressif32":
+# Most platforms can generate the manifest as part of the default 'buildprog' target.
+# Typically this passes success/failure properly.
+mtjson_deps = ["buildprog"]
+if platform.name == "espressif32":
+    # On ESP32, we need to explicitly depend upon the binary to prevent fake-success upon failure.
+    mtjson_deps = ["$BUILD_DIR/${PROGNAME}.bin"]
    # Build littlefs image as part of mtjson target
    # Equivalent to `pio run -t buildfs`
    target_lfs = env.DataToBin(
@@ -309,7 +312,8 @@ if should_skip_manifest:

    env.AddCustomTarget(
        name="mtjson",
-        dependencies=mtjson_deps,
+        # For host/native envs, avoid depending on 'buildprog' (some targets don't define it)
+        dependencies=[],
        actions=[skip_manifest],
        title="Meshtastic Manifest (skipped)",
        description="mtjson generation is skipped for native environments",
@@ -0,0 +1,73 @@
+#!/usr/bin/env bash
+# Regenerate the fake-NodeDB fixtures: produces 250 / 500 / 1000 / 2000-node
+# JSONL seed files + their compiled v25 protobufs.
+#
+# Layout:
+#   test/fixtures/nodedb/seed_v25_<N>.jsonl   — COMMITTED, hand-editable.
+#   build/fixtures/nodedb/nodes_v25_<N>.proto — .gitignored, build artifact.
+#                                               Drop into /prefs/nodes.proto.
+#
+# Daily use:  ./bin/regen-fake-nodedbs.sh
+#   - Recompiles protos from committed seeds (fresh wall-clock timestamps).
+# Intentional seed bump:  REGEN_SEEDS=yes ./bin/regen-fake-nodedbs.sh
+#   - Overwrites the committed JSONL files with freshly-seeded data.
+
+set -euo pipefail
+cd "$(dirname "$0")/.."
+
+# 1) Make sure the Python protobuf bindings exist (in-tree generation; .gitignored).
+if [[ ! -d bin/_generated/meshtastic ]]; then
+    echo "regenerating Python protobuf bindings (one-time)..."
+    ./bin/regen-py-protos.sh
+fi
+
+# 2) Pick a Python interpreter that has the meshtastic deps installed.
+#    Prefer the mcp-server venv (most likely to be set up by the operator).
+PY="python3"
+for cand in mcp-server/.venv/bin/python3 .venv/bin/python3; do
+    if [[ -x "$cand" ]]; then
+        PY="$cand"
+        break
+    fi
+done
+
+# 3) Pinned seeds per size — bump only when you intentionally want different
+#    structural data committed. Parallel arrays so the script works on
+#    macOS bash 3.2 (no `declare -A`).
+SIZES=(250 500 1000 2000)
+SEEDS=(20260511 20260512 20260513 20260514)
+
+REGEN_SEEDS="${REGEN_SEEDS:-no}"
+
+mkdir -p build/fixtures/nodedb test/fixtures/nodedb
+
+for i in 0 1 2 3; do
+    n="${SIZES[$i]}"
+    seed="${SEEDS[$i]}"
+    jsonl=$(printf "test/fixtures/nodedb/seed_v25_%04d.jsonl" "$n")
+    proto=$(printf "build/fixtures/nodedb/nodes_v25_%04d.proto" "$n")
+
+    if [[ "$REGEN_SEEDS" == "yes" || ! -f "$jsonl" ]]; then
+        $PY bin/gen-fake-nodedb-seed.py \
+            --count "$n" \
+            --seed "$seed" \
+            --out "$jsonl" \
+            --centroid 33.1284,-107.2528 \
+            --spread-km 60 \
+            --position-coverage 0.85 \
+            --telemetry-coverage 0.70 \
+            --environment-coverage 0.25 \
+            --status-coverage 0.40
+        echo "  seed:   $jsonl ($(wc -c < "$jsonl") bytes)"
+    fi
+
+    $PY bin/seed-json-to-proto.py --in "$jsonl" --out "$proto"
+    echo "  proto:  $proto ($(wc -c < "$proto") bytes)"
+done
+
+echo ""
+echo "Done. To load on Portduino native:"
+echo "  cp build/fixtures/nodedb/nodes_v25_1000.proto ~/.portduino/default/prefs/nodes.proto"
+echo ""
+echo "To push to a hardware device:"
+echo "  Use the mcp-server tool: push_fake_nodedb(size=1000, target=\"hardware\", port=\"/dev/cu.usbmodemXXXX\", confirm=True)"
@@ -0,0 +1,51 @@
+#!/usr/bin/env bash
+# Regenerate Python protobuf bindings from the in-tree `protobufs/` submodule
+# into `bin/_generated/`. Called by bin/regen-fake-nodedbs.sh; also useful as
+# a standalone refresh after any change to a .proto file.
+#
+# Output is .gitignored — bindings are a build artifact.
+#
+# Namespace rewrite:
+# The .proto files declare `package meshtastic;`, which makes protoc emit
+# imports like `from meshtastic import mesh_pb2`. That conflicts with the
+# PyPI `meshtastic` package (which the mcp-server relies on for its
+# SerialInterface/BLEInterface transport). We post-process the generated
+# files to live under `meshtastic_v25` instead — both the directory layout
+# and all internal imports — so they coexist cleanly with the PyPI package.
+
+set -euo pipefail
+cd "$(dirname "$0")/.."
+
+if ! command -v protoc >/dev/null 2>&1; then
+    echo "ERROR: protoc not found in PATH." >&2
+    echo "  macOS: brew install protobuf" >&2
+    echo "  Ubuntu/Debian: apt install protobuf-compiler" >&2
+    exit 1
+fi
+
+OUT=bin/_generated
+LOCAL_NS=meshtastic_v25
+
+rm -rf "$OUT"
+mkdir -p "$OUT"
+
+# 1) Generate from the in-tree protos. nanopb.proto first so its descriptor
+#    is available for the [(nanopb).*] options on other messages.
+protoc \
+    --proto_path=protobufs \
+    --python_out="$OUT" \
+    protobufs/nanopb.proto \
+    protobufs/meshtastic/*.proto
+
+# 2) Move the generated `meshtastic/` directory to `meshtastic_v25/`.
+mv "$OUT/meshtastic" "$OUT/$LOCAL_NS"
+
+# 3) Rewrite internal imports: any reference to `meshtastic.X_pb2` or
+#    `from meshtastic import X_pb2` becomes `meshtastic_v25.*`.
+python3 bin/_rewrite_proto_namespace.py "$OUT/$LOCAL_NS" "$LOCAL_NS"
+
+# 4) Make the package importable.
+touch "$OUT/__init__.py"
+touch "$OUT/$LOCAL_NS/__init__.py"
+
+echo "regenerated Python protobuf bindings -> $OUT/$LOCAL_NS/ (namespace: $LOCAL_NS)" >&2
@@ -0,0 +1,342 @@
+#!/usr/bin/env python3
+"""Compile a committed seed JSONL into a binary meshtastic_NodeDatabase v25 proto.
+
+The input is produced by `bin/gen-fake-nodedb-seed.py`. Timestamps in the JSONL
+are stored as `*_offset_sec` (seconds before "now"); this script resolves them
+to absolute epochs using `--now-epoch` (default: current wall clock).
+
+Output is a raw `pb_encode`-compatible binary that can be dropped at
+`/prefs/nodes.proto` on the device (Portduino prefs dir or hardware via
+XModem) and loaded by `NodeDB::loadFromDisk` at boot.
+
+Wire format reference:
+  protobufs/meshtastic/deviceonly.proto  (NodeDatabase, NodeInfoLite, sat entries)
+  src/mesh/NodeDB.h:467-484              (bitfield bit positions)
+  src/mesh/NodeDB.cpp:1523-1524          (pb_decode entry point)
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import pathlib
+import sys
+import time
+from typing import Any
+
+# Prefer the in-tree generated Python protobuf bindings (bin/_generated/meshtastic_v25/)
+# because the firmware branch's protos (v25 NodeDatabase satellite arrays, slim
+# NodeInfoLite) are typically newer than what the PyPI `meshtastic` package
+# ships. Run `bin/regen-py-protos.sh` to (re)generate.
+#
+# Namespace note: the local bindings live under `meshtastic_v25` (NOT `meshtastic`)
+# to avoid shadowing the PyPI `meshtastic` package — bin/regen-py-protos.sh
+# post-processes the protoc output to rename the package.
+_HERE = pathlib.Path(__file__).resolve().parent
+_LOCAL_PROTO_DIR = _HERE / "_generated"
+if _LOCAL_PROTO_DIR.is_dir():
+    sys.path.insert(0, str(_LOCAL_PROTO_DIR))
+
+try:
+    from meshtastic_v25.deviceonly_pb2 import (  # type: ignore[import-not-found]
+        NodeDatabase,
+        NodeInfoLite,
+        NodePositionEntry,
+        NodeTelemetryEntry,
+        NodeEnvironmentEntry,
+        NodeStatusEntry,
+        PositionLite,
+    )
+    from meshtastic_v25.mesh_pb2 import HardwareModel, Position, StatusMessage  # type: ignore[import-not-found]
+    from meshtastic_v25.config_pb2 import Config  # type: ignore[import-not-found]
+    from meshtastic_v25.telemetry_pb2 import DeviceMetrics, EnvironmentMetrics  # type: ignore[import-not-found]
+except ImportError as local_err:
+    # Fall back to the PyPI package if in-tree bindings haven't been generated.
+    # Will fail the v25 assertion below if the PyPI package predates the
+    # satellite-DB schema, but at least gives a clear "run regen-py-protos.sh"
+    # error message instead of an opaque ImportError.
+    try:
+        from meshtastic.protobuf.deviceonly_pb2 import (
+            NodeDatabase,
+            NodeInfoLite,
+            NodePositionEntry,
+            NodeTelemetryEntry,
+            NodeEnvironmentEntry,
+            NodeStatusEntry,
+            PositionLite,
+        )
+        from meshtastic.protobuf.mesh_pb2 import HardwareModel, Position, StatusMessage
+        from meshtastic.protobuf.config_pb2 import Config
+        from meshtastic.protobuf.telemetry_pb2 import DeviceMetrics, EnvironmentMetrics
+    except ImportError as pypi_err:
+        print(
+            "ERROR: could not import meshtastic protobuf bindings.\n"
+            "  In-tree generation: run `bin/regen-py-protos.sh` (requires protoc).\n"
+            "  PyPI fallback: `pip install meshtastic` (may lag firmware branch).\n"
+            f"  local error (meshtastic_v25): {local_err}\n"
+            f"  pypi error  (meshtastic.protobuf): {pypi_err}",
+            file=sys.stderr,
+        )
+        sys.exit(1)
+
+# Fail loudly if bindings predate v25 (no satellite arrays).
+assert (
+    hasattr(NodeDatabase, "DESCRIPTOR")
+    and "positions" in NodeDatabase.DESCRIPTOR.fields_by_name
+), (
+    "Loaded meshtastic bindings are older than v25 (NodeDatabase.positions missing). "
+    "Run `bin/regen-py-protos.sh` against the in-tree protobufs/ submodule."
+)
+
+# ---------------------------------------------------------------------------
+# Bitfield bit positions (mirror src/mesh/NodeDB.h:467-484).
+# ---------------------------------------------------------------------------
+BIT_IS_KEY_MANUALLY_VERIFIED = 0
+BIT_IS_MUTED = 1
+BIT_VIA_MQTT = 2
+BIT_IS_FAVORITE = 3
+BIT_IS_IGNORED = 4
+BIT_HAS_USER = 5
+BIT_IS_LICENSED = 6
+BIT_IS_UNMESSAGABLE = 7
+BIT_HAS_IS_UNMESSAGABLE = 8
+
+BITFIELD_LAYOUT = (
+    # JSON key            bit position
+    ("is_key_manually_verified", BIT_IS_KEY_MANUALLY_VERIFIED),
+    ("is_muted", BIT_IS_MUTED),
+    ("via_mqtt", BIT_VIA_MQTT),
+    ("is_favorite", BIT_IS_FAVORITE),
+    ("is_ignored", BIT_IS_IGNORED),
+    ("has_user", BIT_HAS_USER),
+    ("is_licensed", BIT_IS_LICENSED),
+    ("is_unmessagable", BIT_IS_UNMESSAGABLE),
+    ("has_is_unmessagable", BIT_HAS_IS_UNMESSAGABLE),
+)
+
+
+def _pack_bitfield(bf: dict[str, bool]) -> int:
+    out = 0
+    for key, shift in BITFIELD_LAYOUT:
+        if bf.get(key, False):
+            out |= (1 << shift)
+    return out
+
+
+def _validate_node(node: dict[str, Any]) -> None:
+    """Friendly errors so hand-editors get clear feedback."""
+    if "num" not in node or not isinstance(node["num"], str):
+        raise ValueError(f"node missing/invalid 'num' (must be hex string): {node!r}")
+    if "long_name" not in node:
+        raise ValueError(f"node {node['num']}: missing 'long_name'")
+    if len(node["long_name"]) > 24:
+        raise ValueError(
+            f"node {node['num']}: long_name {node['long_name']!r} is "
+            f"{len(node['long_name'])} chars; max 24 (nanopb max_size:25 minus NUL)"
+        )
+    if "short_name" in node:
+        # short_name max_size:5 (incl. NUL) → 4 bytes of content.
+        # Char count is irrelevant — emojis with variation selectors (e.g. ❄️ = 6 B)
+        # would slip past a `len(str) > 4` check. Always measure bytes.
+        b = node["short_name"].encode("utf-8")
+        if len(b) > 4:
+            raise ValueError(
+                f"node {node['num']}: short_name {node['short_name']!r} is "
+                f"{len(b)} bytes UTF-8; max 4 (nanopb max_size:5 minus NUL)"
+            )
+    pk = node.get("public_key_hex", "")
+    if pk and len(pk) != 64:
+        raise ValueError(
+            f"node {node['num']}: public_key_hex must be 64 hex chars or empty; "
+            f"got {len(pk)} chars"
+        )
+    if pk:
+        try:
+            bytes.fromhex(pk)
+        except ValueError as e:
+            raise ValueError(f"node {node['num']}: public_key_hex is not valid hex: {e}")
+
+
+def _resolve_time(
+    node: dict[str, Any],
+    field_absolute: str,
+    field_offset: str,
+    now_epoch: int,
+) -> int:
+    """If `field_absolute` is set, use it; else compute `now_epoch - offset`."""
+    if field_absolute in node and node[field_absolute] is not None:
+        return int(node[field_absolute])
+    offset = node.get(field_offset, 0)
+    return max(0, int(now_epoch) - int(offset))
+
+
+def _build_node_info_lite(node: dict[str, Any], now_epoch: int) -> NodeInfoLite:
+    _validate_node(node)
+    info = NodeInfoLite()
+    info.num = int(node["num"], 16) if isinstance(node["num"], str) else int(node["num"])
+    info.long_name = node.get("long_name", "")
+    info.short_name = node.get("short_name", "")
+    # Enum lookups will raise ValueError on unknown names — that's exactly what we want.
+    info.hw_model = HardwareModel.Value(node.get("hw_model", "UNSET"))
+    info.role = Config.DeviceConfig.Role.Value(node.get("role", "CLIENT"))
+    pk_hex = node.get("public_key_hex", "")
+    if pk_hex:
+        info.public_key = bytes.fromhex(pk_hex)
+    info.snr = float(node.get("snr", 0.0))
+    info.channel = int(node.get("channel", 0))
+    if "hops_away" in node:
+        # `optional uint32 hops_away = 9;` — in Python protobuf, assigning the
+        # field implicitly sets HasField("hops_away") to True. No has_hops_away
+        # setter exists (unlike the C++ nanopb-generated header).
+        info.hops_away = int(node["hops_away"])
+    info.next_hop = int(node.get("next_hop", 0))
+    info.last_heard = _resolve_time(node, "last_heard", "last_heard_offset_sec", now_epoch)
+    info.bitfield = _pack_bitfield(node.get("bitfield", {}))
+    return info
+
+
+def _build_position_entry(num: int, pos: dict[str, Any], now_epoch: int) -> NodePositionEntry:
+    entry = NodePositionEntry()
+    entry.num = num
+    pl = PositionLite()
+    # Firmware stores lat/long as int32 in 1e-7 degrees.
+    pl.latitude_i = int(round(float(pos["latitude"]) * 1e7))
+    pl.longitude_i = int(round(float(pos["longitude"]) * 1e7))
+    pl.altitude = int(pos.get("altitude", 0))
+    pl.time = _resolve_time(pos, "time", "time_offset_sec", now_epoch)
+    pl.location_source = Position.LocSource.Value(pos.get("location_source", "LOC_UNSET"))
+    entry.position.CopyFrom(pl)
+    return entry
+
+
+def _build_telemetry_entry(num: int, tel: dict[str, Any]) -> NodeTelemetryEntry:
+    entry = NodeTelemetryEntry()
+    entry.num = num
+    dm = DeviceMetrics()
+    if "battery_level" in tel:
+        dm.battery_level = int(tel["battery_level"])
+    if "voltage" in tel:
+        dm.voltage = float(tel["voltage"])
+    if "channel_utilization" in tel:
+        dm.channel_utilization = float(tel["channel_utilization"])
+    if "air_util_tx" in tel:
+        dm.air_util_tx = float(tel["air_util_tx"])
+    if "uptime_seconds" in tel:
+        dm.uptime_seconds = int(tel["uptime_seconds"])
+    entry.device_metrics.CopyFrom(dm)
+    return entry
+
+
+def _build_environment_entry(num: int, env: dict[str, Any]) -> NodeEnvironmentEntry:
+    entry = NodeEnvironmentEntry()
+    entry.num = num
+    em = EnvironmentMetrics()
+    if "temperature" in env:
+        em.temperature = float(env["temperature"])
+    if "relative_humidity" in env:
+        em.relative_humidity = float(env["relative_humidity"])
+    if "barometric_pressure" in env:
+        em.barometric_pressure = float(env["barometric_pressure"])
+    if "iaq" in env:
+        em.iaq = int(env["iaq"])
+    entry.environment_metrics.CopyFrom(em)
+    return entry
+
+
+def _build_status_entry(num: int, status: dict[str, Any]) -> NodeStatusEntry:
+    # `StatusMessage` (mesh.proto:1445) has a single `string status` field.
+    entry = NodeStatusEntry()
+    entry.num = num
+    sm = StatusMessage()
+    if "status" in status:
+        sm.status = str(status["status"])
+    entry.status.CopyFrom(sm)
+    return entry
+
+
+def compile_jsonl_to_proto(jsonl_path: pathlib.Path, now_epoch: int) -> bytes:
+    """Read a seed JSONL and return the encoded NodeDatabase bytes."""
+    lines = jsonl_path.read_text(encoding="utf-8").splitlines()
+    if not lines:
+        raise ValueError(f"{jsonl_path} is empty")
+    meta_line = lines[0]
+    meta_obj = json.loads(meta_line)
+    meta = meta_obj.get("_meta", {})
+    version = meta.get("version")
+    if version != 25:
+        raise ValueError(
+            f"{jsonl_path}: meta version is {version!r}; this compiler "
+            f"requires version=25. Regenerate the seed with the matching tooling."
+        )
+
+    db = NodeDatabase()
+    db.version = 25
+
+    for ln, raw in enumerate(lines[1:], start=2):
+        raw = raw.strip()
+        if not raw:
+            continue
+        try:
+            node = json.loads(raw)
+        except json.JSONDecodeError as e:
+            raise ValueError(f"{jsonl_path}:{ln} JSON parse error: {e}")
+
+        num = int(node["num"], 16) if isinstance(node["num"], str) else int(node["num"])
+
+        # Header
+        info = _build_node_info_lite(node, now_epoch)
+        db.nodes.append(info)
+
+        # Satellites (nullable)
+        if node.get("position"):
+            db.positions.append(_build_position_entry(num, node["position"], now_epoch))
+        if node.get("telemetry"):
+            db.telemetry.append(_build_telemetry_entry(num, node["telemetry"]))
+        if node.get("environment"):
+            db.environment.append(_build_environment_entry(num, node["environment"]))
+        if node.get("status"):
+            db.status.append(_build_status_entry(num, node["status"]))
+
+    return db.SerializeToString()
+
+
+def main(argv: list[str]) -> int:
+    p = argparse.ArgumentParser(
+        description="Compile a seed JSONL into a binary v25 NodeDatabase proto.",
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter,
+    )
+    p.add_argument("--in", dest="in_path", required=True, help="Input seed JSONL.")
+    p.add_argument("--out", required=True, help="Output binary .proto path.")
+    p.add_argument(
+        "--now-epoch",
+        type=int,
+        default=None,
+        help="Pin 'now' to this Unix epoch (for byte-identical CI). Default: time.time().",
+    )
+    args = p.parse_args(argv)
+
+    in_path = pathlib.Path(args.in_path)
+    if not in_path.is_file():
+        print(f"input not found: {in_path}", file=sys.stderr)
+        return 2
+
+    now_epoch = args.now_epoch if args.now_epoch is not None else int(time.time())
+
+    try:
+        encoded = compile_jsonl_to_proto(in_path, now_epoch)
+    except ValueError as e:
+        print(f"ERROR: {e}", file=sys.stderr)
+        return 3
+
+    out_path = pathlib.Path(args.out)
+    out_path.parent.mkdir(parents=True, exist_ok=True)
+    out_path.write_bytes(encoded)
+    print(
+        f"compiled {in_path} -> {out_path} ({len(encoded)} bytes, now_epoch={now_epoch})",
+        file=sys.stderr,
+    )
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main(sys.argv[1:]))
@@ -0,0 +1,118 @@
+#!/bin/bash
+
+# Script to show commits in develop that are not in master
+# with their associated PR info and commit hashes
+# 
+# Usage:
+#   ./show-unmerged-prs.sh            # Show all unmerged commits
+#   ./show-unmerged-prs.sh --bugfix   # Show only bugfix-labeled PRs
+
+set -e
+
+REPO="firmware"
+OWNER="meshtastic"
+BASE_BRANCH="master"
+HEAD_BRANCH="develop"
+LIMIT=100
+FILTER_LABEL=""
+
+# Parse arguments
+for arg in "$@"; do
+    case $arg in
+        --bugfix)
+            FILTER_LABEL="bugfix"
+            shift
+            ;;
+        --feature)
+            FILTER_LABEL="feature"
+            shift
+            ;;
+        --help)
+            echo "Usage: $0 [OPTIONS]"
+            echo "Options:"
+            echo "  --bugfix     Show only PRs labeled with 'bugfix'"
+            echo "  --feature    Show only PRs labeled with 'feature'"
+            echo "  --help       Show this help message"
+            exit 0
+            ;;
+    esac
+done
+
+if [ -n "$FILTER_LABEL" ]; then
+    echo "Fetching commits in $HEAD_BRANCH that are not in $BASE_BRANCH (filtered by label: $FILTER_LABEL)..."
+else
+    echo "Fetching commits in $HEAD_BRANCH that are not in $BASE_BRANCH..."
+fi
+echo ""
+
+# Check if gh CLI is available
+if ! command -v gh &> /dev/null; then
+    echo "ERROR: GitHub CLI (gh) not found. Please install it first."
+    echo "Visit: https://cli.github.com/"
+    exit 1
+fi
+
+# Get commits in develop that are not in master
+# For each commit, try to find associated PR
+git fetch origin develop master 2>/dev/null || true
+
+# Use git to get the list of commits
+commits=$(git log --pretty=format:"%H|%s" origin/master..origin/develop | head -n $LIMIT)
+
+count=0
+displayed=0
+echo "Commits in $HEAD_BRANCH not in $BASE_BRANCH:"
+echo "=============================================="
+echo ""
+
+while IFS='|' read -r hash subject; do
+    ((count++))
+    
+    # Try to find the PR for this commit
+    # Extract PR number, title, description, and labels
+    pr_response=$(gh api -X GET "/repos/$OWNER/$REPO/commits/$hash/pulls" \
+        -H "Accept: application/vnd.github.v3+json" 2>/dev/null | \
+        jq -r '.[0] | "\(.number)|\(.title)|\(.body // "No description")|\(.labels | map(.name) | join(","))"' 2>/dev/null || echo "||||")
+    
+    if [ -z "$pr_response" ] || [ "$pr_response" = "||||" ]; then
+        # If no PR found, skip if filter is active, otherwise show the commit
+        if [ -z "$FILTER_LABEL" ]; then
+            ((displayed++))
+            echo "[$displayed] Commit: $hash"
+            echo "    Subject: $subject"
+            echo "    PR: Not found in GitHub"
+            echo ""
+        fi
+    else
+        IFS='|' read -r pr_num pr_title pr_desc pr_labels <<< "$pr_response"
+        
+        # Check if filter matches
+        if [ -n "$FILTER_LABEL" ]; then
+            # Only show if the label is in the labels list
+            if ! echo "$pr_labels" | grep -q "$FILTER_LABEL"; then
+                continue
+            fi
+        fi
+        
+        ((displayed++))
+        echo "[$displayed] PR #$pr_num - $pr_title"
+        echo "    Commit: $hash"
+        if [ -n "$pr_desc" ] && [ "$pr_desc" != "No description" ]; then
+            # Truncate description to 200 chars
+            desc_short="${pr_desc:0:200}"
+            [ ${#pr_desc} -gt 200 ] && desc_short+="..."
+            echo "    Description: $desc_short"
+        fi
+        if [ -n "$pr_labels" ] && [ "$pr_labels" != "" ]; then
+            echo "    Labels: $pr_labels"
+        fi
+        echo ""
+    fi
+done <<< "$commits"
+
+echo ""
+if [ -n "$FILTER_LABEL" ]; then
+    echo "Done. Showing $displayed PRs with label '$FILTER_LABEL' from $displayed commits checked."
+else
+    echo "Done. Showing $displayed commits from $HEAD_BRANCH not in $BASE_BRANCH."
+fi
@@ -7,7 +7,7 @@
    "extra_flags": [
      "-D CDEBYTE_EORA_S3",
      "-D ARDUINO_USB_CDC_ON_BOOT=1",
-      "-D ARDUINO_USB_MODE=0",
+      "-D ARDUINO_USB_MODE=1",
      "-D ARDUINO_RUNNING_CORE=1",
      "-D ARDUINO_EVENT_RUNNING_CORE=1",
      "-D BOARD_HAS_PSRAM"
@@ -0,0 +1,42 @@
+{
+  "build": {
+    "arduino": {
+      "ldscript": "esp32s3_out.ld",
+      "memory_type": "qio_opi"
+    },
+    "core": "esp32",
+    "extra_flags": [
+      "-D BOARD_HAS_PSRAM",
+      "-D ARDUINO_USB_CDC_ON_BOOT=0",
+      "-D ARDUINO_USB_MODE=0",
+      "-D ARDUINO_RUNNING_CORE=1",
+      "-D ARDUINO_EVENT_RUNNING_CORE=0"
+    ],
+    "f_cpu": "240000000L",
+    "f_flash": "80000000L",
+    "flash_mode": "qio",
+    "psram_type": "qio_opi",
+    "hwids": [["0x303A", "0x1001"]],
+    "mcu": "esp32s3",
+    "variant": "ELECROW-ThinkNode-M7"
+  },
+  "connectivity": ["wifi", "bluetooth", "lora"],
+  "debug": {
+    "default_tool": "esp-builtin",
+    "onboard_tools": ["esp-builtin"],
+    "openocd_target": "esp32s3.cfg"
+  },
+  "frameworks": ["arduino", "espidf"],
+  "name": "ELECROW ThinkNode M7",
+  "upload": {
+    "flash_size": "8MB",
+    "maximum_ram_size": 524288,
+    "maximum_size": 8388608,
+    "use_1200bps_touch": true,
+    "wait_for_upload_port": true,
+    "require_upload_port": true,
+    "speed": 921600
+  },
+  "url": "https://www.elecrow.com",
+  "vendor": "ELECROW"
+}
@@ -6,7 +6,7 @@
    "core": "esp32",
    "extra_flags": [
      "-DARDUINO_USB_CDC_ON_BOOT=1",
-      "-DARDUINO_USB_MODE=0",
+      "-DARDUINO_USB_MODE=1",
      "-DARDUINO_RUNNING_CORE=1",
      "-DARDUINO_EVENT_RUNNING_CORE=1",
      "-DBOARD_HAS_PSRAM"
@@ -8,7 +8,7 @@
    "extra_flags": [
      "-DBOARD_HAS_PSRAM",
      "-DARDUINO_USB_CDC_ON_BOOT=1",
-      "-DARDUINO_USB_MODE=0",
+      "-DARDUINO_USB_MODE=1",
      "-DARDUINO_RUNNING_CORE=1",
      "-DARDUINO_EVENT_RUNNING_CORE=1"
    ],
@@ -0,0 +1,43 @@
+{
+  "build": {
+    "arduino": {
+      "ldscript": "esp32s3_out.ld",
+      "partitions": "default_16MB.csv",
+      "memory_type": "qio_opi"
+    },
+    "core": "esp32",
+    "extra_flags": [
+      "-DBOARD_HAS_PSRAM",
+      "-DARDUINO_USB_CDC_ON_BOOT=1",
+      "-DARDUINO_USB_MODE=1",
+      "-DARDUINO_RUNNING_CORE=1",
+      "-DARDUINO_EVENT_RUNNING_CORE=1"
+    ],
+    "f_cpu": "240000000L",
+    "f_flash": "80000000L",
+    "flash_mode": "qio",
+    "psram_type": "opi",
+    "hwids": [["0x303A", "0x1001"]],
+    "mcu": "esp32s3",
+    "variant": "heltec_v4_r8"
+  },
+  "connectivity": ["wifi", "bluetooth", "lora"],
+  "debug": {
+    "default_tool": "esp-builtin",
+    "onboard_tools": ["esp-builtin"],
+    "openocd_target": "esp32s3.cfg"
+  },
+  "frameworks": ["arduino", "espidf"],
+  "name": "heltec_wifi_lora_32 v4 r8 (16 MB FLASH, 8 MB PSRAM)",
+  "upload": {
+    "flash_size": "16MB",
+    "maximum_ram_size": 327680,
+    "maximum_size": 16777216,
+    "use_1200bps_touch": true,
+    "wait_for_upload_port": true,
+    "require_upload_port": true,
+    "speed": 921600
+  },
+  "url": "https://heltec.org/",
+  "vendor": "heltec"
+}
@@ -9,7 +9,7 @@
    "extra_flags": [
      "-DBOARD_HAS_PSRAM",
      "-DARDUINO_USB_CDC_ON_BOOT=1",
-      "-DARDUINO_USB_MODE=0",
+      "-DARDUINO_USB_MODE=1",
      "-DARDUINO_RUNNING_CORE=1",
      "-DARDUINO_EVENT_RUNNING_CORE=1"
    ],
@@ -9,7 +9,7 @@
    "extra_flags": [
      "-DBOARD_HAS_PSRAM",
      "-DARDUINO_USB_CDC_ON_BOOT=1",
-      "-DARDUINO_USB_MODE=0",
+      "-DARDUINO_USB_MODE=1",
      "-DARDUINO_RUNNING_CORE=1",
      "-DARDUINO_EVENT_RUNNING_CORE=1"
    ],
@@ -9,7 +9,7 @@
    "extra_flags": [
      "-DBOARD_HAS_PSRAM",
      "-DARDUINO_USB_CDC_ON_BOOT=1",
-      "-DARDUINO_USB_MODE=0",
+      "-DARDUINO_USB_MODE=1",
      "-DARDUINO_RUNNING_CORE=1",
      "-DARDUINO_EVENT_RUNNING_CORE=1"
    ],
@@ -8,7 +8,7 @@
    "extra_flags": [
      "-DHELTEC_WIRELESS_TRACKER",
      "-DARDUINO_USB_CDC_ON_BOOT=1",
-      "-DARDUINO_USB_MODE=0",
+      "-DARDUINO_USB_MODE=1",
      "-DARDUINO_RUNNING_CORE=1",
      "-DARDUINO_EVENT_RUNNING_CORE=1"
    ],
@@ -7,7 +7,7 @@
    "core": "esp32",
    "extra_flags": [
      "-DARDUINO_USB_CDC_ON_BOOT=1",
-      "-DARDUINO_USB_MODE=0",
+      "-DARDUINO_USB_MODE=1",
      "-DARDUINO_RUNNING_CORE=1",
      "-DARDUINO_EVENT_RUNNING_CORE=1"
    ],
@@ -0,0 +1,26 @@
+{
+  "build": {
+    "cpu": "cortex-m33",
+    "f_cpu": "128000000L",
+    "mcu": "nrf54l15",
+    "zephyr": {
+      "variant": "nrf54l15dk/nrf54l15/cpuapp"
+    }
+  },
+  "connectivity": ["bluetooth"],
+  "debug": {
+    "default_tools": ["jlink"],
+    "jlink_device": "nRF54L15_M33",
+    "svd_path": "nrf54l15.svd"
+  },
+  "frameworks": ["zephyr"],
+  "name": "Nordic nRF54L15-DK (PCA10156)",
+  "upload": {
+    "maximum_ram_size": 262144,
+    "maximum_size": 1572864,
+    "protocol": "jlink",
+    "protocols": ["jlink"]
+  },
+  "url": "https://www.nordicsemi.com/Products/nRF54L15",
+  "vendor": "Nordic Semiconductor"
+}
@@ -8,7 +8,7 @@
    "extra_flags": [
      "-DBOARD_HAS_PSRAM",
      "-DARDUINO_USB_CDC_ON_BOOT=1",
-      "-DARDUINO_USB_MODE=0",
+      "-DARDUINO_USB_MODE=1",
      "-DARDUINO_RUNNING_CORE=1",
      "-DARDUINO_EVENT_RUNNING_CORE=0"
    ],
@@ -0,0 +1,41 @@
+{
+  "build": {
+    "arduino": {
+      "ldscript": "esp32s3_out.ld",
+      "memory_type": "qio_opi"
+    },
+    "core": "esp32",
+    "extra_flags": [
+      "-DBOARD_HAS_PSRAM",
+      "-DARDUINO_USB_CDC_ON_BOOT=1",
+      "-DARDUINO_USB_MODE=1",
+      "-DARDUINO_RUNNING_CORE=1",
+      "-DARDUINO_EVENT_RUNNING_CORE=0"
+    ],
+    "f_cpu": "240000000L",
+    "f_flash": "80000000L",
+    "flash_mode": "qio",
+    "hwids": [["0x303A", "0x1001"]],
+    "mcu": "esp32s3",
+    "variant": "station-g3"
+  },
+  "connectivity": ["wifi", "bluetooth", "lora"],
+  "debug": {
+    "default_tool": "esp-builtin",
+    "onboard_tools": ["esp-builtin"],
+    "openocd_target": "esp32s3.cfg"
+  },
+  "frameworks": ["arduino", "espidf"],
+  "name": "BQ Station G3",
+  "upload": {
+    "flash_size": "16MB",
+    "maximum_ram_size": 327680,
+    "maximum_size": 16777216,
+    "use_1200bps_touch": true,
+    "wait_for_upload_port": true,
+    "require_upload_port": true,
+    "speed": 921600
+  },
+  "url": "",
+  "vendor": "BQ Consulting"
+}
@@ -9,7 +9,7 @@
      "-DBOARD_HAS_PSRAM",
      "-DLILYGO_TBEAM_1W",
      "-DARDUINO_USB_CDC_ON_BOOT=1",
-      "-DARDUINO_USB_MODE=0",
+      "-DARDUINO_USB_MODE=1",
      "-DARDUINO_RUNNING_CORE=1",
      "-DARDUINO_EVENT_RUNNING_CORE=1"
    ],
@@ -8,7 +8,7 @@
    "extra_flags": [
      "-DBOARD_HAS_PSRAM",
      "-DARDUINO_USB_CDC_ON_BOOT=1",
-      "-DARDUINO_USB_MODE=0",
+      "-DARDUINO_USB_MODE=1",
      "-DARDUINO_RUNNING_CORE=1",
      "-DARDUINO_EVENT_RUNNING_CORE=1"
    ],
@@ -8,7 +8,7 @@
      "-DBOARD_HAS_PSRAM",
      "-DLILYGO_TBEAM_S3_CORE",
      "-DARDUINO_USB_CDC_ON_BOOT=1",
-      "-DARDUINO_USB_MODE=0",
+      "-DARDUINO_USB_MODE=1",
      "-DARDUINO_RUNNING_CORE=1",
      "-DARDUINO_EVENT_RUNNING_CORE=1"
    ],
@@ -7,7 +7,7 @@
    "extra_flags": [
      "-DLILYGO_T3S3_V1",
      "-DARDUINO_USB_CDC_ON_BOOT=1",
-      "-DARDUINO_USB_MODE=0",
+      "-DARDUINO_USB_MODE=1",
      "-DARDUINO_RUNNING_CORE=1",
      "-DARDUINO_EVENT_RUNNING_CORE=1",
      "-DBOARD_HAS_PSRAM"
@@ -10,7 +10,7 @@
      "-DBOARD_HAS_PSRAM",
      "-DUNPHONE_SPIN=9",
      "-DARDUINO_USB_CDC_ON_BOOT=1",
-      "-DARDUINO_USB_MODE=0",
+      "-DARDUINO_USB_MODE=1",
      "-DARDUINO_RUNNING_CORE=1",
      "-DARDUINO_EVENT_RUNNING_CORE=1"
    ],
@@ -1,3 +1,9 @@
+meshtasticd (2.7.24.0) unstable; urgency=medium
+
+  * Version 2.7.24
+
+ -- GitHub Actions <github-actions[bot]@users.noreply.github.com>  Fri, 08 May 2026 10:44:12 +0000
+
 meshtasticd (2.7.23.0) unstable; urgency=medium

  * Version 2.7.23
@@ -1,4 +1,5 @@
 #!/usr/bin/bash
+set -e
 export DEBEMAIL="jbennett@incomsystems.biz"
 export PLATFORMIO_LIBDEPS_DIR=pio/libdeps
 export PLATFORMIO_PACKAGES_DIR=pio/packages
@@ -0,0 +1,7 @@
+# Name,   Type, SubType, Offset,  Size, Flags
+nvs,      data, nvs,     0x9000,  0x5000,
+otadata,  data, ota,     0xe000,  0x2000,
+app0,     app,  ota_0,   0x10000, 0x640000,
+app1,     app,  ota_1,   0x650000,0x640000,
+spiffs,   data, spiffs,  0xc90000,0x360000,
+coredump, data, coredump,0xFF0000,0x10000,
@@ -0,0 +1,7 @@
+# Name,   Type, SubType, Offset,  Size, Flags
+nvs,      data, nvs,     0x9000,  0x5000,
+otadata,  data, ota,     0xe000,  0x2000,
+app0,     app,  ota_0,   0x10000, 0x330000,
+app1,     app,  ota_1,   0x340000,0x330000,
+spiffs,   data, spiffs,  0x670000,0x180000,
+coredump, data, coredump,0x7F0000,0x10000,
@@ -70,17 +70,6 @@ def esp32_create_combined_bin(source, target, env):

 env.AddPostAction("$BUILD_DIR/${PROGNAME}.bin", esp32_create_combined_bin)

-esp32_kind = env.GetProjectOption("custom_esp32_kind")
-if esp32_kind == "esp32":
-    # Free up some IRAM by removing auxiliary SPI flash chip drivers.
-    # Wrapped stub symbols are defined in src/platform/esp32/iram-quirk.c.
-    env.Append(
-        LINKFLAGS=[
-            "-Wl,--wrap=esp_flash_chip_gd",
-            "-Wl,--wrap=esp_flash_chip_issi",
-            "-Wl,--wrap=esp_flash_chip_winbond",
-        ]
-    )
-else:
-    # For newer ESP32 targets, using newlib nano works better.
-    env.Append(LINKFLAGS=["--specs=nano.specs", "-u", "_printf_float"])
+# Enable Newlib Nano formatting to save space
+# ...but allow printf float support (compromise)
+env.Append(LINKFLAGS=["--specs=nano.specs", "-u", "_printf_float"])
@@ -0,0 +1,23 @@
+#!/usr/bin/env python3
+# trunk-ignore-all(ruff/F821)
+# trunk-ignore-all(flake8/F821): For SConstruct imports
+
+# force linker response file instead of command line arguments
+
+Import("env")
+
+
+def wrap_with_tempfile(command_key):
+    command = env.get(command_key)
+    if not command or not isinstance(command, str):
+        return
+    if "TEMPFILE(" in command:
+        return
+    env.Replace(**{command_key: "${TEMPFILE('%s')}" % command})
+
+
+# Force SCons to spill long commands into response files on this target.
+env.Replace(MAXLINELENGTH=8192)
+
+for key in ("LINKCOM", "CXXLINKCOM", "SHLINKCOM", "SHCXXLINKCOM"):
+    wrap_with_tempfile(key)
@@ -0,0 +1,140 @@
+#!/usr/bin/env python3
+# trunk-ignore-all(ruff/F821)
+# trunk-ignore-all(flake8/F821): For SConstruct imports
+#
+# post:extra_scripts/nrf54l15_linker.py
+#
+# Fix for Zephyr two-pass link on nRF54L15:
+# platformio-build.py registers env.Depends("$PROG_PATH", final_ld_script) but
+# the SCons dependency chain is broken (final_ld_script Command never runs).
+# This script adds a PreAction on the final firmware binary that runs the gcc
+# preprocessing command directly (extracted from build.ninja) to generate
+# zephyr/linker.cmd before the link step.
+#
+# PlatformIO bundles an old Ninja that can't handle multi-output depslog rules,
+# so we parse the COMMAND line from build.ninja and run just the gcc -E part,
+# skipping the cmake_transform_depfile step (only needed for Ninja deps tracking).
+
+import os
+import re
+import subprocess
+
+Import("env")
+
+if env.get("PIOENV") != "nrf54l15dk":
+    pass  # Only for the nrf54l15dk environment
+else:
+
+    def _extract_gcc_command(ninja_build):
+        """Parse build.ninja to find the gcc -E command that generates linker.cmd.
+
+        The rule format depends on the host:
+          Windows (CMake's RunCMake wraps every command):
+            COMMAND = cmd.exe /C "cd /D DIR && arm-none-eabi-gcc.exe ... -o linker.cmd && cmake.exe -E cmake_transform_depfile ..."
+          POSIX (Linux/macOS — no wrapper):
+            COMMAND = cd DIR && arm-none-eabi-gcc ... -o linker.cmd && cmake -E cmake_transform_depfile ...
+
+        Returns (gcc_cmd_string, cwd_path) or raises RuntimeError.
+        """
+        in_rule = False
+        with open(ninja_build, "r", encoding="utf-8", errors="replace") as f:
+            for line in f:
+                # Detect start of the linker.cmd custom command rule
+                if not in_rule:
+                    if "build zephyr/linker.cmd" in line and "CUSTOM_COMMAND" in line:
+                        in_rule = True
+                    continue
+
+                stripped = line.strip()
+                if not stripped.startswith("COMMAND = "):
+                    continue
+
+                command_val = stripped[len("COMMAND = ") :]
+
+                # On Windows the value is wrapped in `cmd.exe /C "..."` — strip
+                # the wrapper. On POSIX hosts the inner sequence is the value
+                # itself (no quoting layer).
+                m = re.search(r'/C\s+"(.*)"\s*$', command_val)
+                inner = m.group(1) if m else command_val
+                parts = inner.split(" && ")
+
+                cwd = None
+                gcc_cmd = None
+                for part in parts:
+                    part = part.strip()
+                    if part.startswith("cd /D "):  # Windows form
+                        cwd = part[len("cd /D ") :]
+                    elif part.startswith("cd "):  # POSIX form
+                        cwd = part[len("cd ") :]
+                    elif "arm-none-eabi-gcc" in part:
+                        gcc_cmd = part
+
+                if not gcc_cmd:
+                    raise RuntimeError(
+                        "nRF54L15 linker fix: arm-none-eabi-gcc command not found in:\n%s"
+                        % inner[:400]
+                    )
+
+                return gcc_cmd, cwd
+
+        raise RuntimeError(
+            "nRF54L15 linker fix: 'build zephyr/linker.cmd' rule not found in build.ninja"
+        )
+
+    def _generate_linker_cmd(target, source, env):
+        """Generate zephyr/linker.cmd via direct gcc invocation before the final link."""
+        build_dir = env.subst("$BUILD_DIR")
+        zephyr_dir = os.path.join(build_dir, "zephyr")
+        linker_cmd = os.path.join(zephyr_dir, "linker.cmd")
+
+        if os.path.exists(linker_cmd):
+            return  # Already present — nothing to do
+
+        ninja_build = os.path.join(build_dir, "build.ninja")
+        if not os.path.exists(ninja_build):
+            raise RuntimeError(
+                "nRF54L15 linker fix: build.ninja not found at %s\n"
+                "Run a full build first so CMake generates the Ninja files."
+                % ninja_build
+            )
+
+        gcc_cmd, cwd = _extract_gcc_command(ninja_build)
+        run_cwd = cwd if cwd else zephyr_dir
+
+        print(
+            "==> nRF54L15: Generating zephyr/linker.cmd (LINKER_ZEPHYR_FINAL) via GCC"
+        )
+        # gcc_cmd comes verbatim from our own build.ninja (never user input) and
+        # contains Windows-style paths with spaces that cannot be safely argv-split
+        # with shlex, so we run it via the platform shell. nosec/nosemgrep below
+        # acknowledge this deliberate, scoped use of shell=True.
+        result = subprocess.run(  # nosec B602
+            gcc_cmd,
+            shell=True,  # nosemgrep: python.lang.security.audit.subprocess-shell-true.subprocess-shell-true
+            cwd=run_cwd,
+            capture_output=True,
+            text=True,
+        )
+        if result.returncode != 0:
+            print("GCC stdout:", result.stdout[:2000])
+            print("GCC stderr:", result.stderr[:2000])
+            raise RuntimeError(
+                "nRF54L15 linker fix: GCC failed to generate linker.cmd (rc=%d)"
+                % result.returncode
+            )
+        if not os.path.exists(linker_cmd):
+            raise RuntimeError(
+                "nRF54L15 linker fix: GCC returned 0 but linker.cmd was not created at %s"
+                % linker_cmd
+            )
+        print("==> linker.cmd generated successfully")
+
+    # Use PIOMAINPROG (set by ZephyrBuildProgram) to get the exact SCons node
+    prog = env.get("PIOMAINPROG")
+    if prog:
+        env.AddPreAction(prog, _generate_linker_cmd)
+    else:
+        print(
+            "[nrf54l15_linker] WARNING: PIOMAINPROG not set, falling back to $PROG_PATH"
+        )
+        env.AddPreAction(env.subst("$PROG_PATH"), _generate_linker_cmd)
@@ -7,6 +7,12 @@ __pycache__/
 dist/
 build/

+# Persistent device-log capture (recorder + Datadog cursor).
+# Cross-session JSONL streams written by the autouse Recorder singleton
+# (see src/meshtastic_mcp/recorder/). Lives outside tests/ so the pytest
+# fixture truncate doesn't touch it.
+.mtlog/
+
 # Test harness artifacts
 tests/report.html
 tests/junit.xml
@@ -24,3 +30,6 @@ tests/.tui-runs
 tests/.history/
 # Reproducer bundles (TUI `x` export on failed tests).
 tests/reproducers/
+# UI-tier camera captures + per-test transcripts. Regenerated every run;
+# left on disk for human review between runs.
+tests/ui_captures/
@@ -61,7 +61,7 @@ Replace `<firmware-repo>` with the absolute path, e.g. `/Users/you/GitHub/firmwa

 Same `mcpServers` block, but in `~/Library/Application Support/Claude/claude_desktop_config.json` (macOS) or `%APPDATA%\Claude\claude_desktop_config.json` (Windows).

-## Tools (38)
+## Tools (43)

 ### Discovery & metadata

@@ -130,6 +130,34 @@ _The tool tables below document 38 currently registered MCP server tools._
 | `picotool_load`       | Load a UF2                                                |
 | `picotool_raw`        | Pass-through                                              |

+### USB power control (uhubctl)
+
+| Tool            | What it does                                                |
+| --------------- | ----------------------------------------------------------- |
+| `uhubctl_list`  | Enumerate USB hubs + attached-device VID/PID (read-only)    |
+| `uhubctl_power` | Drive a hub port `on` or `off`; `off` requires confirm=True |
+| `uhubctl_cycle` | Off → wait `delay_s` → on; confirm=True required            |
+
+Target a port by explicit `(location, port)` (raw uhubctl syntax like
+`location="1-1.3", port=2`) or by `role` (`"nrf52"`, `"esp32s3"`). Role
+lookup checks `MESHTASTIC_UHUBCTL_LOCATION_<ROLE>` +
+`MESHTASTIC_UHUBCTL_PORT_<ROLE>` env vars first, then auto-detects via VID
+against `uhubctl`'s output.
+
+Requires [`uhubctl`](https://github.com/mvp/uhubctl) on PATH:
+
+```bash
+brew install uhubctl        # macOS
+apt install uhubctl         # Debian/Ubuntu
+```
+
+Modern macOS + PPPS-capable hubs generally work without root. On Linux
+without udev rules, or on old macOS with driver quirks, you may need
+`sudo`. If uhubctl returns a permission error the MCP tool raises a
+clear `UhubctlError` pointing at the
+[udev-rules / sudo fallback](https://github.com/mvp/uhubctl#linux-usb-permissions)
+rather than auto-`sudo`'ing mid-run.
+
 ## Safety

 - **All destructive flash/admin tools require `confirm=True`** as a tool-level gate, on top of any permission prompt from Claude.
@@ -138,15 +166,73 @@ _The tool tables below document 38 currently registered MCP server tools._

 ## Environment variables

-| Var                        | Default                                                     | Purpose                                                             |
-| -------------------------- | ----------------------------------------------------------- | ------------------------------------------------------------------- |
-| `MESHTASTIC_FIRMWARE_ROOT` | walks up from cwd for `platformio.ini`                      | Pin the firmware repo                                               |
-| `MESHTASTIC_PIO_BIN`       | `~/.platformio/penv/bin/pio` → `$PATH` `pio` → `platformio` | Override `pio` location                                             |
-| `MESHTASTIC_ESPTOOL_BIN`   | `<firmware>/.venv/bin/esptool` → `$PATH`                    | Override esptool                                                    |
-| `MESHTASTIC_NRFUTIL_BIN`   | `$PATH`                                                     | Override nrfutil                                                    |
-| `MESHTASTIC_PICOTOOL_BIN`  | `$PATH`                                                     | Override picotool                                                   |
-| `MESHTASTIC_MCP_SEED`      | `mcp-<user>-<host>`                                         | PSK seed for test-harness session (CI override)                     |
-| `MESHTASTIC_MCP_FLASH_LOG` | `<mcp-server>/tests/flash.log`                              | Tee target for pio/esptool/nrfutil subprocess output (TUI tails it) |
+| Var                        | Default                                                     | Purpose                                                                                                          |
+| -------------------------- | ----------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------- |
+| `MESHTASTIC_FIRMWARE_ROOT` | walks up from cwd for `platformio.ini`                      | Pin the firmware repo                                                                                            |
+| `MESHTASTIC_PIO_BIN`       | `~/.platformio/penv/bin/pio` → `$PATH` `pio` → `platformio` | Override `pio` location                                                                                          |
+| `MESHTASTIC_ESPTOOL_BIN`   | `<firmware>/.venv/bin/esptool` → `$PATH`                    | Override esptool                                                                                                 |
+| `MESHTASTIC_NRFUTIL_BIN`   | `$PATH`                                                     | Override nrfutil                                                                                                 |
+| `MESHTASTIC_PICOTOOL_BIN`  | `$PATH`                                                     | Override picotool                                                                                                |
+| `MESHTASTIC_MCP_SEED`      | `mcp-<user>-<host>`                                         | PSK seed for test-harness session (CI override)                                                                  |
+| `MESHTASTIC_MCP_FLASH_LOG` | `<mcp-server>/tests/flash.log`                              | Tee target for pio/esptool/nrfutil subprocess output (TUI tails it)                                              |
+| `MESHTASTIC_MCP_TCP_HOST`  | unset                                                       | `host` or `host:port` of a `meshtasticd` daemon to surface as a TCP device (see "TCP / native-host nodes" below) |
+
+## TCP / native-host nodes
+
+The `native-macos` and `native` PlatformIO envs build a headless `meshtasticd`
+binary that runs on the host (Apple Silicon / Intel macOS, or Linux Portduino).
+The daemon exposes the meshtastic TCP API on port `4403` rather than a USB
+serial endpoint — point the MCP server at it via `MESHTASTIC_MCP_TCP_HOST`:
+
+```bash
+# 1. Build + run a daemon on this host (see variants/native/portduino/platformio.ini
+#    for full Homebrew prereqs and CH341 LoRa-adapter setup).
+pio run -e native-macos
+~/.meshtasticd/meshtasticd
+
+# 2. Point the MCP server at it.
+export MESHTASTIC_MCP_TCP_HOST=localhost     # or host:port, default port 4403
+```
+
+**First-run gotcha — MAC address.** `meshtasticd` derives its MAC from the
+USB adapter's serial-number / product strings. Many cheap CH341 dongles
+(MeshStick included — VID 0x1A86 / PID 0x5512) ship with `iSerialNumber=0`
+and `iProduct=0`, so the daemon aborts on boot with `*** Blank MAC Address
+not allowed!`. Set the MAC explicitly in `config.yaml`:
+
+```yaml
+# Under General:
+MACAddress: 02:CA:FE:BA:BE:01
+```
+
+Use a locally-administered address (first byte's second-LSB set, e.g.
+`02:*` / `06:*` / `0A:*` / `0E:*`) to avoid colliding with a real OUI.
+
+There is also a `--hwid AA:BB:CC:DD:EE:FF` CLI flag visible in
+`meshtasticd --help`, but it is **currently broken** in
+`MAC_from_string()` (`src/platform/portduino/PortduinoGlue.cpp`): the
+function strips colons from its parameter but then reads bytes from the
+global `portduino_config.mac_address`, so `--hwid` is silently overridden
+when `MACAddress:` is also set, and crashes the daemon (uncaught
+`std::invalid_argument: stoi: no conversion`) when it isn't. Use the YAML
+form until that's fixed upstream.
+
+`list_devices` will surface the daemon as `tcp://localhost:4403` with
+`likely_meshtastic=True`, so `device_info`, `list_nodes`, `get_config`,
+`set_config`, `set_owner`, `send_text`, `userprefs_*`, and the admin RPCs
+auto-select it when no `port` is passed. Pass `port="tcp://other-host:9999"`
+explicitly to target a different daemon.
+
+**Tools that don't apply to a TCP/native node** (no USB hardware to operate
+on) raise a clear `ConnectionError` rather than failing mysteriously:
+`pio_flash`, `erase_and_flash`, `update_flash`, `touch_1200bps`,
+`serial_open` (use info/admin tools directly), and the vendor escape hatches
+`esptool_*`, `nrfutil_*`, `picotool_*`. `pio_flash` against a `native*` env
+similarly raises — there's no upload step; use `build` and run the binary
+directly.
+
+The pytest harness in `tests/` still assumes USB-attached devices per role —
+TCP-aware fixtures are not part of this surface yet.

 ## Hardware Test Suite

@@ -182,10 +268,22 @@ in the pre-flight header.
 - **`unit`** — pure Python, no hardware. boards / PIO wrapper /
  userPrefs-parse / testing-profile fixtures.
 - **`mesh`** — 2-device mesh: formation, broadcast delivery, direct+ACK,
-  traceroute, bidirectional. Parametrized over both directions.
+  traceroute, bidirectional. Parametrized over both directions. Includes
+  `test_peer_offline_recovery` which uses uhubctl to power-cycle one peer
+  mid-conversation and verifies the mesh recovers (skips without uhubctl).
 - **`telemetry`** — periodic telemetry broadcast + on-demand request/reply
  (`TELEMETRY_APP` with `wantResponse=True`).
 - **`monitor`** — boot log has no panic markers within 60 s of reboot.
+- **`recovery`** — `uhubctl` power-cycle round-trip: verifies the hub port
+  can be toggled off/on, the device re-enumerates with the same
+  `my_node_num`, and NVS-resident config (region, channel, modem preset)
+  survives a hard reset. Requires `uhubctl` on PATH; skips cleanly otherwise.
+- **`ui`** — input-broker-driven screen navigation (`AdminMessage.send_input_event`
+  injection → `Screen::handleInputEvent` → frame transition). Parametrized
+  on the screen-bearing role (heltec-v3 OLED). Captures images via USB
+  webcam + OCRs them for HTML-report evidence. Requires `pip install -e '.[ui]'`
+  and `MESHTASTIC_UI_CAMERA_DEVICE_ESP32S3=<index>`; tier is auto-deselected
+  if `cv2` isn't importable.
 - **`fleet`** — PSK-seed isolation: two labs with different seeds never
  overlap.
 - **`admin`** — owner persistence across reboot, channel URL round-trip,
@@ -193,6 +291,42 @@ in the pre-flight header.
 - **`provisioning`** — region/channel baking, userPrefs survive
  `factory_reset(full=False)`.

+#### UI tier setup
+
+The `tests/ui/` tier drives the on-device OLED via the firmware's existing
+`AdminMessage.send_input_event` RPC (no firmware changes required) and
+verifies transitions via a macro-gated log line + camera + OCR. Summary:
+
+1. Install extras: `pip install -e 'mcp-server/.[ui]'` — pulls in
+   `opencv-python-headless`, `numpy`, `easyocr`, `Pillow`. First easyocr
+   run downloads ~100 MB of models to `~/.EasyOCR/`; an autouse session
+   fixture pre-warms the reader so per-test OCR is <100 ms after that.
+2. Point a USB webcam at the heltec-v3 OLED. Discover its index:
+   ```bash
+   .venv/bin/python -c "import cv2; [print(i, cv2.VideoCapture(i).read()[0]) for i in range(5)]"
+   ```
+3. Export the per-role device env var:
+   ```bash
+   export MESHTASTIC_UI_CAMERA_DEVICE_ESP32S3=0
+   ```
+4. Run:
+   ```bash
+   ./run-tests.sh tests/ui -v
+   ```
+   Captures land under `tests/ui_captures/<session_seed>/<test_id>/`, one
+   PNG + `.ocr.txt` per `frame_capture()` call, with a per-test
+   `transcript.md` stepping through event → frame → OCR. The HTML report
+   embeds the full image strip inline (pass or fail).
+
+On macOS, `cv2.VideoCapture(0)` triggers the TCC Camera permission prompt
+on first use. Pre-grant Terminal (or your IDE's terminal) before running.
+The `OpenCVBackend` fails fast on 10 consecutive black frames so a silent
+permission denial surfaces as a clear error, not an empty PNG strip.
+
+No camera? Set `MESHTASTIC_UI_CAMERA_BACKEND=null` (or leave the device var
+unset). Tests still exercise the event-injection path and log assertions;
+captures just become 1×1 black PNGs.
+
 ### Artifacts (regenerated every run, under `tests/`)

 - `report.html` — self-contained pytest-html report. Each test gets a
@@ -217,6 +351,14 @@ Key bindings: `r` re-run focused, `f` filter, `d` failure detail, `g` open
 `report.html`, `x` export reproducer bundle, `l` cycle fw-log filter, `q`
 quit (SIGINT → SIGTERM → SIGKILL escalation).

+Set `MESHTASTIC_UI_TUI_CAMERA=1` to mount a bottom-of-screen **UI camera**
+panel. Left side: the latest capture PNG rendered as Unicode half-blocks
+(via `rich-pixels`, works in any terminal — no kitty/sixel required).
+Right side: live transcript tail ("step 3 — frame 4/8 name=nodelist_nodes
+— OCR: Nodes 2/2") so you can see every event-injection and its result
+as each UI test runs. Requires the `[ui]` extras for image rendering; the
+transcript alone works without them.
+
 ### Slash commands

 Three AI-assisted workflows are wired up for Claude Code operators
@@ -23,6 +23,21 @@ test = [
  # consumers; revisit if install cost pushes back.
  "textual>=0.50",
 ]
+# UI test tier + `capture_screen` MCP tool. Optional because the ML OCR
+# model alone is ~100 MB and camera hardware is user-supplied.
+#   pip install -e '.[ui]'     — full (OpenCV + easyocr)
+#   pip install -e '.[ui-min]' — image capture only, no OCR
+ui = [
+  "opencv-python-headless>=4.9",
+  "numpy>=1.26",
+  "easyocr>=1.7",
+  "Pillow>=10.0",
+  # Renders the latest camera capture as Unicode half-blocks in the TUI
+  # (MESHTASTIC_UI_TUI_CAMERA=1). Terminal-agnostic — no kitty / sixel
+  # dependency. Pure Python, tiny.
+  "rich-pixels>=3.0",
+]
+ui-min = ["opencv-python-headless>=4.9", "numpy>=1.26"]

 [project.scripts]
 meshtastic-mcp = "meshtastic_mcp.__main__:main"
@@ -217,6 +217,40 @@ if [[ $# -eq 0 ]]; then
 		-v --tb=short
 fi

+# UI tier requires opencv-python-headless (and ideally easyocr). If it's
+# not installed, auto-deselect tests/ui so operators without the [ui]
+# extra still get a green run. Printed in yellow; silent when cv2 is
+# present.
+_cv2_ok=0
+if "$VENV_PY" -c "import cv2" >/dev/null 2>&1; then
+	_cv2_ok=1
+fi
+_running_ui=0
+for _arg in "$@"; do
+	case "$_arg" in
+	*tests/ui* | tests/) _running_ui=1 ;;
+	*) ;;
+	esac
+done
+if [[ $_running_ui -eq 1 && $_cv2_ok -eq 0 ]]; then
+	printf '\033[33m[pre-flight] tests/ui tier detected, but opencv-python-headless is not installed — deselecting.\033[0m\n'
+	printf '             install with: .venv/bin/pip install -e "mcp-server/.[ui]"\n'
+	echo
+	set -- "$@" --ignore=tests/ui
+fi
+
+# Recovery tier needs `uhubctl` on PATH — it power-cycles devices via USB
+# hub PPPS. The tier's conftest already skips cleanly, so this is just a
+# friendly heads-up before the skip happens. `baked_single`'s auto-
+# recovery hook also benefits from having uhubctl available across the
+# whole suite.
+if ! command -v uhubctl >/dev/null 2>&1; then
+	printf "\033[33m[pre-flight] uhubctl not found on PATH — recovery tier will skip, and\n"
+	printf "             wedged-device auto-recovery is disabled.\033[0m\n"
+	printf "             install with: brew install uhubctl (macOS) or apt install uhubctl (Debian/Ubuntu).\n"
+	echo
+fi
+
 # Always emit `tests/reportlog.jsonl` (unless the operator explicitly passed
 # their own `--report-log=...`). Consumers — notably the
 # `meshtastic-mcp-test-tui` TUI — tail the reportlog for live per-test state.
@@ -0,0 +1,217 @@
+{
+  "title": "Meshtastic Firmware — Recorder Stream",
+  "description": "Live view of `.mtlog/` streams shipped by `mtlog_to_datadog.py`. Heap, packet volume, log levels, errors. One row per port.",
+  "widgets": [
+    {
+      "definition": {
+        "title": "Free heap (bytes)",
+        "type": "timeseries",
+        "show_legend": true,
+        "requests": [
+          {
+            "queries": [
+              {
+                "name": "free_heap",
+                "data_source": "metrics",
+                "query": "avg:mesh.local.heap_free_bytes{service:meshtastic-firmware} by {port}"
+              }
+            ],
+            "response_format": "timeseries",
+            "display_type": "line"
+          }
+        ],
+        "yaxis": { "label": "bytes" }
+      }
+    },
+    {
+      "definition": {
+        "title": "Heap slope (bytes/min) — last 1h",
+        "type": "query_value",
+        "precision": 0,
+        "requests": [
+          {
+            "queries": [
+              {
+                "name": "slope",
+                "data_source": "metrics",
+                "query": "derivative(avg:mesh.local.heap_free_bytes{service:meshtastic-firmware})",
+                "aggregator": "avg"
+              }
+            ],
+            "response_format": "scalar"
+          }
+        ],
+        "conditional_formats": [
+          { "comparator": "<", "value": -100, "palette": "white_on_red" },
+          { "comparator": "<", "value": 0, "palette": "white_on_yellow" },
+          { "comparator": ">=", "value": 0, "palette": "white_on_green" }
+        ]
+      }
+    },
+    {
+      "definition": {
+        "title": "Total heap (bytes)",
+        "type": "timeseries",
+        "requests": [
+          {
+            "queries": [
+              {
+                "name": "total_heap",
+                "data_source": "metrics",
+                "query": "avg:mesh.local.heap_total_bytes{service:meshtastic-firmware} by {port}"
+              }
+            ],
+            "response_format": "timeseries",
+            "display_type": "line"
+          }
+        ]
+      }
+    },
+    {
+      "definition": {
+        "title": "Battery level (%)",
+        "type": "timeseries",
+        "requests": [
+          {
+            "queries": [
+              {
+                "name": "battery",
+                "data_source": "metrics",
+                "query": "avg:mesh.device.battery_level{service:meshtastic-firmware} by {port}"
+              }
+            ],
+            "response_format": "timeseries",
+            "display_type": "line"
+          }
+        ],
+        "yaxis": { "min": "0", "max": "105" }
+      }
+    },
+    {
+      "definition": {
+        "title": "Air utilization (TX %)",
+        "type": "timeseries",
+        "requests": [
+          {
+            "queries": [
+              {
+                "name": "airutil",
+                "data_source": "metrics",
+                "query": "avg:mesh.device.air_util_tx{service:meshtastic-firmware} by {port}"
+              }
+            ],
+            "response_format": "timeseries",
+            "display_type": "line"
+          }
+        ]
+      }
+    },
+    {
+      "definition": {
+        "title": "Channel utilization (%)",
+        "type": "timeseries",
+        "requests": [
+          {
+            "queries": [
+              {
+                "name": "chutil",
+                "data_source": "metrics",
+                "query": "avg:mesh.device.channel_utilization{service:meshtastic-firmware} by {port}"
+              }
+            ],
+            "response_format": "timeseries",
+            "display_type": "line"
+          }
+        ]
+      }
+    },
+    {
+      "definition": {
+        "title": "Log volume by level",
+        "type": "timeseries",
+        "show_legend": true,
+        "requests": [
+          {
+            "response_format": "timeseries",
+            "display_type": "bars",
+            "queries": [
+              {
+                "name": "log_count",
+                "data_source": "logs",
+                "indexes": ["*"],
+                "compute": { "aggregation": "count" },
+                "search": { "query": "service:meshtastic-firmware" },
+                "group_by": [
+                  {
+                    "facet": "@level",
+                    "limit": 10,
+                    "sort": { "order": "desc", "aggregation": "count" }
+                  }
+                ]
+              }
+            ]
+          }
+        ]
+      }
+    },
+    {
+      "definition": {
+        "title": "Recent ERROR / CRIT firmware logs",
+        "type": "list_stream",
+        "requests": [
+          {
+            "response_format": "event_list",
+            "query": {
+              "data_source": "logs_stream",
+              "query_string": "service:meshtastic-firmware (status:error OR @level:ERROR OR @level:CRIT)",
+              "indexes": [],
+              "sort": { "column": "timestamp", "order": "desc" }
+            },
+            "columns": [
+              { "field": "timestamp", "width": "auto" },
+              { "field": "host", "width": "auto" },
+              { "field": "@port", "width": "auto" },
+              { "field": "@level", "width": "auto" },
+              { "field": "@thread", "width": "auto" },
+              { "field": "message", "width": "stretch" }
+            ]
+          }
+        ]
+      }
+    },
+    {
+      "definition": {
+        "title": "Recorder marker events",
+        "type": "list_stream",
+        "requests": [
+          {
+            "response_format": "event_list",
+            "query": {
+              "data_source": "logs_stream",
+              "query_string": "service:meshtastic-firmware @level:MARK",
+              "indexes": [],
+              "sort": { "column": "timestamp", "order": "desc" }
+            },
+            "columns": [
+              { "field": "timestamp", "width": "auto" },
+              { "field": "host", "width": "auto" },
+              { "field": "message", "width": "stretch" }
+            ]
+          }
+        ]
+      }
+    }
+  ],
+  "template_variables": [
+    {
+      "name": "port",
+      "prefix": "port",
+      "available_values": [],
+      "default": "*"
+    },
+    { "name": "host", "prefix": "host", "available_values": [], "default": "*" }
+  ],
+  "layout_type": "ordered",
+  "notify_list": [],
+  "reflow_type": "auto"
+}
@@ -0,0 +1,389 @@
+#!/usr/bin/env python3
+"""Forward selected recorder JSONL streams to Datadog.
+
+Reads `.mtlog/logs.jsonl` and `.mtlog/telemetry.jsonl`, ships logs to the
+Logs Intake API and telemetry numerics to the Metrics v2 series API.
+Resumes from `.mtlog/.dd-cursor.json` so a daemon restart doesn't
+duplicate rows already shipped from the current live files.
+
+This forwarder does not currently backfill rotated `.jsonl.gz` archives.
+If the recorder rotates before this process drains the live file, or the
+forwarder is down across a rotation, those older rows are skipped.
+
+Usage:
+    DD_API_KEY=... ./scripts/mtlog_to_datadog.py --tail
+    ./scripts/mtlog_to_datadog.py --once          # catch up + exit
+    ./scripts/mtlog_to_datadog.py --since 3600   # backfill last hour from start
+
+Default `DD_SITE` is `us5.datadoghq.com` — the team's Datadog instance.
+Override via `DD_SITE=...` env var or `--site` flag for one-offs.
+
+The forwarder is a separate process by design — a Datadog outage or
+auth failure must not backpressure the recorder. We exit non-zero on
+fatal config errors (missing API key) and keep retrying on transient
+network/HTTP errors.
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import os
+import socket
+import sys
+import time
+from pathlib import Path
+from typing import Any, Iterator
+
+try:
+    import requests
+except ImportError:
+    print(
+        "requests is required. Install it in the mcp-server venv: "
+        "uv pip install requests",
+        file=sys.stderr,
+    )
+    sys.exit(2)
+
+
+_DEFAULT_LOG_DIR = Path(__file__).resolve().parents[1] / ".mtlog"
+_LOG_INTAKE_TPL = "https://http-intake.logs.{site}/api/v2/logs"
+_METRICS_TPL = "https://api.{site}/api/v2/series"
+_LOG_BATCH = 50
+_METRICS_BATCH = 100
+_MAX_RETRIES = 5
+_RETRY_BASE_S = 1.5
+
+
+# --- streaming JSONL with byte-position cursor -------------------------
+
+
+class _StreamReader:
+    """Reads a single rotating JSONL with cursor-based resume.
+
+    This tails only the live `.jsonl` file. The recorder rotates files
+    (live `.jsonl` → `.YYYYMMDD-HHMMSS-uuuuuu-NNNNN.jsonl.gz`), which means
+    the live file shrinks abruptly. We detect that via inode change OR live
+    size < cursor position, and reset the live-file cursor to 0.
+    """
+
+    def __init__(self, path: Path, cursor: dict[str, Any]):
+        self.path = path
+        self.cursor = cursor
+
+    def _state(self) -> tuple[int, int]:
+        """Return (inode, size) for the live file. (0, 0) if missing."""
+        try:
+            st = self.path.stat()
+            return (st.st_ino, st.st_size)
+        except FileNotFoundError:
+            return (0, 0)
+
+    def iter_new_records(self) -> Iterator[dict[str, Any]]:
+        ino, size = self._state()
+        last_ino = self.cursor.get("ino")
+        last_pos = int(self.cursor.get("pos") or 0)
+        if ino == 0:
+            return
+        if last_ino is not None and last_ino != ino:
+            # Rotation happened. Start over.
+            last_pos = 0
+        if last_pos > size:
+            # Live file truncated/shrunk under us — recorder rotated.
+            last_pos = 0
+        try:
+            with self.path.open("r", encoding="utf-8") as fh:
+                fh.seek(last_pos)
+                for line in fh:
+                    line = line.rstrip("\n")
+                    if not line:
+                        continue
+                    try:
+                        yield json.loads(line)
+                    except json.JSONDecodeError:
+                        continue
+                    last_pos = fh.tell()
+        except FileNotFoundError:
+            return
+        self.cursor["ino"] = ino
+        self.cursor["pos"] = last_pos
+
+
+def _load_cursor(path: Path) -> dict[str, Any]:
+    if not path.exists():
+        return {}
+    try:
+        return json.loads(path.read_text())
+    except (OSError, json.JSONDecodeError):
+        return {}
+
+
+def _save_cursor(path: Path, data: dict[str, Any]) -> None:
+    tmp = path.with_suffix(".json.tmp")
+    tmp.write_text(json.dumps(data, separators=(",", ":")))
+    tmp.replace(path)
+
+
+# --- Datadog clients ---------------------------------------------------
+
+
+class _DDSession:
+    """Pool one HTTPS session, share retry logic."""
+
+    def __init__(self, api_key: str, site: str, hostname: str) -> None:
+        self.api_key = api_key
+        self.site = site
+        self.hostname = hostname
+        self.session = requests.Session()
+        self.session.headers.update(
+            {
+                "DD-API-KEY": api_key,
+                "Content-Type": "application/json",
+            }
+        )
+
+    def _post(self, url: str, payload: Any) -> bool:
+        for attempt in range(_MAX_RETRIES):
+            try:
+                resp = self.session.post(url, json=payload, timeout=30)
+            except requests.RequestException as e:
+                _wait_retry(attempt, f"network error: {e}")
+                continue
+            if 200 <= resp.status_code < 300:
+                return True
+            if resp.status_code in (408, 429, 500, 502, 503, 504):
+                _wait_retry(
+                    attempt,
+                    f"HTTP {resp.status_code} retrying",
+                )
+                continue
+            print(
+                f"datadog refused: {resp.status_code} {resp.text[:200]}",
+                file=sys.stderr,
+            )
+            return False
+        return False
+
+    def send_logs(self, records: list[dict[str, Any]]) -> int:
+        if not records:
+            return 0
+        url = _LOG_INTAKE_TPL.format(site=self.site)
+        sent = 0
+        for i in range(0, len(records), _LOG_BATCH):
+            batch = records[i : i + _LOG_BATCH]
+            if self._post(url, batch):
+                sent += len(batch)
+        return sent
+
+    def send_metrics(self, series: list[dict[str, Any]]) -> int:
+        if not series:
+            return 0
+        url = _METRICS_TPL.format(site=self.site)
+        sent = 0
+        for i in range(0, len(series), _METRICS_BATCH):
+            batch = series[i : i + _METRICS_BATCH]
+            if self._post(url, {"series": batch}):
+                sent += len(batch)
+        return sent
+
+
+def _wait_retry(attempt: int, reason: str) -> None:
+    wait = _RETRY_BASE_S * (2**attempt)
+    print(
+        f"  retry {attempt + 1}/{_MAX_RETRIES} in {wait:.1f}s ({reason})",
+        file=sys.stderr,
+    )
+    time.sleep(wait)
+
+
+# --- record → datadog payload ------------------------------------------
+
+
+def _log_record_to_dd(rec: dict[str, Any], host: str) -> dict[str, Any]:
+    line = rec.get("line") or ""
+    tags = [
+        f"role:{rec.get('role')}",
+        f"port:{rec.get('port')}",
+    ]
+    level = rec.get("level")
+    if level:
+        tags.append(f"level:{level}")
+    tag = rec.get("tag")
+    if tag:
+        tags.append(f"thread:{tag}")
+    return {
+        "ddsource": "meshtastic-firmware",
+        "service": "meshtastic-firmware",
+        "hostname": host,
+        "message": line,
+        "ddtags": ",".join(t for t in tags if t and "None" not in t),
+        "timestamp": int((rec.get("ts") or time.time()) * 1000),
+        "level": level,
+    }
+
+
+def _telemetry_record_to_metrics(
+    rec: dict[str, Any], host: str
+) -> list[dict[str, Any]]:
+    fields = rec.get("fields") or {}
+    if not isinstance(fields, dict):
+        return []
+    variant = rec.get("variant") or "unknown"
+    ts = int(rec.get("ts") or time.time())
+    out: list[dict[str, Any]] = []
+    tags = []
+    if rec.get("port"):
+        tags.append(f"port:{rec['port']}")
+    if rec.get("role"):
+        tags.append(f"role:{rec['role']}")
+    if rec.get("from_node"):
+        tags.append(f"from_node:{rec['from_node']}")
+    tags.append(f"variant:{variant}")
+    for field, value in fields.items():
+        if not isinstance(value, (int, float)) or isinstance(value, bool):
+            continue
+        metric = f"mesh.{variant}.{_metric_safe(field)}"
+        out.append(
+            {
+                "metric": metric,
+                "type": 3,  # GAUGE
+                "points": [{"timestamp": ts, "value": float(value)}],
+                "tags": tags,
+                "resources": [{"type": "host", "name": host}],
+            }
+        )
+    return out
+
+
+def _metric_safe(name: str) -> str:
+    # Lowercase, replace non-alnum with underscore for safe metric names.
+    return "".join(c.lower() if c.isalnum() else "_" for c in name)
+
+
+# --- main loop ---------------------------------------------------------
+
+
+def run(
+    log_dir: Path,
+    *,
+    once: bool,
+    since_seconds: float | None,
+    poll_interval: float,
+    dd: _DDSession,
+) -> int:
+    cursor_path = log_dir / ".dd-cursor.json"
+    cursors = _load_cursor(cursor_path)
+
+    # `--since` overrides cursor: rewind to (now-since) timestamp.
+    # We can't seek by timestamp directly (cursor is byte position), so
+    # we just reset cursors to 0 and let the time filter in iter_new
+    # drop older records.
+    cutoff_ts: float | None = None
+    if since_seconds is not None:
+        cursors = {}
+        cutoff_ts = time.time() - since_seconds
+
+    sent_total = {"logs": 0, "telemetry": 0}
+
+    while True:
+        # logs.jsonl → DD logs
+        log_cursor = cursors.setdefault("logs", {})
+        log_batch: list[dict[str, Any]] = []
+        for rec in _StreamReader(log_dir / "logs.jsonl", log_cursor).iter_new_records():
+            if cutoff_ts and (rec.get("ts") or 0) < cutoff_ts:
+                continue
+            log_batch.append(_log_record_to_dd(rec, dd.hostname))
+        if log_batch:
+            n = dd.send_logs(log_batch)
+            sent_total["logs"] += n
+            print(f"logs: sent {n}/{len(log_batch)}")
+
+        # telemetry.jsonl → DD metrics
+        telem_cursor = cursors.setdefault("telemetry", {})
+        metric_series: list[dict[str, Any]] = []
+        for rec in _StreamReader(
+            log_dir / "telemetry.jsonl", telem_cursor
+        ).iter_new_records():
+            if cutoff_ts and (rec.get("ts") or 0) < cutoff_ts:
+                continue
+            metric_series.extend(_telemetry_record_to_metrics(rec, dd.hostname))
+        if metric_series:
+            n = dd.send_metrics(metric_series)
+            sent_total["telemetry"] += n
+            print(f"telemetry: sent {n}/{len(metric_series)} metric points")
+
+        _save_cursor(cursor_path, cursors)
+
+        if once:
+            print(f"done. logs={sent_total['logs']} metrics={sent_total['telemetry']}")
+            return 0
+        time.sleep(poll_interval)
+
+
+def main(argv: list[str] | None = None) -> int:
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument(
+        "--log-dir",
+        default=str(_DEFAULT_LOG_DIR),
+        help="Path to .mtlog/ (default: mcp-server/.mtlog)",
+    )
+    mode = parser.add_mutually_exclusive_group()
+    mode.add_argument("--once", action="store_true", help="Catch up then exit")
+    mode.add_argument(
+        "--tail",
+        action="store_true",
+        help="Daemon: poll forever (default)",
+    )
+    parser.add_argument(
+        "--since",
+        type=float,
+        default=None,
+        help="Backfill last N seconds. Resets cursor.",
+    )
+    parser.add_argument(
+        "--poll-interval",
+        type=float,
+        default=5.0,
+        help="Seconds between tail polls (default 5)",
+    )
+    parser.add_argument(
+        "--site",
+        default=os.environ.get("DD_SITE", "us5.datadoghq.com"),
+        help=(
+            "Datadog site. Default is the team's instance (us5.datadoghq.com). "
+            "Override via DD_SITE env var or this flag."
+        ),
+    )
+    parser.add_argument(
+        "--host",
+        default=socket.gethostname(),
+        help="Hostname tag (default: socket.gethostname())",
+    )
+    args = parser.parse_args(argv)
+
+    api_key = os.environ.get("DD_API_KEY")
+    if not api_key:
+        print("DD_API_KEY env var required.", file=sys.stderr)
+        return 2
+
+    log_dir = Path(args.log_dir)
+    if not log_dir.exists():
+        print(
+            f"log dir {log_dir} does not exist — start the mcp-server first.",
+            file=sys.stderr,
+        )
+        return 2
+
+    dd = _DDSession(api_key=api_key, site=args.site, hostname=args.host)
+    once = args.once and not args.tail
+    return run(
+        log_dir,
+        once=once,
+        since_seconds=args.since,
+        poll_interval=args.poll_interval,
+        dd=dd,
+    )
+
+
+if __name__ == "__main__":
+    sys.exit(main())
@@ -356,6 +356,46 @@ def shutdown(
    return {"ok": True, "shutting_down_in_s": seconds}


+def send_input_event(
+    event_code: int | str,
+    kb_char: int = 0,
+    touch_x: int = 0,
+    touch_y: int = 0,
+    port: str | None = None,
+) -> dict[str, Any]:
+    """Inject an InputBroker event (button press / key / gesture) into the UI.
+
+    Wraps `AdminMessage.send_input_event` (handled in firmware at
+    src/modules/AdminModule.cpp::handleSendInputEvent). Local-only — no PKI
+    warmup needed since the admin message is addressed to `my_node_num`.
+
+    `event_code` accepts an int, a case-insensitive name
+    (`"RIGHT"` / `"input_broker_right"`), or an `InputEventCode`. The
+    firmware-side enum lives in src/input/InputBroker.h and is mirrored in
+    `meshtastic_mcp.input_events`.
+    """
+    from meshtastic.protobuf import admin_pb2  # type: ignore[import-untyped]
+
+    from .input_events import coerce_event_code
+
+    code = coerce_event_code(event_code)
+    if not 0 <= kb_char <= 255:
+        raise ValueError(f"kb_char out of u8 range: {kb_char}")
+    if not 0 <= touch_x <= 65535:
+        raise ValueError(f"touch_x out of u16 range: {touch_x}")
+    if not 0 <= touch_y <= 65535:
+        raise ValueError(f"touch_y out of u16 range: {touch_y}")
+
+    with connect(port=port) as iface:
+        msg = admin_pb2.AdminMessage()
+        msg.send_input_event.event_code = code
+        msg.send_input_event.kb_char = kb_char
+        msg.send_input_event.touch_x = touch_x
+        msg.send_input_event.touch_y = touch_y
+        iface.localNode._sendAdmin(msg)
+    return {"ok": True, "event_code": code, "kb_char": kb_char}
+
+
 def factory_reset(
    port: str | None = None, confirm: bool = False, full: bool = False
 ) -> dict[str, Any]:
@@ -0,0 +1,286 @@
+"""Cross-platform USB-webcam capture for UI tests + the `capture_screen` tool.
+
+Backends:
+- `opencv` — cv2.VideoCapture (AVFoundation on macOS, V4L2 on Linux).
+- `ffmpeg` — subprocess shelling out to the system `ffmpeg` binary. Slower
+  per frame, but zero Python deps beyond stdlib.
+- `null` — no-op stub returning a 1×1 black PNG. Used when no camera is
+  configured; keeps code paths alive without forcing every operator to
+  hook up hardware.
+
+Environment variables (read at `get_camera()` call time):
+- `MESHTASTIC_UI_CAMERA_BACKEND` — one of `opencv` / `ffmpeg` / `null` /
+  `auto` (default). `auto` picks opencv if `cv2` imports, else ffmpeg if
+  `ffmpeg --version` resolves, else null.
+- `MESHTASTIC_UI_CAMERA_DEVICE` — generic default (index or path).
+- `MESHTASTIC_UI_CAMERA_DEVICE_<ROLE>` — per-role override, e.g.
+  `MESHTASTIC_UI_CAMERA_DEVICE_ESP32S3=0` for the OLED-bearing heltec-v3.
+  Role suffix is uppercased before lookup.
+
+Dependencies land in the optional `[ui]` extra; imports are lazy so clients
+without `opencv-python-headless` installed can still import this module.
+"""
+
+from __future__ import annotations
+
+import io
+import os
+import shutil
+import subprocess
+import sys
+import time
+import warnings
+from pathlib import Path
+from typing import Protocol
+
+
+class CameraError(RuntimeError):
+    """Raised when a camera backend fails to initialize or capture."""
+
+
+class CameraBackend(Protocol):
+    name: str
+
+    def capture(self) -> bytes:
+        """Return one PNG-encoded frame."""
+        ...
+
+    def close(self) -> None: ...
+
+
+# ---------- OpenCV backend -------------------------------------------------
+
+
+class OpenCVBackend:
+    name = "opencv"
+
+    def __init__(self, device: int | str, warmup_frames: int = 5) -> None:
+        try:
+            import cv2  # type: ignore[import-untyped]  # noqa: PLC0415
+        except ImportError as exc:
+            raise CameraError(
+                "opencv backend requested but `cv2` is not installed. "
+                "Install the mcp-server [ui] extra: pip install -e '.[ui]'"
+            ) from exc
+
+        self._cv2 = cv2
+        device_arg: int | str
+        if isinstance(device, str) and device.isdigit():
+            device_arg = int(device)
+        else:
+            device_arg = device
+        self._cap = cv2.VideoCapture(device_arg)
+        if not self._cap.isOpened():
+            raise CameraError(
+                f"cv2.VideoCapture({device_arg!r}) failed to open. "
+                "On macOS check TCC Camera permission; on Linux check /dev/video* and v4l2 access."
+            )
+
+        # Drop the first few frames — auto-exposure + white-balance settle.
+        for _ in range(warmup_frames):
+            self._cap.read()
+        # Detect a stuck black-frame camera early rather than silently
+        # producing all-black captures.
+        ok, frame = self._cap.read()
+        if not ok or frame is None:
+            self._cap.release()
+            raise CameraError(f"camera {device_arg!r} opened but returned no frames")
+
+    def capture(self) -> bytes:
+        cv2 = self._cv2
+        ok, frame = self._cap.read()
+        if not ok or frame is None:
+            raise CameraError("cv2.VideoCapture.read() returned no frame")
+        success, buf = cv2.imencode(".png", frame)
+        if not success:
+            raise CameraError("cv2.imencode('.png', ...) failed")
+        return bytes(buf)
+
+    def close(self) -> None:
+        try:
+            self._cap.release()
+        except Exception:  # noqa: BLE001
+            pass
+
+
+# ---------- ffmpeg subprocess backend --------------------------------------
+
+
+class FfmpegBackend:
+    name = "ffmpeg"
+
+    def __init__(self, device: int | str) -> None:
+        if shutil.which("ffmpeg") is None:
+            raise CameraError("ffmpeg backend requested but `ffmpeg` is not on PATH")
+
+        self._device = str(device)
+        # Platform-specific -f flag:
+        #   macOS → avfoundation (index like "0")
+        #   Linux → v4l2        (device like "/dev/video0" or "0")
+        if sys.platform == "darwin":
+            self._input_format = "avfoundation"
+            self._input_spec = self._device  # bare index for avfoundation
+        else:
+            self._input_format = "v4l2"
+            self._input_spec = (
+                self._device
+                if self._device.startswith("/dev/")
+                else f"/dev/video{self._device}"
+            )
+
+    def capture(self) -> bytes:
+        cmd = [
+            "ffmpeg",
+            "-hide_banner",
+            "-loglevel",
+            "error",
+            "-f",
+            self._input_format,
+            "-i",
+            self._input_spec,
+            "-frames:v",
+            "1",
+            "-f",
+            "image2pipe",
+            "-vcodec",
+            "png",
+            "-",
+        ]
+        try:
+            out = subprocess.run(
+                cmd, capture_output=True, check=True, timeout=15  # noqa: S603
+            )
+        except subprocess.CalledProcessError as exc:
+            raise CameraError(
+                f"ffmpeg capture failed (rc={exc.returncode}): {exc.stderr.decode(errors='replace')[:200]}"
+            ) from exc
+        except subprocess.TimeoutExpired as exc:
+            raise CameraError("ffmpeg capture timed out after 15s") from exc
+        return out.stdout
+
+    def close(self) -> None:
+        pass  # stateless — each capture spawns a new process
+
+
+# ---------- Null backend ---------------------------------------------------
+
+
+# A tiny valid 1×1 transparent PNG so callers always get a decodable image.
+_BLACK_1X1_PNG = bytes.fromhex(
+    "89504e470d0a1a0a0000000d49484452000000010000000108060000001f15c489"
+    "0000000d49444154789c6300010000000500010d0a2db40000000049454e44ae426082"
+)
+
+
+class NullBackend:
+    name = "null"
+
+    def capture(self) -> bytes:
+        return _BLACK_1X1_PNG
+
+    def close(self) -> None:
+        pass
+
+
+# ---------- Factory --------------------------------------------------------
+
+
+def _resolve_device(role: str | None) -> str | None:
+    if role:
+        specific = os.environ.get(f"MESHTASTIC_UI_CAMERA_DEVICE_{role.upper()}")
+        if specific:
+            return specific
+    return os.environ.get("MESHTASTIC_UI_CAMERA_DEVICE")
+
+
+def get_camera(role: str | None = None) -> CameraBackend:
+    """Return a CameraBackend for the given device role (e.g. `"esp32s3"`).
+
+    Falls back to `NullBackend` if no camera is configured or the selected
+    backend fails to init — tests should treat captures as best-effort
+    evidence, not a blocker.
+    """
+    backend = os.environ.get("MESHTASTIC_UI_CAMERA_BACKEND", "auto").lower()
+    device = _resolve_device(role)
+
+    if backend in ("null", "none") or device is None:
+        return NullBackend()
+
+    if backend == "auto":
+        # Prefer opencv if importable; fall back to ffmpeg; else null.
+        try:
+            import cv2  # type: ignore[import-untyped]  # noqa: F401,PLC0415
+
+            backend = "opencv"
+        except ImportError:
+            backend = "ffmpeg" if shutil.which("ffmpeg") else "null"
+
+    if backend == "opencv":
+        try:
+            return OpenCVBackend(device)
+        except CameraError as exc:
+            warnings.warn(
+                f"camera backend {backend!r} failed to initialize for device "
+                f"{device!r}: {exc}; falling back to null backend",
+                RuntimeWarning,
+                stacklevel=2,
+            )
+            return NullBackend()
+    if backend == "ffmpeg":
+        try:
+            return FfmpegBackend(device)
+        except CameraError as exc:
+            warnings.warn(
+                f"camera backend {backend!r} failed to initialize for device "
+                f"{device!r}: {exc}; falling back to null backend",
+                RuntimeWarning,
+                stacklevel=2,
+            )
+            return NullBackend()
+    if backend == "null":
+        return NullBackend()
+
+    raise CameraError(f"unknown MESHTASTIC_UI_CAMERA_BACKEND: {backend!r}")
+
+
+def save_capture(png_bytes: bytes, path: Path) -> None:
+    path.parent.mkdir(parents=True, exist_ok=True)
+    path.write_bytes(png_bytes)
+
+
+def capture_to_file(role: str | None, path: Path) -> dict[str, object]:
+    """One-shot: open camera, capture, write PNG, close. Returns metadata."""
+    started = time.monotonic()
+    cam = get_camera(role)
+    try:
+        data = cam.capture()
+    finally:
+        cam.close()
+    save_capture(data, path)
+    return {
+        "backend": cam.name,
+        "path": str(path),
+        "bytes": len(data),
+        "elapsed_s": round(time.monotonic() - started, 3),
+    }
+
+
+def _is_png(data: bytes) -> bool:
+    return data.startswith(b"\x89PNG\r\n\x1a\n")
+
+
+# Exposed so callers can sanity-check a capture without a full PIL import.
+__all__ = [
+    "CameraBackend",
+    "CameraError",
+    "FfmpegBackend",
+    "NullBackend",
+    "OpenCVBackend",
+    "capture_to_file",
+    "get_camera",
+    "save_capture",
+]
+
+# Keep `io` import used (pyflakes is picky) via a small guard used at import
+# time to normalize stdin/stdout if a subclass ever needs it.
+_ = io.BytesIO  # noqa: SLF001
@@ -0,0 +1,83 @@
+"""UI-capture transcript tailer for ``meshtastic-mcp-test-tui``.
+
+Watches ``tests/ui_captures/<session_seed>/`` for new transcript lines
+(one per ``frame_capture()`` call from the UI tier) and posts them to
+the TUI. Enabled by ``MESHTASTIC_UI_TUI_CAMERA=1``.
+
+Design mirrors ``_flashlog.py``:
+- Daemon thread, cooperative stop via ``threading.Event``.
+- Tolerates the captures directory not existing yet (UI tier hasn't run).
+- Per-file seek state so we only forward genuinely-new lines.
+"""
+
+from __future__ import annotations
+
+import pathlib
+import threading
+import time
+from typing import Callable
+
+
+class UiCaptureTailer(threading.Thread):
+    """Recursively watch a captures root for new `transcript.md` lines.
+
+    Invokes ``post(test_id, line)`` for each new line, where ``test_id``
+    is derived from the path — the sanitized nodeid directory name.
+    """
+
+    def __init__(
+        self,
+        root: pathlib.Path,
+        post: Callable[[str, str], None],
+        stop: threading.Event,
+        *,
+        poll_interval: float = 0.5,
+    ) -> None:
+        super().__init__(daemon=True, name="uicap-tail")
+        self._root = root
+        self._post = post
+        self._stop = stop
+        self._poll_interval = poll_interval
+        # path → byte offset we've already read through
+        self._offsets: dict[pathlib.Path, int] = {}
+
+    def run(self) -> None:
+        while not self._stop.is_set():
+            try:
+                self._scan_once()
+            except Exception:
+                # Best-effort tailer — never bring down the TUI because a
+                # directory vanished mid-scan.
+                pass
+            time.sleep(self._poll_interval)
+
+    def _scan_once(self) -> None:
+        if not self._root.is_dir():
+            return
+        for transcript in self._root.rglob("transcript.md"):
+            test_id = transcript.parent.name
+            offset = self._offsets.get(transcript, 0)
+            try:
+                size = transcript.stat().st_size
+            except OSError:
+                continue
+            if size < offset:
+                # File truncated / rewritten — reset and re-emit.
+                offset = 0
+            if size == offset:
+                continue
+            try:
+                with transcript.open("rb") as fh:
+                    fh.seek(offset)
+                    chunk = fh.read(size - offset).decode("utf-8", errors="replace")
+            except OSError:
+                continue
+            for line in chunk.splitlines():
+                line = line.rstrip()
+                if not line or line.startswith("#"):
+                    continue
+                try:
+                    self._post(test_id, line)
+                except Exception:
+                    return
+            self._offsets[transcript] = size
@@ -518,6 +518,7 @@ def _build_app(
    from . import _fwlog as _fwlog_mod
    from . import _history as _history_mod
    from . import _reproducer as _reproducer_mod
+    from . import _uicap as _uicap_mod

    # ---------------- Messages ----------------

@@ -548,6 +549,16 @@ def _build_app(
            self.line = line
            super().__init__()

+    class UiCaptureLine(tx.Message):
+        """Live line from the UI-tier camera transcript — one per
+        `frame_capture()` call. Posted only when the camera panel is
+        enabled via `MESHTASTIC_UI_TUI_CAMERA=1`."""
+
+        def __init__(self, test_id: str, line: str) -> None:
+            self.test_id = test_id
+            self.line = line
+            super().__init__()
+
    class DeviceSnapshot(tx.Message):
        def __init__(self, rows: list[DeviceRow]) -> None:
            self.rows = rows
@@ -871,6 +882,10 @@ def _build_app(
        #pytest-pane { height: 50%; border-bottom: solid $primary-background; }
        #fwlog-header { height: 1; padding: 0 1; background: $panel; }
        #fwlog-pane { height: 1fr; }
+        #uicap-header { height: 1; padding: 0 1; background: $boost; }
+        #uicap-pane { height: 14; border-top: solid $primary-background; }
+        #uicap-image { width: 36; border-right: solid $primary-background; padding: 0 1; }
+        #uicap-log { width: 1fr; height: 14; }
        Tree { height: 100%; }
        RichLog { height: 100%; }
        #device-table { height: auto; max-height: 6; }
@@ -912,6 +927,11 @@ def _build_app(
            self._device_worker: DevicePollerWorker | None = None
            self._fwlog_worker: _fwlog_mod.FirmwareLogTailer | None = None
            self._flashlog_worker: _flashlog_mod.FlashLogTailer | None = None
+            self._uicap_worker: _uicap_mod.UiCaptureTailer | None = None
+            # Env-gated; only mounts the UI-capture panel when operator asks for it.
+            self._ui_camera_enabled = bool(
+                int(os.environ.get("MESHTASTIC_UI_TUI_CAMERA", "0") or "0")
+            )
            self._tree_filter: str = ""
            self._sigint_count = 0
            # Firmware-log port filter: None = all, else exact port match.
@@ -959,6 +979,22 @@ def _build_app(
                            wrap=True,
                            max_lines=5000,
                        )
+                    if self._ui_camera_enabled:
+                        yield tx.Static(
+                            "UI camera — latest capture + transcript   (MESHTASTIC_UI_TUI_CAMERA=1)",
+                            id="uicap-header",
+                        )
+                        with tx.Horizontal(id="uicap-pane"):
+                            yield tx.Static(
+                                "(waiting…)", id="uicap-image", markup=False
+                            )
+                            yield tx.RichLog(
+                                id="uicap-log",
+                                highlight=False,
+                                markup=False,
+                                wrap=True,
+                                max_lines=500,
+                            )
            yield tx.DataTable(id="device-table", show_cursor=False)
            yield tx.Footer()

@@ -1023,6 +1059,21 @@ def _build_app(
                stop=self._stop,
            )
            self._flashlog_worker.start()
+            # UI-capture transcript tailer — only runs when the camera panel
+            # is enabled. Watches tests/ui_captures/**/transcript.md for new
+            # lines as UI tests execute.
+            if self._ui_camera_enabled:
+                captures_root = self._root / "mcp-server" / "tests" / "ui_captures"
+                # When the TUI is launched from inside mcp-server (the usual
+                # case), `self._root` is already mcp-server/, so adjust:
+                if not captures_root.parent.name == "mcp-server":
+                    captures_root = self._root / "tests" / "ui_captures"
+                self._uicap_worker = _uicap_mod.UiCaptureTailer(
+                    root=captures_root,
+                    post=lambda tid, line: self.post_message(UiCaptureLine(tid, line)),
+                    stop=self._stop,
+                )
+                self._uicap_worker.start()
            self._spawn_pytest(self._pytest_args)
            # Header tick (seed / runtime / sparkline re-renders at 1 Hz).
            # Also refreshes the device-status column so the per-test elapsed
@@ -1217,6 +1268,84 @@ def _build_app(
            log = self.query_one("#pytest-log", tx.RichLog)
            log.write(f"[flash] {message.line}")

+        def on_ui_capture_line(self, message: Any) -> None:
+            """Route a UI-capture transcript line into the camera panel.
+
+            Each line is already formatted by frame_capture — e.g.
+            `1. **initial** — frame 2/8 name=home — OCR: ...`. We write
+            the text into the RichLog AND try to render the corresponding
+            PNG on the left side (requires rich-pixels, Pillow).
+            """
+            if not self._ui_camera_enabled:
+                return
+            try:
+                log_panel = self.query_one("#uicap-log", tx.RichLog)
+            except Exception:
+                return
+            log_panel.write(f"[{message.test_id}] {message.line}")
+            self._render_latest_ui_capture(message.test_id, message.line)
+
+        def _render_latest_ui_capture(self, test_id: str, line: str) -> None:
+            """Find the PNG that corresponds to `line` and render it on the
+            left of the uicap pane. Soft-fails if rich-pixels isn't
+            installed or the PNG isn't found — operator still has the text
+            transcript on the right.
+            """
+            try:
+                from PIL import Image  # type: ignore[import-untyped]
+                from rich_pixels import Pixels  # type: ignore[import-untyped]
+            except ImportError:
+                return
+
+            # Transcript lines look like `1. **label** — ...`. Pull the leading
+            # integer to locate the capture file.
+            import re as _re
+
+            m = _re.match(r"\s*(\d+)\.\s", line)
+            if not m:
+                return
+            step = int(m.group(1))
+
+            # Captures directory is sibling of tests/ — mirror the path the
+            # tailer watches. Search both likely layouts (in-mcp-server vs.
+            # firmware-root invocation).
+            candidates = [
+                self._root / "tests" / "ui_captures",
+                self._root / "mcp-server" / "tests" / "ui_captures",
+            ]
+            captures_root = next((p for p in candidates if p.is_dir()), None)
+            if captures_root is None:
+                return
+
+            # Drill into <session_seed>/<test_id>/ — test_id is the
+            # sanitized nodeid the tailer already passed through.
+            matches = list(captures_root.rglob(f"{test_id}/{step:03d}-*.png"))
+            if not matches:
+                return
+            png_path = matches[-1]
+
+            try:
+                img = Image.open(png_path).convert("RGB")
+                # Resize to fit ~32 cells wide × ~12 rows tall (half-block
+                # renderer gives 2× vertical resolution, so 32×24 px input
+                # lands at ~32×12 cells). Keep aspect ratio.
+                target_w = 60
+                w, h = img.size
+                target_h = max(1, int(h * (target_w / max(1, w))))
+                # Clamp: the image panel is 14 rows; half-blocks give 2 rows
+                # per vertical cell, so cap pixel height at ~26.
+                target_h = min(target_h, 26)
+                img = img.resize((target_w, target_h))
+                pixels = Pixels.from_image(img)
+            except Exception:
+                return
+
+            try:
+                image_widget = self.query_one("#uicap-image", tx.Static)
+                image_widget.update(pixels)
+            except Exception:
+                pass
+
        def on_firmware_log_line(self, message: Any) -> None:
            rec = message.record
            port = rec.get("port")
@@ -135,3 +135,14 @@ def picotool_bin() -> Path:
        ("picotool",),
        "Install via `brew install picotool` or build from https://github.com/raspberrypi/picotool.",
    )
+
+
+def uhubctl_bin() -> Path:
+    return _hw_tool(
+        "MESHTASTIC_UHUBCTL_BIN",
+        ("uhubctl",),
+        "Install via `brew install uhubctl` (macOS) or `apt install uhubctl` "
+        "(Debian/Ubuntu). On Linux without the udev rules, or on older macOS "
+        "with certain hubs, you may need to run via `sudo`: "
+        "https://github.com/mvp/uhubctl#linux-usb-permissions",
+    )
@@ -1,4 +1,4 @@
-"""Context manager for meshtastic.SerialInterface connections.
+"""Context manager for meshtastic interface connections (serial + TCP).

 Every info/admin tool goes through `connect(port)` so we have a single place
 that:
@@ -6,8 +6,16 @@ that:
  - fails fast if a serial_session is already holding the port,
  - guarantees `.close()` is called, even on exception.

-The `SerialInterface` blocks on construction waiting for the node database;
-that's fine for v1 since every tool is a short-lived request.
+Two transports:
+  - Serial: USB-attached firmware on `/dev/cu.*` / `/dev/ttyUSB*` / `COM*`.
+  - TCP: a `meshtasticd` daemon (e.g. the native macOS / Linux Portduino
+    headless build) addressed as `tcp://host[:port]` (default port 4403).
+    Surfaced by `devices.list_devices()` when `MESHTASTIC_MCP_TCP_HOST` is
+    set, so `resolve_port(None)` auto-selects it like a USB candidate.
+
+Both `SerialInterface` and `TCPInterface` block on construction waiting for
+the node database; that's fine for v1 since every tool is a short-lived
+request.
 """

 from __future__ import annotations
@@ -17,20 +25,107 @@ from typing import Iterator

 from . import devices, registry

+DEFAULT_TCP_PORT = 4403
+TCP_SCHEME = "tcp://"
+TCP_HOST_ENV = "MESHTASTIC_MCP_TCP_HOST"
+

 class ConnectionError(RuntimeError):
    pass


+def is_tcp_port(port: str | None) -> bool:
+    return bool(port) and port.startswith(TCP_SCHEME)
+
+
+def parse_tcp_port(port: str) -> tuple[str, int]:
+    """Parse `tcp://host[:port]` → (host, port). Defaults to 4403.
+
+    Validates host shape (non-empty, no path separators) and port range
+    (1..65535). Raises `ConnectionError` on malformed input — never lets
+    a raw `ValueError` bubble up to a tool surface.
+    """
+    if not port.startswith(TCP_SCHEME):
+        raise ConnectionError(
+            f"Invalid TCP endpoint {port!r}: expected '{TCP_SCHEME}host[:port]'."
+        )
+    rest = port[len(TCP_SCHEME) :]
+    if ":" in rest:
+        host, port_str = rest.rsplit(":", 1)
+        try:
+            tcp_port = int(port_str)
+        except ValueError as e:
+            raise ConnectionError(
+                f"Invalid TCP endpoint {port!r}: port {port_str!r} is not an integer."
+            ) from e
+    else:
+        host, tcp_port = rest, DEFAULT_TCP_PORT
+    if not host:
+        raise ConnectionError(f"Invalid TCP endpoint {port!r}: empty host.")
+    if any(c in host for c in ("/", "\\")):
+        raise ConnectionError(
+            f"Invalid TCP endpoint {port!r}: host {host!r} contains a path "
+            "separator. TCP hostnames cannot contain '/' or '\\' — did you "
+            "pass a serial port path or a Windows drive path by mistake?"
+        )
+    if not (1 <= tcp_port <= 65535):
+        raise ConnectionError(
+            f"Invalid TCP endpoint {port!r}: port {tcp_port} out of range "
+            "(must be 1..65535)."
+        )
+    return host, tcp_port
+
+
+def normalize_tcp_endpoint(endpoint: str) -> str:
+    r"""Normalize `host`, `host:port`, or `tcp://host[:port]` → canonical
+    `tcp://host:port` form. One place that owns the lock-key shape.
+
+    Defers all validation to `parse_tcp_port`, so path-like inputs
+    (`/dev/cu.foo`, `C:\Windows\…`), empty hosts, non-integer ports,
+    and out-of-range ports raise `ConnectionError` here too.
+    """
+    if endpoint.startswith(TCP_SCHEME):
+        canonical = endpoint
+    elif ":" in endpoint:
+        canonical = f"{TCP_SCHEME}{endpoint}"
+    else:
+        canonical = f"{TCP_SCHEME}{endpoint}:{DEFAULT_TCP_PORT}"
+    host, port = parse_tcp_port(canonical)
+    return f"{TCP_SCHEME}{host}:{port}"
+
+
+def reject_if_tcp(port: str | None, tool_name: str) -> None:
+    """Raise if `port` is a TCP endpoint — for tools that need real USB
+    hardware (flash, bootloader, vendor escape hatches, serial monitor).
+
+    Only checks the explicit arg; auto-selection via env var is the caller's
+    responsibility to handle if it matters.
+    """
+    if is_tcp_port(port):
+        raise ConnectionError(
+            f"{tool_name} is not applicable to TCP/native nodes ({port}). "
+            "This tool requires USB-attached hardware."
+        )
+
+
 def resolve_port(port: str | None) -> str:
-    """Pick a port: explicit > sole likely_meshtastic candidate > error."""
+    """Pick a port: explicit > sole likely_meshtastic candidate > error.
+
+    A `tcp://` string passes through (after canonicalization). When `port`
+    is None and no USB candidates are present, `MESHTASTIC_MCP_TCP_HOST`
+    is consulted via `devices.list_devices()`.
+    """
    if port:
+        if is_tcp_port(port):
+            return normalize_tcp_endpoint(port)
        return port
    candidates = [d for d in devices.list_devices() if d["likely_meshtastic"]]
    if not candidates:
        raise ConnectionError(
-            "No Meshtastic devices detected. Plug one in or pass `port` explicitly. "
-            "Run `list_devices` with include_unknown=True to see all serial ports."
+            "No Meshtastic devices detected. Plug one in, set "
+            f"{TCP_HOST_ENV}=<host[:port]> for a meshtasticd daemon, "
+            "or pass `port` explicitly. Run `list_devices` with "
+            "include_unknown=True to see all serial ports."
        )
    if len(candidates) > 1:
        ports = ", ".join(c["port"] for c in candidates)
@@ -43,17 +138,62 @@ def resolve_port(port: str | None) -> str:

@contextmanager
 def connect(port: str | None = None, timeout_s: float = 8.0) -> Iterator:
-    """Open a `meshtastic.SerialInterface` and always close it.
+    """Open a meshtastic interface (serial or TCP) and always close it.

-    Raises `ConnectionError` immediately if another serial session holds the
-    port (a `pio device monitor` in `serial_sessions/`, for instance).
+    For serial: raises `ConnectionError` immediately if another serial
+    session holds the port (a `pio device monitor` in `serial_sessions/`).
+    For TCP: no exclusive-access requirement, so the serial-session check
+    is skipped — but the `port_lock` still serializes parallel `connect()`
+    calls to the same daemon endpoint.
+
+    `timeout_s` is plumbed through to both `SerialInterface(timeout=...)`
+    and `TCPInterface(timeout=...)`. The meshtastic library uses the value
+    as the reply-wait deadline for `localNode.waitForConfig()` during
+    construction and for any subsequent admin RPC. `int()`-converted at
+    the boundary because the upstream API expects whole seconds.
    """
+    resolved = resolve_port(port)
+    timeout = int(timeout_s)
+
+    if is_tcp_port(resolved):
+        from meshtastic.tcp_interface import (
+            TCPInterface,  # type: ignore[import-untyped]
+        )
+
+        host, tcp_port = parse_tcp_port(resolved)
+        lock = registry.port_lock(resolved)
+        if not lock.acquire(blocking=False):
+            raise ConnectionError(
+                f"TCP endpoint {resolved} is busy — another device operation "
+                "is in flight. Retry shortly."
+            )
+
+        iface = None
+        try:
+            iface = TCPInterface(
+                hostname=host,
+                portNumber=tcp_port,
+                connectNow=True,
+                noProto=False,
+                timeout=timeout,
+            )
+            yield iface
+        finally:
+            if iface is not None:
+                try:
+                    iface.close()
+                except Exception:
+                    pass
+            try:
+                lock.release()
+            except RuntimeError:
+                pass
+        return
+
    from meshtastic.serial_interface import (
        SerialInterface,  # type: ignore[import-untyped]
    )

-    resolved = resolve_port(port)
-
    active = registry.active_session_for_port(resolved)
    if active is not None:
        raise ConnectionError(
@@ -70,7 +210,12 @@ def connect(port: str | None = None, timeout_s: float = 8.0) -> Iterator:

    iface = None
    try:
-        iface = SerialInterface(devPath=resolved, connectNow=True, noProto=False)
+        iface = SerialInterface(
+            devPath=resolved,
+            connectNow=True,
+            noProto=False,
+            timeout=timeout,
+        )
        yield iface
    finally:
        if iface is not None:
@@ -1,13 +1,18 @@
-"""USB/serial device discovery.
+"""USB/serial + TCP device discovery.

 Combines the canonical `meshtastic.util.findPorts()` allowlist/blocklist with
 the richer metadata (`serial.tools.list_ports.comports()`) so callers see
 VID/PID, descriptions, and manufacturer strings alongside the "is this likely
 a Meshtastic device" signal.
+
+If `MESHTASTIC_MCP_TCP_HOST=<host[:port]>` is set, a synthetic entry for the
+`meshtasticd` daemon at that endpoint is prepended to the result, so
+`resolve_port(None)` auto-selects it like a USB candidate.
 """

 from __future__ import annotations

+import os
 from typing import Any

 from serial.tools import list_ports
@@ -19,6 +24,45 @@ def _to_hex(value: int | None) -> str | None:
    return f"0x{value:04x}"


+def _tcp_endpoint_from_env() -> dict[str, Any] | None:
+    """Synthesize a TCP device entry from MESHTASTIC_MCP_TCP_HOST, if set.
+
+    If the env var is malformed (non-integer port, path-like host, etc.),
+    return an entry with `likely_meshtastic=False` and the parser error in
+    the description, rather than raising — `list_devices` is the diagnostic
+    tool a user reaches for when their env var isn't working, so it must
+    not crash on misconfiguration.
+    """
+    host = os.environ.get("MESHTASTIC_MCP_TCP_HOST")
+    if not host:
+        return None
+    # Lazy import to avoid a circular dependency (connection imports devices).
+    from . import connection
+
+    try:
+        port = connection.normalize_tcp_endpoint(host)
+        description = "meshtasticd (TCP)"
+        likely = True
+    except connection.ConnectionError as e:
+        # Surface the raw env-var value plus the parser's reason so the
+        # user can see exactly what they set and why it was rejected.
+        # Don't double the scheme if the user already prefixed `tcp://`.
+        port = host if host.startswith(connection.TCP_SCHEME) else f"tcp://{host}"
+        description = f"meshtasticd (TCP) — invalid MESHTASTIC_MCP_TCP_HOST: {e}"
+        likely = False
+    return {
+        "port": port,
+        "vid": None,
+        "pid": None,
+        "description": description,
+        "manufacturer": None,
+        "product": None,
+        "serial_number": None,
+        "likely_meshtastic": likely,
+        "blacklisted": False,
+    }
+
+
 def list_devices(include_unknown: bool = False) -> list[dict[str, Any]]:
    """Return enriched info for serial ports, flagging Meshtastic candidates.

@@ -70,6 +114,22 @@ def list_devices(include_unknown: bool = False) -> list[dict[str, Any]]:
            }
        )

-    # Stable ordering: likely_meshtastic first, then by port path
-    results.sort(key=lambda r: (not r["likely_meshtastic"], r["port"]))
+    # Append the TCP endpoint (if env var set) and sort everything together.
+    tcp_entry = _tcp_endpoint_from_env()
+    if tcp_entry is not None:
+        results.append(tcp_entry)
+
+    # Stable ordering: likely_meshtastic first; within rank, TCP wins over
+    # USB (explicit env-var configuration takes precedence over USB
+    # enumeration); then by port path. A misconfigured TCP entry has
+    # likely_meshtastic=False and lands among the other ignored entries —
+    # it does NOT pre-empt real USB devices at the top of the list.
+    results.sort(
+        key=lambda r: (
+            not r["likely_meshtastic"],
+            not r["port"].startswith("tcp://"),
+            r["port"],
+        )
+    )
+
    return results
@@ -0,0 +1,382 @@
+"""Fake NodeDB fixture push — Portduino file copy + hardware XModem upload.
+
+The fixture pipeline is two-stage:
+  1. `bin/gen-fake-nodedb-seed.py` produces a deterministic JSONL describing N
+     fake-but-realistic peers. Committed under `test/fixtures/nodedb/`.
+  2. `bin/seed-json-to-proto.py` compiles JSONL → binary v25 NodeDatabase
+     protobuf with fresh wall-clock timestamps.
+
+This module exposes `push_fake_nodedb(...)`, the MCP tool that:
+  - target="portduino": compiles the JSONL into the device's prefs dir on
+    the local filesystem (`~/.portduino/<config>/prefs/nodes.proto`).
+  - target="hardware":  compiles to a temp file, then streams it over the
+    XModem protocol (via the meshtastic SerialInterface/BLEInterface +
+    `meshtastic.xmodempacket` pubsub topic) to `/prefs/nodes.proto` on the
+    device. Triggers a reboot so the firmware loads the new state on next
+    boot.
+
+XModem wire details (mirrors firmware impl at src/xmodem.cpp:115-260):
+  * 128-byte chunks; final chunk padded to 128 B with 0x1A (SUB) bytes.
+  * CRC16-CCITT (poly 0x1021, init 0x0000).
+  * SOH/seq=0 carries the destination filename in `buffer.bytes`. ACK if
+    `FSCom.open(filename, FILE_O_WRITE)` succeeds; NAK otherwise.
+  * SOH/seq≥1 carries a 128-byte chunk. ACK = advance; NAK = retransmit.
+  * EOT after the last chunk flushes + closes the file on-device.
+
+Hardware push requires `confirm=True` (mirrors factory_reset / erase_and_flash
+in the .github/copilot-instructions.md "never do these without asking" list).
+"""
+
+from __future__ import annotations
+
+import dataclasses
+import hashlib
+import pathlib
+import queue
+import shutil
+import subprocess
+import sys
+import tempfile
+import time
+from typing import Any, Literal
+
+from .connection import connect, is_tcp_port
+
+# Resolve repo root so the tool works regardless of mcp-server cwd.
+_REPO_ROOT = pathlib.Path(__file__).resolve().parents[3]
+_SEED_DIR = _REPO_ROOT / "test" / "fixtures" / "nodedb"
+_COMPILE_SCRIPT = _REPO_ROOT / "bin" / "seed-json-to-proto.py"
+
+_DEFAULT_NODES_FILENAME = "/prefs/nodes.proto"
+_XMODEM_CHUNK = 128
+_XMODEM_SUB = 0x1A
+_ACK_TIMEOUT_INIT_S = 5.0
+_ACK_TIMEOUT_CHUNK_S = 2.0
+_MAX_CHUNK_RETRIES = 5
+
+_VALID_SIZES = (250, 500, 1000, 2000)
+
+
+class FixtureError(RuntimeError):
+    """Raised for any fixture-push failure (compile, transport, ack timeout, …)."""
+
+
+# ---------------------------------------------------------------------------
+# CRC16-CCITT (poly 0x1021, init 0x0000). Matches the firmware's `crc16_ccitt`.
+# Hand-rolled to avoid the optional `crcmod` dep.
+# ---------------------------------------------------------------------------
+def _crc16_ccitt(data: bytes, *, init: int = 0x0000) -> int:
+    crc = init
+    for b in data:
+        crc ^= b << 8
+        for _ in range(8):
+            if crc & 0x8000:
+                crc = ((crc << 1) ^ 0x1021) & 0xFFFF
+            else:
+                crc = (crc << 1) & 0xFFFF
+    return crc
+
+
+# ---------------------------------------------------------------------------
+# Compile step — shells out to bin/seed-json-to-proto.py so the MCP module
+# doesn't have to duplicate the proto-encoding logic.
+# ---------------------------------------------------------------------------
+def _compile_proto(jsonl_path: pathlib.Path, out_path: pathlib.Path) -> None:
+    if not _COMPILE_SCRIPT.is_file():
+        raise FixtureError(f"compile script missing at {_COMPILE_SCRIPT}")
+    cmd = [
+        sys.executable,
+        str(_COMPILE_SCRIPT),
+        "--in",
+        str(jsonl_path),
+        "--out",
+        str(out_path),
+    ]
+    try:
+        subprocess.run(cmd, check=True, capture_output=True, text=True)
+    except subprocess.CalledProcessError as exc:
+        raise FixtureError(
+            f"seed-json-to-proto.py failed (exit {exc.returncode}):\n"
+            f"  stdout: {exc.stdout}\n  stderr: {exc.stderr}"
+        ) from exc
+
+
+def _resolve_seed_jsonl(size: int, custom: str | None) -> pathlib.Path:
+    if custom is not None:
+        p = pathlib.Path(custom).expanduser().resolve()
+        if not p.is_file():
+            raise FixtureError(f"custom_seed_jsonl not found: {p}")
+        return p
+    p = _SEED_DIR / f"seed_v25_{size:04d}.jsonl"
+    if not p.is_file():
+        raise FixtureError(
+            f"missing committed seed at {p}. "
+            f"Run `./bin/regen-fake-nodedbs.sh` to generate it."
+        )
+    return p
+
+
+# ---------------------------------------------------------------------------
+# Portduino push — file copy into ~/.portduino/<config>/prefs/
+# ---------------------------------------------------------------------------
+def _portduino_prefs_dir(config_name: str) -> pathlib.Path:
+    home = pathlib.Path.home()
+    return home / ".portduino" / config_name / "prefs"
+
+
+def _push_portduino(
+    size: int,
+    jsonl: pathlib.Path,
+    portduino_config: str,
+    backup_existing: bool,
+) -> dict[str, Any]:
+    prefs = _portduino_prefs_dir(portduino_config)
+    prefs.mkdir(parents=True, exist_ok=True)
+    target = prefs / "nodes.proto"
+    backed_up_to: str | None = None
+    if backup_existing and target.is_file():
+        ts = int(time.time())
+        backup = prefs / f"nodes.proto.bak.{ts}"
+        shutil.move(str(target), str(backup))
+        backed_up_to = str(backup)
+    _compile_proto(jsonl, target)
+    raw = target.read_bytes()
+    return {
+        "transport": "portduino",
+        "path": str(target),
+        "bytes": len(raw),
+        "sha256": hashlib.sha256(raw).hexdigest(),
+        "jsonl_source": str(jsonl),
+        "backed_up_to": backed_up_to,
+    }
+
+
+# ---------------------------------------------------------------------------
+# Hardware push — XModem over BLE/serial via the meshtastic Python interface.
+# ---------------------------------------------------------------------------
+@dataclasses.dataclass
+class _AckEvent:
+    control: int
+    seq: int
+
+
+def _wait_for_response(q: "queue.Queue[_AckEvent]", timeout_s: float) -> _AckEvent:
+    try:
+        return q.get(timeout=timeout_s)
+    except queue.Empty as exc:
+        raise FixtureError(
+            f"XModem response timeout after {timeout_s:.1f}s — device not responding"
+        ) from exc
+
+
+def _push_hardware(
+    size: int,
+    jsonl: pathlib.Path,
+    port: str | None,
+    reboot_after: bool,
+) -> dict[str, Any]:
+    # Lazy imports so the module loads even when the meshtastic deps aren't
+    # available (e.g. CI in a Python env without the package installed).
+    try:
+        from meshtastic.protobuf import mesh_pb2, xmodem_pb2
+        from pubsub import pub
+    except ImportError as exc:  # pragma: no cover — dep missing
+        raise FixtureError(
+            f"hardware push requires the meshtastic + pypubsub packages: {exc}"
+        ) from exc
+
+    if is_tcp_port(port):
+        raise FixtureError(
+            "hardware push over TCP/portduino is not supported — use "
+            "target='portduino' to drop the fixture directly into the prefs dir."
+        )
+
+    # Compile the fixture to a temp file with fresh timestamps.
+    with tempfile.NamedTemporaryFile(suffix=".proto", delete=False) as tf:
+        proto_path = pathlib.Path(tf.name)
+    try:
+        _compile_proto(jsonl, proto_path)
+        payload = proto_path.read_bytes()
+    finally:
+        proto_path.unlink(missing_ok=True)
+
+    sha256 = hashlib.sha256(payload).hexdigest()
+    total_bytes = len(payload)
+
+    # Subscribe to XModem responses BEFORE we open the interface, so we don't
+    # race the first ACK that arrives during the SOH/seq=0 handshake.
+    #
+    # NB: the signature MUST declare every kwarg pypubsub will see for this
+    # topic, or pubsub locks the topic spec to a smaller set (whichever
+    # subscribe arrives first) and then *rejects* the meshtastic library's
+    # publish call with `SenderUnknownMsgDataError: unknown ... interface`.
+    # The meshtastic lib publishes both `packet=` and `interface=`
+    # (mesh_interface.py:1389-1395), so both must appear here.
+    response_q: "queue.Queue[_AckEvent]" = queue.Queue()
+
+    def _on_xmodem(packet: Any = None, interface: Any = None, **_kw: Any) -> None:
+        if packet is None:
+            return
+        response_q.put(_AckEvent(control=int(packet.control), seq=int(packet.seq)))
+
+    pub.subscribe(_on_xmodem, "meshtastic.xmodempacket")
+
+    chunks_sent = 0
+    retried = 0
+    rebooted = False
+
+    XMC = xmodem_pb2.XModem.Control
+    try:
+        with connect(port=port) as iface:
+            # 1) Send the filename (SOH, seq=0).
+            init_pkt = xmodem_pb2.XModem(
+                control=XMC.Value("SOH"),
+                seq=0,
+                buffer=_DEFAULT_NODES_FILENAME.encode("utf-8"),
+            )
+            iface._sendToRadio(mesh_pb2.ToRadio(xmodemPacket=init_pkt))
+            ack = _wait_for_response(response_q, _ACK_TIMEOUT_INIT_S)
+            if ack.control != XMC.Value("ACK"):
+                raise FixtureError(
+                    f"device refused filename {_DEFAULT_NODES_FILENAME!r} "
+                    f"(got control={ack.control}, expected ACK). "
+                    f"Filesystem full or permissions issue?"
+                )
+
+            # 2) Stream the payload in 128 B chunks.
+            for offset in range(0, total_bytes, _XMODEM_CHUNK):
+                chunk = payload[offset : offset + _XMODEM_CHUNK]
+                if len(chunk) < _XMODEM_CHUNK:
+                    # Pad final chunk to 128 B with SUB. The trailing 0x1A bytes
+                    # become part of the file on-device, but nanopb ignores
+                    # bytes past the end of the top-level message.
+                    chunk = chunk + bytes([_XMODEM_SUB] * (_XMODEM_CHUNK - len(chunk)))
+                seq = ((offset // _XMODEM_CHUNK) + 1) % 256
+                # Retry loop on NAK / timeout.
+                attempts = 0
+                while True:
+                    pkt = xmodem_pb2.XModem(
+                        control=XMC.Value("SOH"),
+                        seq=seq,
+                        buffer=chunk,
+                        crc16=_crc16_ccitt(chunk),
+                    )
+                    iface._sendToRadio(mesh_pb2.ToRadio(xmodemPacket=pkt))
+                    ack = _wait_for_response(response_q, _ACK_TIMEOUT_CHUNK_S)
+                    if ack.control == XMC.Value("ACK"):
+                        chunks_sent += 1
+                        break
+                    if ack.control == XMC.Value("NAK"):
+                        attempts += 1
+                        retried += 1
+                        if attempts >= _MAX_CHUNK_RETRIES:
+                            # Abort: send CAN so the firmware removes the half-
+                            # written file via FSCom.remove(filename).
+                            iface._sendToRadio(
+                                mesh_pb2.ToRadio(
+                                    xmodemPacket=xmodem_pb2.XModem(
+                                        control=XMC.Value("CAN")
+                                    )
+                                )
+                            )
+                            raise FixtureError(
+                                f"chunk seq={seq} NAK'd {attempts} times; "
+                                f"aborted transfer (file removed on-device)."
+                            )
+                        continue  # retry the same chunk
+                    raise FixtureError(
+                        f"unexpected XModem control={ack.control} on seq={seq}"
+                    )
+
+            # 3) Tell the device we're done.
+            iface._sendToRadio(
+                mesh_pb2.ToRadio(
+                    xmodemPacket=xmodem_pb2.XModem(control=XMC.Value("EOT"))
+                )
+            )
+            ack = _wait_for_response(response_q, _ACK_TIMEOUT_CHUNK_S)
+            if ack.control != XMC.Value("ACK"):
+                raise FixtureError(f"EOT not ACKed (got control={ack.control})")
+
+            # 4) Reboot so loadFromDisk picks up the new file.
+            if reboot_after:
+                iface.localNode.reboot(secs=1)
+                rebooted = True
+    finally:
+        try:
+            pub.unsubscribe(_on_xmodem, "meshtastic.xmodempacket")
+        except Exception:
+            pass
+
+    return {
+        "transport": "hardware",
+        "port": port,
+        "filename_on_device": _DEFAULT_NODES_FILENAME,
+        "bytes": total_bytes,
+        "chunks_sent": chunks_sent,
+        "retried": retried,
+        "sha256": sha256,
+        "jsonl_source": str(jsonl),
+        "rebooted": rebooted,
+    }
+
+
+# ---------------------------------------------------------------------------
+# Public entry point — registered as an MCP tool in server.py.
+# ---------------------------------------------------------------------------
+def push_fake_nodedb(
+    size: int,
+    target: Literal["portduino", "hardware"] = "portduino",
+    *,
+    port: str | None = None,
+    portduino_config: str = "default",
+    backup_existing: bool = True,
+    confirm: bool = False,
+    reboot_after: bool = True,
+    custom_seed_jsonl: str | None = None,
+) -> dict[str, Any]:
+    """Compile a fresh-timestamp NodeDatabase fixture and push it to a device.
+
+    Args:
+      size: 250, 500, 1000, or 2000 — selects which committed seed JSONL to use.
+      target: "portduino" (file copy to ~/.portduino/<config>/prefs/) or
+              "hardware" (XModem upload to /prefs/nodes.proto + reboot).
+      port: required for target="hardware". Serial path (e.g. /dev/cu.usbmodemXXXX)
+            or BLE identifier. TCP endpoints are rejected — use target="portduino"
+            instead.
+      portduino_config: which Portduino instance dir under ~/.portduino/. Default "default".
+      backup_existing: portduino only. Move nodes.proto -> nodes.proto.bak.<ts>
+                        if present, so you can roll back.
+      confirm: required True for target="hardware" (writes flash + reboots).
+      reboot_after: hardware only. If True, send a 1-second reboot after the
+                    final ACK so loadFromDisk picks up the new file at next boot.
+      custom_seed_jsonl: override the committed JSONL. Use to push a hand-edited
+                         test scenario.
+
+    Returns:
+        dict with transport, bytes, sha256, etc. — depends on target.
+
+    """
+    if size not in _VALID_SIZES:
+        raise FixtureError(
+            f"size must be one of {_VALID_SIZES}; got {size!r}. "
+            f"Add a new committed seed if you need a different cardinality."
+        )
+
+    jsonl = _resolve_seed_jsonl(size, custom_seed_jsonl)
+
+    if target == "portduino":
+        return _push_portduino(size, jsonl, portduino_config, backup_existing)
+
+    if target == "hardware":
+        if not confirm:
+            raise FixtureError(
+                "hardware push writes flash and triggers a reboot — pass confirm=True."
+            )
+        if not port:
+            raise FixtureError(
+                "target='hardware' requires a port (e.g. /dev/cu.usbmodemXXXX)."
+            )
+        return _push_hardware(size, jsonl, port, reboot_after)
+
+    raise FixtureError(f"unknown target {target!r}; expected 'portduino' or 'hardware'")
@@ -17,7 +17,7 @@ from typing import Any

 import serial

-from . import boards, config, devices, pio, userprefs
+from . import boards, config, connection, devices, pio, userprefs

 # Meshtastic variants use both `esp32s3` and `esp32-s3` style names across
 # variants/*/platformio.ini (no consistency enforced). Accept both spellings.
@@ -46,6 +46,18 @@ def _require_confirm(confirm: bool, operation: str) -> None:
        )


+def _reject_native_env(env: str, operation: str) -> None:
+    """`native*` envs build a host executable, not firmware — there's no
+    upload step. The user wants `build` (or just runs the binary directly).
+    """
+    if env.startswith("native"):
+        raise FlashError(
+            f"{operation} is not applicable for env {env!r}: native envs "
+            "produce a host executable, not flashable firmware. Use `build` "
+            "instead, then run the resulting binary directly."
+        )
+
+
 def _artifacts_for(env: str) -> list[Path]:
    build_dir = config.firmware_root() / ".pio" / "build" / env
    if not build_dir.is_dir():
@@ -96,18 +108,33 @@ def build(
    env: str,
    with_manifest: bool = True,
    userprefs_overrides: dict[str, Any] | None = None,
+    build_flags: dict[str, Any] | None = None,
 ) -> dict[str, Any]:
    """Run `pio run -e <env>` and return artifact paths.

    `userprefs_overrides` (optional): dict of `USERPREFS_<KEY>: value` to inject
    into userPrefs.jsonc for this build only. File is restored byte-for-byte
    on exit. Use `userprefs_set()` for persistent changes.
+
+    `build_flags` (optional): dict of `-D<NAME>=<VALUE>` macros to set for
+    this build only via `PLATFORMIO_BUILD_FLAGS`. Common useful flag:
+    `{"DEBUG_HEAP": 1}` enables per-thread leak detection + `[heap N]`
+    prefix on every log line. Combines with the recorder so heap shows
+    up at log cadence (much higher resolution than the ~60 s LocalStats
+    packet) — see `recorder/parsers.py:_HEAP_PREFIX_RE`. Bool values
+    expand to bare `-D<NAME>` (presence-only flags).
    """
    args = ["run", "-e", env]
    if with_manifest:
        args.extend(["-t", "mtjson"])
+    extra_env = _build_flags_env(build_flags) if build_flags else None
    with userprefs.temporary_overrides(userprefs_overrides) as effective:
-        result = pio.run(args, timeout=pio.TIMEOUT_BUILD, check=False)
+        result = pio.run(
+            args,
+            timeout=pio.TIMEOUT_BUILD,
+            check=False,
+            extra_env=extra_env,
+        )
    return {
        "exit_code": result.returncode,
        "artifacts": [str(p) for p in _artifacts_for(env)],
@@ -115,9 +142,27 @@ def build(
        "stderr_tail": pio.tail_lines(result.stderr, 200),
        "duration_s": round(result.duration_s, 2),
        "userprefs": _userprefs_summary(effective),
+        "build_flags": dict(build_flags) if build_flags else None,
    }


+def _build_flags_env(build_flags: dict[str, Any]) -> dict[str, str]:
+    """Translate `{"DEBUG_HEAP": 1, "FOO": "bar"}` → `{"PLATFORMIO_BUILD_FLAGS":
+    "-DDEBUG_HEAP=1 -DFOO=bar"}`. Bool True → bare `-D<NAME>`; False/None drop
+    the flag entirely. Other types stringify."""
+    parts: list[str] = []
+    for key, value in build_flags.items():
+        if value is False or value is None:
+            continue
+        if value is True:
+            parts.append(f"-D{key}")
+        else:
+            parts.append(f"-D{key}={value}")
+    if not parts:
+        return {}
+    return {"PLATFORMIO_BUILD_FLAGS": " ".join(parts)}
+
+
 def clean(env: str) -> dict[str, Any]:
    """Run `pio run -e <env> -t clean`."""
    result = pio.run(["run", "-e", env, "-t", "clean"], timeout=120, check=False)
@@ -134,18 +179,29 @@ def flash(
    port: str,
    confirm: bool = False,
    userprefs_overrides: dict[str, Any] | None = None,
+    build_flags: dict[str, Any] | None = None,
 ) -> dict[str, Any]:
    """`pio run -e <env> -t upload --upload-port <port>`. All architectures.

    `userprefs_overrides` (optional): see `build()` — the rebuild-before-upload
    that pio performs will pick up the injected values.
+
+    `build_flags` (optional): same shape as `build()` — `PLATFORMIO_BUILD_FLAGS`
+    is exported for the rebuild-before-upload, so the uploaded firmware
+    actually carries the flags. Without this propagation, `pio run -t upload`
+    would relink without the env var and silently drop them. Common use:
+    `build_flags={"DEBUG_HEAP": 1}` for the leak-hunt path.
    """
    _require_confirm(confirm, "flash")
+    _reject_native_env(env, "flash")
+    connection.reject_if_tcp(port, "flash")
+    extra_env = _build_flags_env(build_flags) if build_flags else None
    with userprefs.temporary_overrides(userprefs_overrides) as effective:
        result = pio.run(
            ["run", "-e", env, "-t", "upload", "--upload-port", port],
            timeout=pio.TIMEOUT_UPLOAD,
            check=False,
+            extra_env=extra_env,
        )
    return {
        "exit_code": result.returncode,
@@ -153,6 +209,7 @@ def flash(
        "stderr_tail": pio.tail_lines(result.stderr, 200),
        "duration_s": round(result.duration_s, 2),
        "userprefs": _userprefs_summary(effective),
+        "build_flags": dict(build_flags) if build_flags else None,
    }


@@ -200,6 +257,7 @@ def erase_and_flash(
    in that case) since a cached factory.bin would not reflect the new prefs.
    """
    _require_confirm(confirm, "erase_and_flash")
+    connection.reject_if_tcp(port, "erase_and_flash")
    _check_esp32_env(env)

    if userprefs_overrides and skip_build:
@@ -257,6 +315,7 @@ def update_flash(
    overrides are provided we always force a rebuild.
    """
    _require_confirm(confirm, "update_flash")
+    connection.reject_if_tcp(port, "update_flash")
    _check_esp32_env(env)

    if userprefs_overrides and skip_build:
@@ -391,6 +450,7 @@ def touch_1200bps(

    Returns `{ok, former_port, new_port, new_port_vid_pid, attempts}`.
    """
+    connection.reject_if_tcp(port, "touch_1200bps")
    before_list = devices.list_devices(include_unknown=True)
    before_ports = {d["port"] for d in before_list}

@@ -16,7 +16,7 @@ import subprocess
 from pathlib import Path
 from typing import Any, Sequence

-from . import config, pio
+from . import config, connection, pio

 _TIMEOUT_SHORT = 30
 _TIMEOUT_LONG = 600
@@ -102,6 +102,7 @@ def _parse_esptool_chip_info(stdout: str) -> dict[str, Any]:


 def esptool_chip_info(port: str) -> dict[str, Any]:
+    connection.reject_if_tcp(port, "esptool_chip_info")
    binary = config.esptool_bin()
    # `chip_id` prints chip + mac + crystal + features. `flash_id` adds flash.
    combined = _run(binary, ["--port", port, "flash_id"], timeout=_TIMEOUT_SHORT)
@@ -116,6 +117,7 @@ def esptool_chip_info(port: str) -> dict[str, Any]:
 def esptool_erase_flash(port: str, confirm: bool = False) -> dict[str, Any]:
    """Full-chip erase. Leaves the device unbootable until reflashed."""
    _require_confirm(confirm, "esptool_erase_flash")
+    connection.reject_if_tcp(port, "esptool_erase_flash")
    binary = config.esptool_bin()
    # esptool v5 uses `erase-flash`, older uses `erase_flash`. Try the new name
    # first; if it fails with unknown command, retry old.
@@ -134,6 +136,7 @@ def esptool_raw(
    """Raw esptool passthrough. Destructive subcommands require confirm=True."""
    if not args:
        raise ToolError("args must not be empty")
+    connection.reject_if_tcp(port, "esptool_raw")
    # Find the first non-flag arg (the subcommand).
    subcommand = next((a for a in args if not a.startswith("-")), None)
    if subcommand and subcommand.replace("-", "_") in {
@@ -156,6 +159,7 @@ NRFUTIL_DESTRUCTIVE = {"dfu", "settings"}

 def nrfutil_dfu(port: str, package_path: str, confirm: bool = False) -> dict[str, Any]:
    _require_confirm(confirm, "nrfutil_dfu")
+    connection.reject_if_tcp(port, "nrfutil_dfu")
    pkg = Path(package_path).expanduser()
    if not pkg.is_file():
        raise ToolError(f"Package not found: {pkg}")
@@ -213,6 +217,7 @@ def _parse_picotool_info(stdout: str) -> dict[str, Any]:
 def picotool_info(port: str | None = None) -> dict[str, Any]:
    """Read device info from a Pico in BOOTSEL mode. `port` is informational
    only — picotool auto-detects."""
+    connection.reject_if_tcp(port, "picotool_info")
    binary = config.picotool_bin()
    res = _run(binary, ["info", "-a"], timeout=_TIMEOUT_SHORT)
    if res["exit_code"] != 0:
@@ -0,0 +1,67 @@
+"""Python mirror of firmware `enum input_broker_event` (src/input/InputBroker.h).
+
+Used by `admin.send_input_event` + `tests/ui/` so callers can say
+`InputEventCode.RIGHT` instead of hard-coding 20. Values MUST stay in sync
+with the firmware enum — unit test `tests/unit/test_input_event_codes.py`
+pins the mapping.
+"""
+
+from __future__ import annotations
+
+from enum import IntEnum
+
+
+class InputEventCode(IntEnum):
+    """Button / key / gesture events dispatched by the firmware InputBroker."""
+
+    NONE = 0
+    SELECT = 10
+    SELECT_LONG = 11
+    UP_LONG = 12
+    DOWN_LONG = 13
+    UP = 17
+    DOWN = 18
+    LEFT = 19
+    RIGHT = 20
+    CANCEL = 24
+    BACK = 27
+    # Auto-incremented values in the C enum (27 + 1, +2, +3):
+    USER_PRESS = 28
+    ALT_PRESS = 29
+    ALT_LONG = 30
+    SHUTDOWN = 0x9B
+    GPS_TOGGLE = 0x9E
+    SEND_PING = 0xAF
+    FN_F1 = 0xF1
+    FN_F2 = 0xF2
+    FN_F3 = 0xF3
+    FN_F4 = 0xF4
+    FN_F5 = 0xF5
+    MATRIXKEY = 0xFE
+    ANYKEY = 0xFF
+
+
+def coerce_event_code(value: int | str | InputEventCode) -> int:
+    """Accept an int, a case-insensitive name, or an `InputEventCode` and return
+    the u8 wire value. Raises ValueError on unknown names / out-of-range ints.
+    """
+    if isinstance(value, InputEventCode):
+        return int(value)
+    if isinstance(value, int):
+        if not 0 <= value <= 255:
+            raise ValueError(f"event_code out of u8 range: {value}")
+        return value
+    if isinstance(value, str):
+        key = value.upper().replace("-", "_")
+        if key.startswith("INPUT_BROKER_"):
+            key = key[len("INPUT_BROKER_") :]
+        try:
+            return int(InputEventCode[key])
+        except KeyError as exc:
+            known = ", ".join(m.name for m in InputEventCode)
+            raise ValueError(
+                f"unknown event code name {value!r}; known: {known}"
+            ) from exc
+    raise TypeError(
+        f"event_code must be int|str|InputEventCode, got {type(value).__name__}"
+    )
@@ -0,0 +1,410 @@
+"""Read-side queries over the recorder's JSONL streams.
+
+Pure functions over `mcp-server/.mtlog/`. Streaming JSONL reader: never
+loads a whole file. Time-bound queries short-circuit as soon as `ts`
+exceeds the requested end. The recorder writes monotonically, so a
+forward scan is cheap; we don't need an index.
+
+All time arguments accept:
+  - epoch seconds (int/float)
+  - relative strings: "-15m", "-2h", "-3d", "now"
+  - ISO-ish absolute strings: "2026-05-07T14:30:00" (naive timestamps are
+    treated as UTC)
+
+Tools that return data ALWAYS cap their output (max_lines / max_points
+/ max), and report whether more matched than was returned.
+"""
+
+from __future__ import annotations
+
+import gzip
+import json
+import re
+import statistics
+import time
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Any, Iterator
+
+from .recorder.recorder import get_recorder
+
+_REL_RE = re.compile(r"^\s*-\s*(\d+(?:\.\d+)?)\s*([smhd])\s*$")
+_REGEX_PREVIEW_MAX = 100
+_REGEX_PREVIEW_TRUNCATE = 97
+
+
+def _parse_time(value: Any, *, now: float | None = None) -> float:
+    """Coerce to epoch seconds. Defaults `now` to `time.time()`."""
+    if value is None:
+        return time.time()
+    if isinstance(value, (int, float)):
+        return float(value)
+    if not isinstance(value, str):
+        raise ValueError(f"invalid time: {value!r}")
+    s = value.strip().lower()
+    if s in ("", "now"):
+        return time.time() if now is None else now
+    m = _REL_RE.match(s)
+    if m:
+        n = float(m.group(1))
+        unit = m.group(2)
+        secs = n * {"s": 1, "m": 60, "h": 3600, "d": 86400}[unit]
+        base = time.time() if now is None else now
+        return base - secs
+    # Try ISO 8601. Accept naive (assume UTC) and Z-suffixed.
+    try:
+        if s.endswith("z"):
+            s = s[:-1] + "+00:00"
+        dt = datetime.fromisoformat(s)
+        if dt.tzinfo is None:
+            dt = dt.replace(tzinfo=timezone.utc)
+        return dt.timestamp()
+    except ValueError as e:
+        raise ValueError(f"unparseable time: {value!r}") from e
+
+
+def _iter_jsonl(path: Path, *, since: float, until: float) -> Iterator[dict[str, Any]]:
+    """Stream records in chronological order: rotated archives first
+    (oldest → newest by lex sort, which is chronological for our
+    `YYYYMMDD-HHMMSS-uuuuuu-NNNNN` archive naming), then the live file
+    last. The "keep last N" pop-front logic in the window queries
+    relies on records arriving in time order across files.
+    """
+    files: list[Path] = []
+    # Gzipped archives are named "<stem>.YYYYMMDD-HHMMSS-uuuuuu-NNNNN.jsonl.gz".
+    for archive in sorted(path.parent.glob(f"{path.stem}.*.jsonl.gz")):
+        files.append(archive)
+    if path.exists():
+        files.append(path)
+    for f in files:
+        opener = gzip.open if f.suffix == ".gz" else open
+        try:
+            with opener(f, "rt", encoding="utf-8") as fh:  # type: ignore[arg-type]
+                for line in fh:
+                    line = line.strip()
+                    if not line:
+                        continue
+                    try:
+                        rec = json.loads(line)
+                    except json.JSONDecodeError:
+                        continue
+                    ts = rec.get("ts")
+                    if not isinstance(ts, (int, float)):
+                        continue
+                    if ts < since:
+                        continue
+                    if ts > until:
+                        # Records are append-monotonic within a file, so
+                        # the rest of this file is also past `until`.
+                        # Archives can still overlap each other, so only
+                        # short-circuit this file, not the whole scan.
+                        break
+                    yield rec
+        except (FileNotFoundError, OSError):
+            continue
+
+
+# -- queries ------------------------------------------------------------
+
+
+def logs_window(
+    start: Any = "-15m",
+    end: Any = "now",
+    *,
+    grep: str | None = None,
+    level: str | None = None,
+    tag: str | None = None,
+    port: str | None = None,
+    max_lines: int = 200,
+) -> dict[str, Any]:
+    """Recent firmware log lines, filtered.
+
+    `level` accepts a single level name or pipe-separated set
+    ("WARN|ERROR|CRIT"). `grep` is a regex (Python re) over the raw
+    `line` field. Returns the last `max_lines` matches.
+    """
+    s = _parse_time(start)
+    e = _parse_time(end)
+    levels = _split_set(level)
+    if grep:
+        try:
+            grep_re = re.compile(grep)
+        except re.error as exc:
+            preview = (
+                grep
+                if len(grep) <= _REGEX_PREVIEW_MAX
+                else f"{grep[:_REGEX_PREVIEW_TRUNCATE]}..."
+            )
+            raise ValueError(f"invalid grep regex {preview!r}: {exc}") from exc
+    else:
+        grep_re = None
+
+    base = get_recorder().base_dir
+    matched = 0
+    out: list[dict[str, Any]] = []
+    for rec in _iter_jsonl(base / "logs.jsonl", since=s, until=e):
+        if levels and rec.get("level") not in levels:
+            continue
+        if tag and rec.get("tag") != tag:
+            continue
+        if port and rec.get("port") != port:
+            continue
+        if grep_re and not grep_re.search(rec.get("line") or ""):
+            continue
+        matched += 1
+        out.append(rec)
+        if len(out) > max_lines:
+            out.pop(0)  # keep the most recent N
+    return {
+        "lines": out,
+        "total_matched": matched,
+        "dropped": max(0, matched - max_lines),
+        "window": {"start": s, "end": e},
+    }
+
+
+def telemetry_timeline(
+    window: Any = "1h",
+    *,
+    variant: str = "local",
+    field: str = "free_heap",
+    port: str | None = None,
+    max_points: int = 200,
+) -> dict[str, Any]:
+    """Timeseries of one telemetry field, downsampled.
+
+    `field` matches both the protobuf snake_case name (`free_heap`,
+    `heap_free_bytes`, `battery_level`) and camelCase (`freeHeap`).
+    Server-side bucket-mean downsamples to ≤ `max_points`. Returns
+    `slope_per_min` (linear regression slope, units/min) so a leak
+    detector can read one number.
+    """
+    end = time.time()
+    if isinstance(window, (int, float)):
+        # Numeric `window` is a duration in seconds — "last N seconds".
+        # Without this branch, `_parse_time(-N)` would treat -N as an
+        # absolute epoch timestamp (i.e., Jan 1 1970 minus N seconds),
+        # producing a wildly negative `start` and matching nothing.
+        start = end - float(window)
+    elif isinstance(window, str) and not window.startswith("-"):
+        # Bare string like "1h" is sugar for "-1h".
+        start = _parse_time(f"-{window}", now=end)
+    else:
+        start = _parse_time(window, now=end)
+
+    base = get_recorder().base_dir
+    raw: list[tuple[float, float]] = []
+    field_aliases = _field_aliases(field)
+    for rec in _iter_jsonl(base / "telemetry.jsonl", since=start, until=end):
+        if rec.get("variant") != variant:
+            continue
+        if port and rec.get("port") != port:
+            continue
+        fields = rec.get("fields") or {}
+        value: Any = None
+        for alias in field_aliases:
+            if alias in fields:
+                value = fields[alias]
+                break
+        if not isinstance(value, (int, float)):
+            continue
+        raw.append((float(rec["ts"]), float(value)))
+
+    if not raw:
+        return {
+            "points": [],
+            "samples": 0,
+            "min": None,
+            "max": None,
+            "slope_per_min": None,
+            "window": {"start": start, "end": end, "variant": variant, "field": field},
+        }
+
+    points = _downsample(raw, max_points=max_points)
+    values = [v for _, v in raw]
+    return {
+        "points": [{"ts": ts, "value": v} for ts, v in points],
+        "samples": len(raw),
+        "min": min(values),
+        "max": max(values),
+        "slope_per_min": _slope_per_min(raw),
+        "window": {"start": start, "end": end, "variant": variant, "field": field},
+    }
+
+
+def packets_window(
+    start: Any = "-5m",
+    end: Any = "now",
+    *,
+    portnum: str | None = None,
+    from_node: str | None = None,
+    to_node: str | None = None,
+    max: int = 200,
+) -> dict[str, Any]:
+    s = _parse_time(start)
+    e = _parse_time(end)
+    portnums = _split_set(portnum)
+    base = get_recorder().base_dir
+    matched = 0
+    out: list[dict[str, Any]] = []
+    for rec in _iter_jsonl(base / "packets.jsonl", since=s, until=e):
+        if portnums and rec.get("portnum") not in portnums:
+            continue
+        if from_node and str(rec.get("from_node")) != str(from_node):
+            continue
+        if to_node and str(rec.get("to_node")) != str(to_node):
+            continue
+        matched += 1
+        out.append(rec)
+        if len(out) > max:
+            out.pop(0)
+    return {
+        "packets": out,
+        "total_matched": matched,
+        "dropped": matched - max if matched > max else 0,
+        "window": {"start": s, "end": e},
+    }
+
+
+def events_window(
+    start: Any = "-1h",
+    end: Any = "now",
+    *,
+    kind: str | None = None,
+    max: int = 200,
+) -> dict[str, Any]:
+    s = _parse_time(start)
+    e = _parse_time(end)
+    kinds = _split_set(kind)
+    base = get_recorder().base_dir
+    matched = 0
+    out: list[dict[str, Any]] = []
+    for rec in _iter_jsonl(base / "events.jsonl", since=s, until=e):
+        if kinds and rec.get("kind") not in kinds:
+            continue
+        matched += 1
+        out.append(rec)
+        if len(out) > max:
+            out.pop(0)
+    return {
+        "events": out,
+        "total_matched": matched,
+        "dropped": matched - max if matched > max else 0,
+        "window": {"start": s, "end": e},
+    }
+
+
+def export(
+    start: Any,
+    end: Any,
+    dest_dir: str,
+    *,
+    streams: list[str] | None = None,
+) -> dict[str, Any]:
+    """Bundle a slice of each requested stream into `dest_dir`.
+
+    For a notebook, a bug report, or a Datadog backfill. Output files
+    are uncompressed JSONL (callers gzip themselves if they want to).
+    """
+    s = _parse_time(start)
+    e = _parse_time(end)
+    selected = streams or ["logs", "telemetry", "packets", "events"]
+    dest = Path(dest_dir)
+    dest.mkdir(parents=True, exist_ok=True)
+
+    base = get_recorder().base_dir
+    paths: dict[str, str] = {}
+    for stream in selected:
+        src = base / f"{stream}.jsonl"
+        if not src.exists() and not list(base.glob(f"{stream}.*.jsonl.gz")):
+            continue
+        out_path = dest / f"{stream}.jsonl"
+        n = 0
+        with out_path.open("w", encoding="utf-8") as fh:
+            for rec in _iter_jsonl(src, since=s, until=e):
+                fh.write(json.dumps(rec, separators=(",", ":")) + "\n")
+                n += 1
+        paths[stream] = str(out_path)
+        paths[f"{stream}_count"] = str(n)
+    return {"dest_dir": str(dest), "paths": paths, "window": {"start": s, "end": e}}
+
+
+# -- helpers ------------------------------------------------------------
+
+
+def _split_set(value: str | None) -> set[str] | None:
+    if not value:
+        return None
+    return {v.strip() for v in value.split("|") if v.strip()}
+
+
+def _field_aliases(field: str) -> list[str]:
+    """Accept snake_case OR camelCase, plus a few legacy aliases."""
+    snake = field
+    camel = _snake_to_camel(field)
+    aliases = {snake, camel}
+    # Old protobuf fields (pre-LocalStats) used different names
+    legacy = {
+        "free_heap": ["free_heap", "freeHeap", "heap_free_bytes", "heapFreeBytes"],
+        "heap_free_bytes": [
+            "heap_free_bytes",
+            "heapFreeBytes",
+            "free_heap",
+            "freeHeap",
+        ],
+        "total_heap": ["total_heap", "totalHeap", "heap_total_bytes", "heapTotalBytes"],
+        "heap_total_bytes": [
+            "heap_total_bytes",
+            "heapTotalBytes",
+            "total_heap",
+            "totalHeap",
+        ],
+    }
+    if field in legacy:
+        aliases.update(legacy[field])
+    return list(aliases)
+
+
+def _snake_to_camel(name: str) -> str:
+    parts = name.split("_")
+    return parts[0] + "".join(p.title() for p in parts[1:])
+
+
+def _downsample(
+    points: list[tuple[float, float]], *, max_points: int
+) -> list[tuple[float, float]]:
+    if len(points) <= max_points:
+        return points
+    # Even-bucket mean. Preserves shape better than nth-sample picking.
+    n = len(points)
+    bucket = n / max_points
+    out: list[tuple[float, float]] = []
+    i = 0
+    for k in range(max_points):
+        end = int((k + 1) * bucket)
+        end = min(end, n)
+        if end <= i:
+            continue
+        chunk = points[i:end]
+        ts = chunk[len(chunk) // 2][0]
+        val = statistics.fmean(v for _, v in chunk)
+        out.append((ts, val))
+        i = end
+    return out
+
+
+def _slope_per_min(points: list[tuple[float, float]]) -> float | None:
+    """Least-squares slope (units per minute). None if too few points."""
+    if len(points) < 2:
+        return None
+    xs = [t for t, _ in points]
+    ys = [v for _, v in points]
+    n = len(xs)
+    mean_x = sum(xs) / n
+    mean_y = sum(ys) / n
+    num = sum((xs[i] - mean_x) * (ys[i] - mean_y) for i in range(n))
+    den = sum((x - mean_x) ** 2 for x in xs)
+    if den == 0:
+        return None
+    slope_per_sec = num / den
+    return slope_per_sec * 60.0
@@ -0,0 +1,147 @@
+"""OCR wrapper for UI tests + the `capture_screen` tool.
+
+Auto-selects a reader in priority order:
+  1. `easyocr` (deep-learning, high quality on OLED screens — but ~100 MB
+     model download on first use).
+  2. `pytesseract` (requires system `tesseract` binary on PATH).
+  3. `null` — returns `""` with a warning. Tests fall back to log + image
+     evidence when OCR is unavailable.
+
+Override via `MESHTASTIC_UI_OCR_BACKEND=easyocr|pytesseract|null|auto`
+(default `auto`).
+
+`ocr_text(png_bytes) -> str` is the only public entry point. The reader is
+constructed lazily on first call and cached, so the easyocr cold-start cost
+only hits once per process.
+"""
+
+from __future__ import annotations
+
+import functools
+import logging
+import os
+import shutil
+import sys
+from typing import Callable
+
+log = logging.getLogger(__name__)
+
+
+def _backend_choice() -> str:
+    return os.environ.get("MESHTASTIC_UI_OCR_BACKEND", "auto").lower()
+
+
+@functools.lru_cache(maxsize=1)
+def _reader() -> tuple[str, Callable[[bytes], str]]:
+    """Return `(backend_name, callable)` for whichever OCR is available."""
+    choice = _backend_choice()
+
+    def _easyocr() -> tuple[str, Callable[[bytes], str]]:
+        import easyocr  # type: ignore[import-untyped]  # noqa: PLC0415
+        import numpy as np  # type: ignore[import-untyped]  # noqa: PLC0415
+
+        reader = easyocr.Reader(["en"], gpu=False, verbose=False)
+
+        def _run(png: bytes) -> str:
+            try:
+                import cv2  # type: ignore[import-untyped]  # noqa: PLC0415
+
+                arr = np.frombuffer(png, dtype=np.uint8)
+                img = cv2.imdecode(arr, cv2.IMREAD_COLOR)
+            except ImportError:
+                # Fall back to PIL if cv2 isn't around.
+                from io import BytesIO  # noqa: PLC0415
+
+                from PIL import Image  # type: ignore[import-untyped]  # noqa: PLC0415
+
+                img = np.array(Image.open(BytesIO(png)).convert("RGB"))
+            try:
+                results = reader.readtext(img, detail=0, paragraph=True)
+            except Exception as exc:  # noqa: BLE001
+                log.warning("easyocr failed: %s", exc)
+                return ""
+            return "\n".join(str(r) for r in results)
+
+        return "easyocr", _run
+
+    def _pytesseract() -> tuple[str, Callable[[bytes], str]]:
+        from io import BytesIO  # noqa: PLC0415
+
+        import pytesseract  # type: ignore[import-untyped]  # noqa: PLC0415
+        from PIL import Image  # type: ignore[import-untyped]  # noqa: PLC0415
+
+        if shutil.which("tesseract") is None:
+            raise ImportError("`tesseract` binary not on PATH")
+
+        def _run(png: bytes) -> str:
+            try:
+                return str(pytesseract.image_to_string(Image.open(BytesIO(png))))
+            except Exception as exc:  # noqa: BLE001
+                log.warning("pytesseract failed: %s", exc)
+                return ""
+
+        return "pytesseract", _run
+
+    def _null() -> tuple[str, Callable[[bytes], str]]:
+        log.warning(
+            "OCR backend is null; install easyocr or tesseract for text extraction"
+        )
+        return "null", lambda _png: ""
+
+    if choice == "easyocr":
+        return _easyocr()
+    if choice == "pytesseract":
+        return _pytesseract()
+    if choice == "null":
+        return _null()
+    if choice != "auto":
+        print(
+            f"[ocr] unknown MESHTASTIC_UI_OCR_BACKEND={choice!r}; falling back to auto",
+            file=sys.stderr,
+        )
+
+    # auto mode
+    try:
+        return _easyocr()
+    except ImportError:
+        pass
+    try:
+        return _pytesseract()
+    except ImportError:
+        pass
+    return _null()
+
+
+def ocr_text(png_bytes: bytes) -> str:
+    """Run OCR on a PNG-encoded image and return the decoded text (possibly empty)."""
+    if not png_bytes:
+        return ""
+    _, run = _reader()
+    return run(png_bytes)
+
+
+def backend_name() -> str:
+    """Return the currently-selected backend name, initializing if necessary."""
+    name, _ = _reader()
+    return name
+
+
+def warm() -> None:
+    """Run one dummy inference so the easyocr cold-start cost is paid upfront.
+
+    Pytest session fixture calls this once so the first real capture doesn't
+    eat the model-load latency.
+    """
+    # A 64×32 white PNG — decodes clean, no text to extract.
+    white_png = bytes.fromhex(
+        "89504e470d0a1a0a0000000d49484452000000400000002008060000007ccac28e"
+        "0000001c49444154785eedc1010d000000c2a0f74f6d0d370000000000000080"
+        "0b010000ffff030000000000000049454e44ae426082"
+    )
+    try:
+        ocr_text(white_png)
+    except Exception as exc:  # noqa: BLE001
+        log.warning("ocr.warm() failed: %s", exc)
+
+
+__all__ = ["backend_name", "ocr_text", "warm"]
@@ -92,6 +92,7 @@ def _run_capturing(
    cwd: Path | None = None,
    timeout: float | None = None,
    tee_header: str | None = None,
+    extra_env: dict[str, str] | None = None,
 ) -> tuple[int, str, str, float]:
    """Run a subprocess, capture stdout+stderr, optionally tee to the flash log.

@@ -99,6 +100,9 @@ def _run_capturing(
    `subprocess.TimeoutExpired` on timeout (callers map this to their own
    domain-specific error).

+    `extra_env` merges into the subprocess environment (parent env stays
+    intact). Used for `PLATFORMIO_BUILD_FLAGS=-DDEBUG_HEAP=1` and similar.
+
    Fast path: `subprocess.run(capture_output=True)` when no flash log is
    configured (unchanged behavior).

@@ -110,6 +114,9 @@ def _run_capturing(
    """
    log_path = _flash_log_path()
    t0 = time.monotonic()
+    env = None
+    if extra_env:
+        env = {**os.environ, **extra_env}

    if log_path is None:
        # Fast path — unchanged.
@@ -119,6 +126,7 @@ def _run_capturing(
            capture_output=True,
            text=True,
            timeout=timeout,
+            env=env,
        )
        return (
            proc.returncode,
@@ -145,6 +153,7 @@ def _run_capturing(
        stderr=subprocess.PIPE,
        text=True,
        bufsize=1,  # line-buffered
+        env=env,
    )
    stdout_chunks: list[str] = []
    stderr_chunks: list[str] = []
@@ -232,12 +241,17 @@ def run(
    cwd: Path | None = None,
    timeout: float | None = TIMEOUT_DEFAULT,
    check: bool = True,
+    extra_env: dict[str, str] | None = None,
 ) -> PioResult:
    """Invoke `pio <args>` and return captured output.

    `cwd` defaults to the firmware root. `check=True` raises `PioError` on
    non-zero exit; set `check=False` to inspect `returncode` manually.

+    `extra_env` merges into the subprocess environment — used for
+    `PLATFORMIO_BUILD_FLAGS=-DDEBUG_HEAP=1` and similar build-time
+    toggles that can't be expressed as command-line args.
+
    If `MESHTASTIC_MCP_FLASH_LOG` is set, output is also tee'd to that file
    line-by-line as it arrives (for live flash progress in the TUI).
    """
@@ -250,6 +264,7 @@ def run(
            cwd=work_dir,
            timeout=timeout,
            tee_header=f"pio {' '.join(args)}",
+            extra_env=extra_env,
        )
    except subprocess.TimeoutExpired as exc:
        raise PioTimeout(f"pio {' '.join(args)} timed out after {timeout}s") from exc
@@ -0,0 +1,19 @@
+"""Persistent device-log capture.
+
+Singleton `Recorder` subscribes once to the meshtastic pubsub fan-out
+(`meshtastic.log.line`, `meshtastic.receive.*`, `meshtastic.connection.*`)
+and appends to four JSONL files under `mcp-server/.mtlog/`. Pubsub is
+process-global so a single subscription captures every active interface
+(serial / TCP / BLE) without any per-connection bookkeeping.
+
+The recorder is opt-in-by-import: importing this package is a no-op; call
+`get_recorder().start()` (which `server.py` does at FastMCP app init) to
+begin writing. `pause()` / `resume()` exist for the rare case the user
+wants a clean stretch of file (e.g. capturing a known-good baseline).
+"""
+
+from __future__ import annotations
+
+from .recorder import Recorder, get_recorder
+
+__all__ = ["Recorder", "get_recorder"]
@@ -0,0 +1,309 @@
+"""Best-effort parsers for log lines and telemetry packets.
+
+Two flavors of log line cross our pubsub subscription:
+  1. Text-mode path (debug_log_api disabled): the meshtastic Python lib
+     accumulates bytes between protobuf frames and emits the full
+     firmware-formatted line, e.g.
+         "INFO  | 12:34:56 12345 [Main] Booting"
+     — level, HH:MM:SS, uptime seconds, thread bracket, then message.
+  2. LogRecord protobuf path (debug_log_api enabled): the lib calls
+     `_handleLogLine(record.message)` with ONLY the message body. The
+     level/source/time fields on the LogRecord are dropped before
+     pubsub fan-out. We get e.g. just "Booting".
+
+Both arrive on `meshtastic.log.line`. The parser tries to recover a
+level + thread when the prefix is present and falls back to level=None
+otherwise. Consumers who want level filtering on protobuf-mode hosts
+should grep the raw `line` field instead.
+
+Telemetry: `meshtastic.receive.telemetry` packets carry one of several
+metric variants in `packet["decoded"]["telemetry"]`. We flatten the
+chosen variant into a {field: value} dict so callers don't have to
+know the protobuf shape.
+"""
+
+from __future__ import annotations
+
+import re
+from typing import Any
+
+# Match: LEVEL | HH:MM:SS UPTIME [Thread] message
+# HH:MM:SS may be ??:??:?? when RTC isn't valid. The level alternation
+# below is the canonical list — DebugConfiguration.h's MESHTASTIC_LOG_LEVEL_*
+# macros must stay in sync with these strings.
+_LINE_RE = re.compile(
+    r"""
+    ^
+    (?P<level>DEBUG|INFO\ |WARN\ |ERROR|CRIT\ |TRACE|HEAP\ )
+    \s*\|\s*
+    (?P<clock>(?:\d{2}:\d{2}:\d{2})|(?:\?{2}:\?{2}:\?{2}))
+    \s+
+    (?P<uptime>\d+)
+    \s+
+    (?:\[(?P<thread>[^\]]+)\]\s+)?
+    (?P<msg>.*)
+    $
+    """,
+    re.VERBOSE,
+)
+
+# DEBUG_HEAP build prepends `[heap N] ` to every message body, AFTER the
+# thread bracket. See src/RedirectablePrint.cpp:175.
+_HEAP_PREFIX_RE = re.compile(r"^\[heap\s+(?P<heap>\d+)\]\s+(?P<rest>.*)$")
+
+# OSThread leak/free detection. See src/concurrency/OSThread.cpp:89-91.
+# Format: "------ Thread NAME leaked heap A -> B (delta) ------"
+#         "++++++ Thread NAME freed heap A -> B (delta) ++++++"
+_THREAD_HEAP_RE = re.compile(
+    r"""
+    ^[\-+]+\s*
+    Thread\s+(?P<thread>\S+)\s+
+    (?P<kind>leaked|freed)\s+heap\s+
+    (?P<before>-?\d+)\s*->\s*(?P<after>-?\d+)\s+
+    \((?P<delta>-?\d+)\)
+    """,
+    re.VERBOSE,
+)
+
+# Power.cpp:908 periodic heap status (DEBUG_HEAP only).
+# Format: "Heap status: FREE/TOTAL bytes free (DELTA), running R/N threads"
+_HEAP_STATUS_RE = re.compile(
+    r"""
+    Heap\s+status:\s+
+    (?P<free>\d+)\s*/\s*(?P<total>\d+)\s+bytes\s+free
+    (?:\s+\((?P<delta>-?\d+)\))?
+    """,
+    re.VERBOSE,
+)
+
+
+_ANSI_RE = re.compile(r"\x1b\[[0-9;]*[A-Za-z]")
+_HEAP_BRACKET_RE = re.compile(r"^heap\s+(?P<heap>\d+)$")
+
+
+def parse_log_line(line: str) -> dict[str, Any]:
+    """Best-effort decompose a raw firmware log line.
+
+    Returns a dict with at least `line` (the original, unmodified — ANSI
+    codes preserved for fidelity). Adds `level`, `tag`, `clock`,
+    `uptime_s`, and `msg` when the full prefix is present.
+
+    Handles two firmware quirks:
+      - LogRecord.message can carry ANSI color escapes from RedirectablePrint
+        (the BLE/StreamAPI path inherited the colored body in some builds).
+        We strip ANSI before regex matching so the prefix survives.
+      - DEBUG_HEAP injects `[heap N]` after the thread bracket. When NO
+        thread name is set, the heap takes the thread bracket position —
+        looks like `[heap 12345] msg`. We detect that shape and move it
+        out of `tag` and into `heap_free`.
+
+    DEBUG_HEAP-build extras (when `[heap N]` is injected): `heap_free`
+    (bytes), and when a `Thread X leaked|freed heap` line is recognized,
+    `heap_event` = {kind, thread, before, after, delta}.
+
+    Never raises.
+    """
+    out: dict[str, Any] = {"line": line}
+    if not line:
+        return out
+
+    # Strip ANSI escapes BEFORE any regex matching. The original `line`
+    # stays in `out["line"]` for fidelity / future grep.
+    clean = _ANSI_RE.sub("", line)
+
+    m = _LINE_RE.match(clean)
+    msg: str | None = None
+    if m:
+        level = m.group("level").rstrip()
+        out["level"] = level
+        out["clock"] = m.group("clock")
+        try:
+            out["uptime_s"] = int(m.group("uptime"))
+        except (TypeError, ValueError):
+            out["uptime_s"] = None
+        thread = m.group("thread")
+        if thread:
+            # If "thread" is actually the heap prefix taking the bracket
+            # position (DEBUG_HEAP build, no thread set), capture heap
+            # and leave tag unset.
+            hb = _HEAP_BRACKET_RE.match(thread.strip())
+            if hb:
+                try:
+                    out["heap_free"] = int(hb.group("heap"))
+                except (TypeError, ValueError):
+                    pass
+            else:
+                out["tag"] = thread
+        msg = m.group("msg")
+        out["msg"] = msg
+    else:
+        # No prefix — bare LogRecord.message body. Inspect the whole
+        # line for DEBUG_HEAP-style content; the heap-prefix and
+        # thread-leak patterns can survive on either path.
+        msg = clean
+
+    # DEBUG_HEAP per-line heap prefix: `[heap 92344] message`.
+    # Sits AFTER the thread bracket and BEFORE the message body, but
+    # for bare LogRecord lines it's at the start. Match it at the
+    # head of `msg`.
+    if msg:
+        hp = _HEAP_PREFIX_RE.match(msg)
+        if hp:
+            try:
+                out["heap_free"] = int(hp.group("heap"))
+            except (TypeError, ValueError):
+                pass
+            else:
+                # Strip the prefix from `msg` so a grep on the message
+                # body doesn't have to know about it.
+                out["msg"] = hp.group("rest")
+                msg = hp.group("rest")
+
+        # Thread-level leak/free detection.
+        thr = _THREAD_HEAP_RE.search(msg)
+        if thr:
+            try:
+                out["heap_event"] = {
+                    "kind": thr.group("kind"),
+                    "thread": thr.group("thread"),
+                    "before": int(thr.group("before")),
+                    "after": int(thr.group("after")),
+                    "delta": int(thr.group("delta")),
+                }
+            except (TypeError, ValueError):
+                pass
+
+        # Power.cpp periodic "Heap status: F/T bytes free (D), running ..."
+        hs = _HEAP_STATUS_RE.search(msg)
+        if hs:
+            try:
+                out["heap_free"] = int(hs.group("free"))
+                out["heap_total"] = int(hs.group("total"))
+                if hs.group("delta") is not None:
+                    out["heap_delta"] = int(hs.group("delta"))
+            except (TypeError, ValueError):
+                pass
+
+    return out
+
+
+# -- Telemetry ----------------------------------------------------------
+
+# Order matters: meshtastic-python decoded packets use the protobuf
+# `oneof variant` field name (snake_case) as the dict key.
+_TELEMETRY_VARIANTS = (
+    ("device_metrics", "device"),
+    ("local_stats", "local"),
+    ("environment_metrics", "environment"),
+    ("power_metrics", "power"),
+    ("air_quality_metrics", "airQuality"),
+    ("health_metrics", "health"),
+    ("host_metrics", "host"),
+)
+
+
+def extract_telemetry(packet: dict[str, Any]) -> dict[str, Any] | None:
+    """Pull the telemetry variant + flat fields out of a `meshtastic.receive.telemetry`
+    packet. Returns None when the shape isn't what we expect — so the
+    caller can fall back to a generic packets.jsonl row.
+    """
+    if not isinstance(packet, dict):
+        return None
+    decoded = packet.get("decoded")
+    if not isinstance(decoded, dict):
+        return None
+    telem = decoded.get("telemetry")
+    if not isinstance(telem, dict):
+        return None
+    # The Python lib produces dict-of-camelCase keys via MessageToDict.
+    # Try both camelCase and snake_case to be robust to lib version drift.
+    for snake, label in _TELEMETRY_VARIANTS:
+        camel = _snake_to_camel(snake)
+        for key in (snake, camel):
+            value = telem.get(key)
+            if isinstance(value, dict):
+                return {
+                    "variant": label,
+                    "fields": {k: _scalarize(v) for k, v in value.items()},
+                    "time": telem.get("time"),
+                }
+    return None
+
+
+def _snake_to_camel(name: str) -> str:
+    parts = name.split("_")
+    return parts[0] + "".join(p.title() for p in parts[1:])
+
+
+def _scalarize(value: Any) -> Any:
+    """Keep telemetry fields JSON-friendly. Lists/dicts pass through
+    untouched; bytes -> hex string; protobuf enums occasionally arrive
+    as ints (fine) or strings (also fine)."""
+    if isinstance(value, (bytes, bytearray, memoryview)):
+        return bytes(value).hex()
+    return value
+
+
+# -- Generic packet summary ---------------------------------------------
+
+
+def summarize_packet(
+    packet: dict[str, Any], *, payload_hex_len: int = 64
+) -> dict[str, Any]:
+    """Reduce a packet dict to a stable, queryable summary. Drops the
+    full payload bytes — the recorder records summaries, not pcaps.
+    """
+    if not isinstance(packet, dict):
+        return {"raw_type": type(packet).__name__}
+    decoded = packet.get("decoded") if isinstance(packet.get("decoded"), dict) else {}
+    portnum = decoded.get("portnum") if isinstance(decoded, dict) else None
+    payload = decoded.get("payload") if isinstance(decoded, dict) else None
+    payload_hex = None
+    payload_size = None
+    if isinstance(payload, (bytes, bytearray, memoryview)):
+        b = bytes(payload)
+        payload_size = len(b)
+        payload_hex = b[:payload_hex_len].hex() if b else ""
+    elif isinstance(payload, str):
+        # Some decoded payloads (text messages) come as decoded strings.
+        payload_size = len(payload)
+        payload_hex = None  # not bytes
+    return {
+        "from_node": packet.get("fromId") or packet.get("from"),
+        "to_node": packet.get("toId") or packet.get("to"),
+        "portnum": portnum,
+        "hop_limit": packet.get("hopLimit"),
+        "want_ack": packet.get("wantAck"),
+        "rx_rssi": packet.get("rxRssi"),
+        "rx_snr": packet.get("rxSnr"),
+        "channel": packet.get("channel"),
+        "id": packet.get("id"),
+        "payload_size": payload_size,
+        "payload_hex_prefix": payload_hex,
+    }
+
+
+# -- Interface identification ------------------------------------------
+
+
+def interface_label(interface: Any) -> dict[str, Any]:
+    """Stable identifier for the meshtastic interface that emitted an event.
+
+    Used as the `port`/`role` tag on every recorded row. SerialInterface
+    has `devPath`; TCPInterface has `hostname`+`portNumber`; BLEInterface
+    has `address`. Falls back to the class name when none of those exist.
+    """
+    if interface is None:
+        return {"port": None, "role": None}
+    dev_path = getattr(interface, "devPath", None)
+    if dev_path:
+        return {"port": str(dev_path), "role": "serial"}
+    hostname = getattr(interface, "hostname", None)
+    if hostname:
+        port_num = getattr(interface, "portNumber", None)
+        endpoint = f"tcp://{hostname}:{port_num}" if port_num else f"tcp://{hostname}"
+        return {"port": endpoint, "role": "tcp"}
+    address = getattr(interface, "address", None)
+    if address:
+        return {"port": str(address), "role": "ble"}
+    return {"port": type(interface).__name__, "role": None}
@@ -0,0 +1,467 @@
+"""Process-global recorder singleton.
+
+Subscribes once to the meshtastic pubsub fan-out and writes four append-only
+JSONL streams under `mcp-server/.mtlog/`. The pubsub fan-out is
+process-global — a single subscription captures every active interface
+without per-connection bookkeeping.
+
+Files:
+  logs.jsonl      — every `meshtastic.log.line` event (best-effort prefix
+                    parsed for level/tag/uptime; raw `line` always preserved)
+  telemetry.jsonl — `meshtastic.receive.telemetry` packets, flattened by
+                    variant (device / local / environment / power / etc.)
+  packets.jsonl   — every other `meshtastic.receive.*` packet, summarized
+                    (portnum, hops, RSSI/SNR, payload size + 64-byte hex)
+  events.jsonl    — connection lifecycle, node-DB updates, and manual
+                    `mark_event` rows. Lower volume; useful for aligning
+                    timelines.
+
+Pause/resume: `pause()` flips a flag; subscriptions stay registered. The
+write methods short-circuit when paused, so we don't lose ordering when
+resumed (we just have a gap). No queueing.
+"""
+
+from __future__ import annotations
+
+import logging
+import os
+import threading
+import time
+from pathlib import Path
+from typing import Any
+
+from . import parsers
+from .rotating import _RotatingJsonl
+
+_DEFAULT_DIR = Path(__file__).resolve().parents[3] / ".mtlog"
+log = logging.getLogger(__name__)
+
+
+class Recorder:
+    """Singleton write-side of the persistent log capture system."""
+
+    def __init__(self, base_dir: Path | None = None) -> None:
+        self.base_dir = Path(base_dir) if base_dir else _DEFAULT_DIR
+        self._lock = threading.RLock()
+        self._started = False
+        self._paused = False
+        self._pause_reason: str | None = None
+        self._started_at: float | None = None
+        self._handlers: list[tuple[str, Any]] = []
+        self._files: dict[str, _RotatingJsonl] = {}
+
+    # -- lifecycle ----------------------------------------------------
+
+    def start(self) -> None:
+        """Idempotent. Safe to call from FastMCP app startup."""
+        with self._lock:
+            if self._started:
+                return
+            self.base_dir.mkdir(parents=True, exist_ok=True)
+            self._files = {
+                "logs": _RotatingJsonl(self.base_dir / "logs.jsonl"),
+                "telemetry": _RotatingJsonl(self.base_dir / "telemetry.jsonl"),
+                "packets": _RotatingJsonl(self.base_dir / "packets.jsonl"),
+                "events": _RotatingJsonl(self.base_dir / "events.jsonl"),
+            }
+            self._wire_pubsub()
+            self._started = True
+            self._started_at = time.time()
+        # Write the recorder_start marker after the initialization block.
+        # `_write_event()` re-checks recorder state via `_files_snapshot()`,
+        # so keeping this out of the setup block avoids nested lifecycle work.
+        self._write_event(kind="recorder_start", label="recorder_started")
+
+    def stop(self) -> None:
+        with self._lock:
+            if not self._started:
+                return
+            self._unwire_pubsub()
+            for f in self._files.values():
+                f.close()
+            self._files = {}
+            self._started = False
+
+    def pause(self, reason: str | None = None) -> None:
+        # Write the pause marker BEFORE flipping the flag — `_write_event`
+        # short-circuits when paused, so the order matters for this event
+        # to actually land in events.jsonl.
+        self._write_event(
+            kind="recorder_pause",
+            label="paused",
+            note=reason,
+        )
+        with self._lock:
+            self._paused = True
+            self._pause_reason = reason
+
+    def resume(self) -> None:
+        # Mirror of `pause()`: clear the flag first, then write the marker
+        # so it isn't suppressed by the still-paused short-circuit.
+        with self._lock:
+            self._paused = False
+            self._pause_reason = None
+        self._write_event(kind="recorder_resume", label="resumed")
+
+    # -- pubsub wiring ------------------------------------------------
+
+    def _wire_pubsub(self) -> None:
+        from pubsub import pub  # type: ignore[import-untyped]
+
+        # Subscribers — one per topic. Each pubsub publisher sends
+        # keyword args matching its handler's signature; pubsub
+        # introspects the function signature to route args.
+        bindings = [
+            ("meshtastic.log.line", self._on_log_line),
+            ("meshtastic.serial.line", self._on_serial_line),
+            ("meshtastic.receive", self._on_receive),
+            ("meshtastic.receive.telemetry", self._on_telemetry),
+            ("meshtastic.connection.established", self._on_connection_established),
+            ("meshtastic.connection.lost", self._on_connection_lost),
+            ("meshtastic.node.updated", self._on_node_updated),
+        ]
+        for topic, handler in bindings:
+            try:
+                pub.subscribe(handler, topic)
+                self._handlers.append((topic, handler))
+            except Exception as exc:
+                # If pubsub refuses one binding (signature mismatch on
+                # an old lib version), log it and keep the rest.
+                log.warning("Recorder failed to subscribe to %s: %s", topic, exc)
+
+    def _unwire_pubsub(self) -> None:
+        from pubsub import pub  # type: ignore[import-untyped]
+
+        for topic, handler in self._handlers:
+            try:
+                pub.unsubscribe(handler, topic)
+            except Exception:
+                pass
+        self._handlers.clear()
+
+    # -- handlers -----------------------------------------------------
+    #
+    # Pubsub callbacks must never raise. Every handler is wrapped in a
+    # try/except that swallows so a bug here can't take down the
+    # SerialInterface receive thread.
+    #
+    # Threading: handlers fire on whatever thread the meshtastic library
+    # dispatches from (varies by interface), while `stop()` clears
+    # `self._files` under `self._lock`. We snapshot `_files` under the
+    # lock at the top of each handler so a concurrent stop can't
+    # KeyError us mid-write. The actual file write goes through
+    # `_RotatingJsonl` which has its own lock.
+
+    def _files_snapshot(self) -> dict[str, _RotatingJsonl] | None:
+        """Atomic-ish view of `self._files`. Returns None when the recorder
+        is paused or stopped, so handlers can early-exit cleanly without
+        racing `stop()`'s clear."""
+        with self._lock:
+            if not self._started or self._paused:
+                return None
+            return dict(self._files)
+
+    def _on_log_line(self, line: str, interface: Any = None) -> None:
+        files = self._files_snapshot()
+        if files is None:
+            return
+        try:
+            tags = parsers.interface_label(interface)
+            parsed = parsers.parse_log_line(str(line))
+            ts = time.time()
+            record: dict[str, Any] = {
+                "ts": ts,
+                "port": tags["port"],
+                "role": tags["role"],
+                "level": parsed.get("level"),
+                "tag": parsed.get("tag"),
+                "uptime_s": parsed.get("uptime_s"),
+                "line": parsed["line"],
+            }
+            # DEBUG_HEAP enrichments (only present when the firmware
+            # was built with -DDEBUG_HEAP=1). Surface as first-class
+            # fields so logs_window can grep/filter on them and so
+            # heap_free synthesizes a telemetry point below.
+            if "heap_free" in parsed:
+                record["heap_free"] = parsed["heap_free"]
+            if "heap_total" in parsed:
+                record["heap_total"] = parsed["heap_total"]
+            if "heap_delta" in parsed:
+                record["heap_delta"] = parsed["heap_delta"]
+            heap_event = parsed.get("heap_event")
+            if heap_event:
+                record["heap_event"] = heap_event
+            files["logs"].write(record)
+
+            # If the line carried a heap snapshot, also write it as a
+            # synthesized LocalStats-shaped row so telemetry_timeline
+            # picks it up at log cadence (much higher resolution than
+            # the ~60 s LocalStats packet). Tagged source=debug_heap so
+            # consumers can filter if mixing scales is unwanted.
+            heap_free = parsed.get("heap_free")
+            if isinstance(heap_free, int):
+                fields: dict[str, Any] = {"heap_free_bytes": heap_free}
+                heap_total = parsed.get("heap_total")
+                if isinstance(heap_total, int):
+                    fields["heap_total_bytes"] = heap_total
+                files["telemetry"].write(
+                    {
+                        "ts": ts,
+                        "port": tags["port"],
+                        "role": tags["role"],
+                        "from_node": None,
+                        "variant": "local",
+                        "fields": fields,
+                        "source": "debug_heap",
+                    }
+                )
+        except Exception:
+            pass
+
+    def _on_serial_line(self, line: str, port: str | None = None) -> None:
+        """Text-mode passive tap. Fired from `serial_session._drain` when a
+        `pio device monitor` subprocess is running.
+
+        Same parse + heap-synthesis path as `_on_log_line`, but receives
+        the raw text-formatted line (full level/clock/uptime/thread/`[heap N]`/
+        body). On DEBUG_HEAP builds in text mode this gives us per-log-line
+        heap data — far higher cadence than LocalStats, and works without
+        protobuf API mode (no SerialInterface required).
+        """
+        files = self._files_snapshot()
+        if files is None:
+            return
+        try:
+            parsed = parsers.parse_log_line(str(line))
+            ts = time.time()
+            record: dict[str, Any] = {
+                "ts": ts,
+                "port": port,
+                "role": "serial_session",
+                "level": parsed.get("level"),
+                "tag": parsed.get("tag"),
+                "uptime_s": parsed.get("uptime_s"),
+                "line": parsed["line"],
+            }
+            if "heap_free" in parsed:
+                record["heap_free"] = parsed["heap_free"]
+            if "heap_total" in parsed:
+                record["heap_total"] = parsed["heap_total"]
+            if "heap_delta" in parsed:
+                record["heap_delta"] = parsed["heap_delta"]
+            heap_event = parsed.get("heap_event")
+            if heap_event:
+                record["heap_event"] = heap_event
+            files["logs"].write(record)
+
+            # Synthesize a heap_free telemetry sample whenever the line
+            # carries one — same logic as _on_log_line, tagged source so
+            # consumers can distinguish text-mode tap from protobuf path.
+            heap_free = parsed.get("heap_free")
+            if isinstance(heap_free, int):
+                fields: dict[str, Any] = {"heap_free_bytes": heap_free}
+                heap_total = parsed.get("heap_total")
+                if isinstance(heap_total, int):
+                    fields["heap_total_bytes"] = heap_total
+                files["telemetry"].write(
+                    {
+                        "ts": ts,
+                        "port": port,
+                        "role": "serial_session",
+                        "from_node": None,
+                        "variant": "local",
+                        "fields": fields,
+                        "source": "debug_heap_serial",
+                    }
+                )
+        except Exception:
+            pass
+
+    def _on_telemetry(self, packet: dict[str, Any], interface: Any = None) -> None:
+        files = self._files_snapshot()
+        if files is None:
+            return
+        try:
+            tags = parsers.interface_label(interface)
+            extracted = parsers.extract_telemetry(packet)
+            if extracted is None:
+                # Couldn't extract a known variant — fall through to the
+                # generic `_on_receive` path, which will still fire for
+                # this packet via the parent topic.
+                return
+            record = {
+                "ts": time.time(),
+                "port": tags["port"],
+                "role": tags["role"],
+                "from_node": packet.get("fromId") or packet.get("from"),
+                "variant": extracted["variant"],
+                "fields": extracted["fields"],
+                "device_time": extracted.get("time"),
+            }
+            files["telemetry"].write(record)
+        except Exception:
+            pass
+
+    def _on_receive(self, packet: dict[str, Any], interface: Any = None) -> None:
+        # Generic-receive fires for EVERY packet. Telemetry packets get
+        # recorded twice (here and in _on_telemetry) — that's intentional:
+        # packets.jsonl is the universal record, telemetry.jsonl is the
+        # structured timeseries view.
+        files = self._files_snapshot()
+        if files is None:
+            return
+        try:
+            tags = parsers.interface_label(interface)
+            summary = parsers.summarize_packet(packet)
+            record = {
+                "ts": time.time(),
+                "port": tags["port"],
+                "role": tags["role"],
+                **summary,
+            }
+            files["packets"].write(record)
+        except Exception:
+            pass
+
+    def _on_connection_established(self, interface: Any = None) -> None:
+        self._write_event(
+            kind="connection_established",
+            interface=interface,
+        )
+
+    def _on_connection_lost(self, interface: Any = None) -> None:
+        self._write_event(
+            kind="connection_lost",
+            interface=interface,
+        )
+
+    def _on_node_updated(
+        self, node: dict[str, Any] | None = None, interface: Any = None
+    ) -> None:
+        # Lower-volume than packets but informative — node ID, hops away,
+        # last heard. Skip the user dict if absent.
+        try:
+            user = (node or {}).get("user") if isinstance(node, dict) else None
+            self._write_event(
+                kind="node_updated",
+                interface=interface,
+                data={
+                    "num": (node or {}).get("num"),
+                    "id": (user or {}).get("id"),
+                    "short": (user or {}).get("shortName"),
+                    "long": (user or {}).get("longName"),
+                    "hops_away": (node or {}).get("hopsAway"),
+                    "snr": (node or {}).get("snr"),
+                    "last_heard": (node or {}).get("lastHeard"),
+                },
+            )
+        except Exception:
+            pass
+
+    # -- public write helpers -----------------------------------------
+
+    def mark_event(
+        self,
+        label: str,
+        note: str | None = None,
+        data: dict[str, Any] | None = None,
+    ) -> dict[str, Any]:
+        """User-facing marker. Writes to events.jsonl AND emits a
+        synthetic logs.jsonl row tagged level=MARK so timelines align.
+        """
+        ts = self._write_event(kind="mark", label=label, note=note, data=data)
+        # Mirror into logs so a single logs_window grep finds it.
+        files = self._files_snapshot()
+        if files is not None:
+            try:
+                files["logs"].write(
+                    {
+                        "ts": ts,
+                        "port": None,
+                        "role": "marker",
+                        "level": "MARK",
+                        "tag": "mark_event",
+                        "line": f"[mark] {label}" + (f" — {note}" if note else ""),
+                    }
+                )
+            except Exception:
+                pass
+        return {"ts": ts, "label": label}
+
+    def _write_event(
+        self,
+        *,
+        kind: str,
+        label: str | None = None,
+        note: str | None = None,
+        interface: Any = None,
+        data: dict[str, Any] | None = None,
+    ) -> float:
+        ts = time.time()
+        # Lifecycle markers (recorder_start, recorder_pause, recorder_resume)
+        # arrive at choreographed moments — `pause()` writes BEFORE flipping
+        # the flag and `resume()` writes AFTER clearing it, so those calls
+        # see _paused=False here. Other event kinds short-circuit when
+        # paused via the snapshot guard below.
+        files = self._files_snapshot()
+        if files is None:
+            return ts
+        try:
+            tags = parsers.interface_label(interface)
+            files["events"].write(
+                {
+                    "ts": ts,
+                    "kind": kind,
+                    "label": label,
+                    "note": note,
+                    "port": tags["port"],
+                    "role": tags["role"],
+                    "data": data,
+                }
+            )
+        except Exception:
+            pass
+        return ts
+
+    # -- introspection ------------------------------------------------
+
+    def status(self) -> dict[str, Any]:
+        with self._lock:
+            return {
+                "running": self._started,
+                "paused": self._paused,
+                "pause_reason": self._pause_reason,
+                "started_at": self._started_at,
+                "base_dir": str(self.base_dir),
+                "files": {name: f.status() for name, f in self._files.items()},
+            }
+
+    def force_rotate_all(self) -> dict[str, Any]:
+        """Test/admin hook: rotate every stream right now."""
+        with self._lock:
+            files = list(self._files.values())
+        for f in files:
+            f.force_rotate()
+        # `status()` re-acquires `self._lock`; release before calling it.
+        return self.status()
+
+
+# -- module-level singleton accessor ------------------------------------
+
+_INSTANCE_LOCK = threading.Lock()
+_INSTANCE: Recorder | None = None
+
+
+def get_recorder() -> Recorder:
+    """Return the process-global Recorder. Created on first call.
+
+    Honors `MESHTASTIC_MCP_LOG_DIR` env var for the base directory
+    (used by tests to redirect to a tmpdir).
+    """
+    global _INSTANCE
+    with _INSTANCE_LOCK:
+        if _INSTANCE is None:
+            override = os.environ.get("MESHTASTIC_MCP_LOG_DIR")
+            base = Path(override) if override else None
+            _INSTANCE = Recorder(base_dir=base)
+        return _INSTANCE
@@ -0,0 +1,163 @@
+"""Append-only JSONL writer with size-capped rotation.
+
+A `_RotatingJsonl` owns one live `.jsonl` file. Writes are line-delimited
+JSON objects (one row per call). When the live file exceeds `max_bytes`,
+it is closed, gzipped to `<name>.YYYYMMDD-HHMMSS-uuuuuu-NNNNN.jsonl.gz`,
+and the live file resets to empty. Old archives past `keep_archives` are
+unlinked oldest-first.
+
+Size check is amortized — `os.fstat` runs every `check_every` writes,
+not per-write, so the hot path stays at one `fh.write` + one `fh.flush`.
+
+Threading: every public method acquires `self._lock`. The recorder runs
+several pubsub handlers on whatever thread the meshtastic library
+dispatches from (varies by interface), and queries from MCP tool calls
+arrive on the FastMCP request thread, so this lock is not optional.
+"""
+
+from __future__ import annotations
+
+import gzip
+import json
+import os
+import shutil
+import threading
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Any
+
+
+class _RotatingJsonl:
+    """Append-only JSONL with size rotation. Thread-safe."""
+
+    def __init__(
+        self,
+        path: Path,
+        *,
+        max_bytes: int = 100 * 1024 * 1024,
+        keep_archives: int = 5,
+        check_every: int = 1000,
+    ) -> None:
+        self.path = path
+        self.max_bytes = max_bytes
+        self.keep_archives = keep_archives
+        self.check_every = check_every
+        self._lock = threading.Lock()
+        self._fh: Any = None
+        self._writes_since_check = 0
+        self._rotations = 0
+        self._lines_written = 0
+        self._last_ts: float | None = None
+        self._open()
+
+    # -- lifecycle ----------------------------------------------------
+
+    def _open(self) -> None:
+        self.path.parent.mkdir(parents=True, exist_ok=True)
+        self._fh = self.path.open("a", encoding="utf-8")
+
+    def close(self) -> None:
+        with self._lock:
+            if self._fh is not None:
+                try:
+                    self._fh.close()
+                finally:
+                    self._fh = None
+
+    # -- write --------------------------------------------------------
+
+    def write(self, record: dict[str, Any]) -> None:
+        """Append one JSON object as a line. Triggers rotation if oversized."""
+        line = json.dumps(record, separators=(",", ":"), default=str) + "\n"
+        with self._lock:
+            if self._fh is None:
+                return
+            try:
+                self._fh.write(line)
+                self._fh.flush()
+            except Exception:
+                # Best-effort: a failed write must not crash the pubsub
+                # handler. Caller has no way to react anyway.
+                return
+            self._lines_written += 1
+            ts = record.get("ts")
+            if isinstance(ts, (int, float)):
+                self._last_ts = float(ts)
+            self._writes_since_check += 1
+            if self._writes_since_check >= self.check_every:
+                self._writes_since_check = 0
+                self._maybe_rotate()
+
+    # -- rotation -----------------------------------------------------
+
+    def _maybe_rotate(self) -> None:
+        # Caller holds self._lock.
+        try:
+            size = os.fstat(self._fh.fileno()).st_size
+        except OSError:
+            return
+        if size < self.max_bytes:
+            return
+        self._rotate_locked()
+
+    def _rotate_locked(self) -> None:
+        # Close, gzip-rename, reopen empty, prune oldest archives.
+        try:
+            self._fh.close()
+        except Exception:
+            pass
+        self._fh = None
+        # Microsecond-resolution timestamp + per-instance counter so back-
+        # to-back rotations (small max_bytes, repeated `force_rotate()`,
+        # or chatty test loops) get unique archive filenames. The lex
+        # sort order of `YYYYMMDD-HHMMSS-uuuuuu-NNNNN` is chronological,
+        # which `_prune_archives()` and `log_query._iter_jsonl()` both
+        # rely on.
+        stamp = datetime.now(timezone.utc).strftime("%Y%m%d-%H%M%S-%f")
+        archive = self.path.with_suffix(f".{stamp}-{self._rotations:05d}.jsonl.gz")
+        try:
+            with self.path.open("rb") as src, gzip.open(archive, "wb") as dst:
+                shutil.copyfileobj(src, dst, length=1024 * 1024)
+            self.path.unlink()
+        except Exception:
+            # Rotation is best-effort. If gzip fails, leave the file
+            # in place and re-open it; we'll try again next check.
+            pass
+        self._open()
+        self._rotations += 1
+        self._prune_archives()
+
+    def _prune_archives(self) -> None:
+        # Match siblings of self.path.name with `.jsonl.gz` suffix.
+        prefix = self.path.stem  # "logs" for "logs.jsonl"
+        # Archive filenames are already lexicographically chronological.
+        # Prune by name, not mtime, so copied/restored files don't reorder.
+        archives = sorted(self.path.parent.glob(f"{prefix}.*.jsonl.gz"))
+        excess = len(archives) - self.keep_archives
+        for old in archives[: max(0, excess)]:
+            try:
+                old.unlink()
+            except OSError:
+                pass
+
+    def force_rotate(self) -> None:
+        """Test/admin hook: rotate immediately regardless of size."""
+        with self._lock:
+            if self._fh is not None:
+                self._rotate_locked()
+
+    # -- introspection ------------------------------------------------
+
+    def status(self) -> dict[str, Any]:
+        with self._lock:
+            try:
+                size = os.fstat(self._fh.fileno()).st_size if self._fh else 0
+            except OSError:
+                size = 0
+            return {
+                "path": str(self.path),
+                "size": size,
+                "lines": self._lines_written,
+                "last_ts": self._last_ts,
+                "rotations": self._rotations,
+            }
@@ -46,7 +46,23 @@ class SerialSession:


 def _drain(session: SerialSession) -> None:
-    """Reader thread: line-by-line pull stdout into buffer."""
+    """Reader thread: line-by-line pull stdout into buffer.
+
+    Each line is also published to the `meshtastic.serial.line` pubsub
+    topic so the persistent recorder can capture it without holding its
+    own port. This is the text-mode tap path: when no SerialInterface is
+    open, the firmware emits full formatted lines (level + clock + uptime
+    + thread + `[heap N]` prefix on DEBUG_HEAP builds + body), and we
+    fan them out to whoever is listening. Pubsub is best-effort —
+    publish failures must never block the reader.
+    """
+    # Lazy import: pubsub isn't required just to import this module
+    # (e.g., during static analysis), and we want a clean test surface.
+    try:
+        from pubsub import pub  # type: ignore[import-untyped]
+    except Exception:  # pragma: no cover - defensive
+        pub = None
+
    assert session.proc.stdout is not None
    try:
        for line in session.proc.stdout:
@@ -54,6 +70,16 @@ def _drain(session: SerialSession) -> None:
            with session.lock:
                session.buffer.append(line_stripped)
                session.total_lines += 1
+            if pub is not None:
+                try:
+                    pub.sendMessage(
+                        "meshtastic.serial.line",
+                        line=line_stripped,
+                        port=session.port,
+                    )
+                except Exception:
+                    # A subscriber raising must not break the reader.
+                    pass
    except Exception:  # pragma: no cover - defensive
        pass
    finally:
@@ -71,6 +97,10 @@ def open_session(
    If `env` is supplied, pio resolves baud and filters from platformio.ini.
    Otherwise uses the supplied `baud` and `filters` (default `['direct']`).
    """
+    # Lazy import to avoid circular: registry imports serial_session.
+    from . import connection
+
+    connection.reject_if_tcp(port, "serial_open")
    args = ["device", "monitor", "--port", port, "--no-reconnect"]
    effective_filters: list[str]
    effective_baud: int = baud
@@ -1,4 +1,4 @@
-"""FastMCP server wiring — 38 tools across 7 categories.
+"""FastMCP server wiring — 43 tools across 9 categories (adds uhubctl power control).

 Each tool handler is a thin delegation to a named module (pio.py, admin.py,
 etc.). Business logic does not live here.
@@ -6,6 +6,7 @@ etc.). Business logic does not live here.

 from __future__ import annotations

+import logging
 from typing import Any

 from mcp.server.fastmcp import FastMCP
@@ -14,17 +15,38 @@ from . import (
    admin,
    boards,
    devices,
+    fixtures,
    flash,
    hw_tools,
    info,
+    log_query,
    registry,
    serial_session,
 )
 from . import userprefs as userprefs_mod
+from .recorder import get_recorder
+
+log = logging.getLogger(__name__)

 app = FastMCP("meshtastic-mcp")


+def _start_recorder() -> None:
+    # Persistent device-log capture. Starts on first import — pubsub fan-out
+    # is process-global, so subscribing here captures every active interface
+    # (whether opened by an MCP tool, a pytest fixture, or a serial_session).
+    # Files land in mcp-server/.mtlog/ (gitignored). See recorder/recorder.py
+    # for the full design. Recorder startup is best-effort: an unwritable
+    # log dir or pubsub mismatch should not take the MCP server down.
+    try:
+        get_recorder().start()
+    except Exception as exc:
+        log.warning("Failed to start persistent recorder: %s", exc)
+
+
+_start_recorder()
+
+
 # ---------- Discovery & metadata ------------------------------------------


@@ -75,6 +97,7 @@ def build(
    env: str,
    with_manifest: bool = True,
    userprefs: dict[str, Any] | None = None,
+    build_flags: dict[str, Any] | None = None,
 ) -> dict[str, Any]:
    """Build firmware for one env via `pio run -e <env>`.

@@ -86,8 +109,21 @@ def build(
    build via userPrefs.jsonc injection. The file is restored after the build
    completes. Use `userprefs_manifest` to discover available keys. Use
    `userprefs_set` for persistent changes.
+
+    `build_flags` (optional): dict of `-D<NAME>=<VALUE>` macros for this build
+    only, injected via `PLATFORMIO_BUILD_FLAGS`. Common pattern:
+    `build_flags={"DEBUG_HEAP": 1}` enables per-thread leak detection + a
+    `[heap N]` prefix on every log line. The recorder picks the prefix up
+    automatically and synthesizes a high-resolution heap timeline that
+    `telemetry_timeline(field="free_heap")` can read alongside the normal
+    ~60 s LocalStats packets. Pair with `/leakhunt` for classification.
    """
-    return flash.build(env, with_manifest=with_manifest, userprefs_overrides=userprefs)
+    return flash.build(
+        env,
+        with_manifest=with_manifest,
+        userprefs_overrides=userprefs,
+        build_flags=build_flags,
+    )


@app.tool()
@@ -105,6 +141,7 @@ def pio_flash(
    port: str,
    confirm: bool = False,
    userprefs: dict[str, Any] | None = None,
+    build_flags: dict[str, Any] | None = None,
 ) -> dict[str, Any]:
    """Flash firmware via `pio run -e <env> -t upload --upload-port <port>`.

@@ -114,8 +151,19 @@ def pio_flash(

    `userprefs` (optional): dict of `USERPREFS_<KEY>: value` baked into this
    build via userPrefs.jsonc injection; restored after upload.
+
+    `build_flags` (optional): dict of `-D<NAME>=<VALUE>` macros for the
+    rebuild-before-upload, e.g. `{"DEBUG_HEAP": 1}`. Required for the flags
+    to actually land in the uploaded firmware — without it, the implicit
+    rebuild relinks without the env var and silently drops them.
    """
-    return flash.flash(env, port, confirm=confirm, userprefs_overrides=userprefs)
+    return flash.flash(
+        env,
+        port,
+        confirm=confirm,
+        userprefs_overrides=userprefs,
+        build_flags=build_flags,
+    )


@app.tool()
@@ -513,6 +561,152 @@ def factory_reset(
    return admin.factory_reset(port=port, confirm=confirm, full=full)


+@app.tool()
+def send_input_event(
+    event_code: int | str,
+    kb_char: int = 0,
+    touch_x: int = 0,
+    touch_y: int = 0,
+    port: str | None = None,
+) -> dict[str, Any]:
+    """Inject an InputBroker event (button / key / gesture) into the device UI.
+
+    Drives the same code path as a physical button press. Accepts a numeric
+    event code (0..255) or a name like `"RIGHT"`, `"SELECT"`, `"FN_F1"`.
+
+    Common codes: SELECT=10, UP=17, DOWN=18, LEFT=19, RIGHT=20, CANCEL=24,
+    BACK=27, FN_F1..F5=241..245.
+    """
+    return admin.send_input_event(
+        event_code=event_code,
+        kb_char=kb_char,
+        touch_x=touch_x,
+        touch_y=touch_y,
+        port=port,
+    )
+
+
+@app.tool()
+def capture_screen(role: str | None = None, ocr: bool = True) -> dict[str, Any]:
+    """Grab a frame from the USB webcam pointed at the device screen.
+
+    Returns PNG bytes (base64), optional OCR text, and backend metadata.
+    Requires the `[ui]` extras (opencv-python-headless) and a camera
+    configured via `MESHTASTIC_UI_CAMERA_DEVICE[_<ROLE>]`. Falls back to a
+    1×1 black PNG from the null backend when no camera is configured.
+    """
+    import base64
+
+    from . import camera as camera_mod
+
+    cam = camera_mod.get_camera(role)
+    try:
+        png = cam.capture()
+    finally:
+        cam.close()
+
+    result: dict[str, Any] = {
+        "backend": cam.name,
+        "bytes": len(png),
+        "image_base64": base64.b64encode(png).decode("ascii"),
+    }
+    if ocr:
+        from . import ocr as ocr_mod
+
+        result["ocr_backend"] = ocr_mod.backend_name()
+        result["ocr_text"] = ocr_mod.ocr_text(png)
+    return result
+
+
+# ---------- USB power control (uhubctl) -----------------------------------
+
+
+@app.tool()
+def uhubctl_list() -> list[dict[str, Any]]:
+    """List every USB hub + per-port device attachment as seen by `uhubctl`.
+
+    Read-only — no confirm required. Each hub entry includes its location
+    (`1-1.3`), descriptor, whether it supports Per-Port Power Switching,
+    and a list of populated ports with VID:PID of attached devices.
+    Useful for pre-flight checks before a destructive power-cycle call.
+    """
+    from . import uhubctl as uhubctl_mod
+
+    return uhubctl_mod.list_hubs()
+
+
+@app.tool()
+def uhubctl_power(
+    action: str,
+    location: str | None = None,
+    port: int | None = None,
+    role: str | None = None,
+    confirm: bool = False,
+) -> dict[str, Any]:
+    """Power a USB hub port on or off via `uhubctl -a on|off`.
+
+    Target the port by either (`location`, `port`) — raw uhubctl syntax,
+    e.g. `location="1-1.3", port=2` — OR by `role` ("nrf52", "esp32s3").
+    Role lookup honors `MESHTASTIC_UHUBCTL_LOCATION_<ROLE>` +
+    `_PORT_<ROLE>` env vars first, falls back to VID auto-detection.
+
+    `action="off"` requires `confirm=True` (destructive — the attached
+    device will immediately disappear from the OS).
+    """
+    from . import uhubctl as uhubctl_mod
+
+    action_lower = action.lower()
+    if action_lower not in {"on", "off"}:
+        raise ValueError(f"action must be 'on' or 'off', got {action!r}")
+    if action_lower == "off" and not confirm:
+        raise uhubctl_mod.UhubctlError(
+            "uhubctl_power action='off' requires confirm=True"
+        )
+    loc, p = _resolve_uhubctl_target(location, port, role)
+    if action_lower == "on":
+        return uhubctl_mod.power_on(loc, p)
+    return uhubctl_mod.power_off(loc, p)
+
+
+@app.tool()
+def uhubctl_cycle(
+    location: str | None = None,
+    port: int | None = None,
+    role: str | None = None,
+    delay_s: int = 2,
+    confirm: bool = False,
+) -> dict[str, Any]:
+    """Power a USB hub port off, wait `delay_s` seconds, then on.
+
+    The typical hard-reset sequence — shorter than off+on as two RPCs
+    because uhubctl handles the timing in-process. Target by (location,
+    port) or by role (see `uhubctl_power`). Requires `confirm=True`.
+    """
+    from . import uhubctl as uhubctl_mod
+
+    if not confirm:
+        raise uhubctl_mod.UhubctlError("uhubctl_cycle requires confirm=True")
+    if delay_s < 0 or delay_s > 60:
+        raise ValueError(f"delay_s must be 0..60, got {delay_s}")
+    loc, p = _resolve_uhubctl_target(location, port, role)
+    return uhubctl_mod.cycle(loc, p, delay_s=delay_s)
+
+
+def _resolve_uhubctl_target(
+    location: str | None, port: int | None, role: str | None
+) -> tuple[str, int]:
+    """Shared arg-resolution for uhubctl_power + uhubctl_cycle."""
+    from . import uhubctl as uhubctl_mod
+
+    if role is not None:
+        if location is not None or port is not None:
+            raise ValueError("pass either `role` OR (`location` + `port`), not both")
+        return uhubctl_mod.resolve_target(role)
+    if location is None or port is None:
+        raise ValueError("must pass `role` or both `location` and `port`")
+    return (location, int(port))
+
+
 # ---------- Direct hardware tools -----------------------------------------


@@ -588,3 +782,227 @@ def picotool_load(uf2_path: str, confirm: bool = False) -> dict[str, Any]:
 def picotool_raw(args: list[str], confirm: bool = False) -> dict[str, Any]:
    """Pass-through to `picotool`. load/reboot/save/erase require confirm=True."""
    return hw_tools.picotool_raw(args, confirm=confirm)
+
+
+# ---------- Persistent device-log capture (recorder) ----------------------
+#
+# The recorder is autouse — it starts at server import and continuously
+# writes every meshtastic pubsub event to JSONL files under .mtlog/. These
+# tools are query-only over those files, plus a few lifecycle controls.
+
+
+@app.tool()
+def logs_window(
+    start: str = "-15m",
+    end: str = "now",
+    grep: str | None = None,
+    level: str | None = None,
+    tag: str | None = None,
+    port: str | None = None,
+    max_lines: int = 200,
+) -> dict[str, Any]:
+    """Recent firmware log lines from the persistent recorder.
+
+    Filters by time window, regex over the line, level (single or
+    pipe-separated set like "WARN|ERROR|CRIT"), thread-name tag, and
+    interface port. Returns up to max_lines most-recent matches.
+
+    Time strings: "-15m", "-2h", "-3d", "now", or ISO 8601.
+
+    Note: lines arriving via the LogRecord protobuf path (when
+    set_debug_log_api(True) is on) come without level prefix — the
+    meshtastic Python lib drops record.level before fan-out. For those,
+    `level` filter won't match; use `grep` instead.
+    """
+    return log_query.logs_window(
+        start=start,
+        end=end,
+        grep=grep,
+        level=level,
+        tag=tag,
+        port=port,
+        max_lines=max_lines,
+    )
+
+
+@app.tool()
+def telemetry_timeline(
+    window: str = "1h",
+    variant: str = "local",
+    field: str = "free_heap",
+    port: str | None = None,
+    max_points: int = 200,
+) -> dict[str, Any]:
+    """Time series of one telemetry field, downsampled to <= max_points.
+
+    `variant` ∈ device, local, environment, power, airQuality, health, host.
+    `field` accepts snake_case or camelCase; common aliases (free_heap ↔
+    heap_free_bytes) are normalized.
+
+    Returns slope_per_min (linear-regression slope, units/minute) so a
+    leak detector can read one number — negative slope on free_heap over
+    a long window indicates a real leak.
+
+    LocalStats variant ("local") cadence is ~60 s (whatever the device's
+    `device_update_interval` is set to), so a 1 h window gives ~60 raw
+    points. Bucket-mean downsampling preserves shape.
+    """
+    return log_query.telemetry_timeline(
+        window=window,
+        variant=variant,
+        field=field,
+        port=port,
+        max_points=max_points,
+    )
+
+
+@app.tool()
+def packets_window(
+    start: str = "-5m",
+    end: str = "now",
+    portnum: str | None = None,
+    from_node: str | None = None,
+    to_node: str | None = None,
+    max: int = 200,
+) -> dict[str, Any]:
+    """Recent mesh packets recorded by the recorder.
+
+    Each row is a summary (portnum, from/to, hop_limit, RSSI/SNR, payload
+    size + first 64 bytes hex) — full payload bytes are not stored.
+    `portnum` accepts a pipe-separated set like "TEXT_MESSAGE_APP|POSITION_APP".
+    """
+    return log_query.packets_window(
+        start=start,
+        end=end,
+        portnum=portnum,
+        from_node=from_node,
+        to_node=to_node,
+        max=max,
+    )
+
+
+@app.tool()
+def events_window(
+    start: str = "-1h",
+    end: str = "now",
+    kind: str | None = None,
+    max: int = 200,
+) -> dict[str, Any]:
+    """Return recorder events: connection lifecycle, node updates, and `mark_event` markers.
+
+    `kind` ∈ recorder_start, recorder_pause, recorder_resume,
+    connection_established, connection_lost, node_updated, mark.
+    Pipe-separated sets ("connection_lost|connection_established") work.
+    """
+    return log_query.events_window(start=start, end=end, kind=kind, max=max)
+
+
+@app.tool()
+def mark_event(
+    label: str,
+    note: str | None = None,
+    data: dict[str, Any] | None = None,
+) -> dict[str, Any]:
+    """Drop a named marker into events.jsonl AND logs.jsonl.
+
+    Useful for aligning a timeline around a known stimulus: call before
+    and after a stress workload, then query telemetry_timeline /
+    logs_window with the markers' timestamps as bounds.
+
+    The marker also lands in logs.jsonl with level=MARK so a single
+    grep over logs picks it up.
+    """
+    return get_recorder().mark_event(label=label, note=note, data=data)
+
+
+@app.tool()
+def recorder_status() -> dict[str, Any]:
+    """Return recorder runtime info: running, paused, file sizes, last_ts per stream.
+
+    Use this to sanity-check that capture is working before you trust a
+    `logs_window` / `telemetry_timeline` result.
+    """
+    return get_recorder().status()
+
+
+@app.tool()
+def recorder_pause(reason: str | None = None) -> dict[str, Any]:
+    """Pause writes to all four streams. Pubsub subscriptions stay active —
+    we just drop events on the floor while paused. Resume with `recorder_resume`.
+
+    Use when capturing a known-good baseline that you don't want to
+    pollute with pre-test noise. Default state is recording; this is
+    rarely needed.
+    """
+    get_recorder().pause(reason=reason)
+    return {"ok": True, "paused": True, "reason": reason}
+
+
+@app.tool()
+def recorder_resume() -> dict[str, Any]:
+    """Resume writes after `recorder_pause`. No-op if already running."""
+    get_recorder().resume()
+    return {"ok": True, "paused": False}
+
+
+@app.tool()
+def recorder_export(
+    start: str,
+    end: str,
+    dest_dir: str,
+    streams: list[str] | None = None,
+) -> dict[str, Any]:
+    """Bundle a slice of the recorder's streams into `dest_dir`.
+
+    Writes one uncompressed JSONL per requested stream (logs / telemetry /
+    packets / events). Useful for: attaching to a bug report, feeding a
+    notebook, or backfilling Datadog after the fact.
+    """
+    return log_query.export(
+        start=start,
+        end=end,
+        dest_dir=dest_dir,
+        streams=streams,
+    )
+
+
+# ---------- Fixture / test-data push --------------------------------------
+
+
+@app.tool()
+def push_fake_nodedb(
+    size: int,
+    target: str = "portduino",
+    port: str | None = None,
+    portduino_config: str = "default",
+    backup_existing: bool = True,
+    confirm: bool = False,
+    reboot_after: bool = True,
+    custom_seed_jsonl: str | None = None,
+) -> dict[str, Any]:
+    """Push a fake-NodeDB v25 fixture (250/500/1000/2000 nodes) onto a device.
+
+    Two transports:
+      target="portduino" — file copy to ~/.portduino/<portduino_config>/prefs/nodes.proto.
+                            Fast, no device connection needed.
+      target="hardware"  — XModem upload over serial/BLE to /prefs/nodes.proto.
+                            Requires `port` + `confirm=True`. Triggers a reboot
+                            so loadFromDisk picks up the new file at next boot.
+
+    Compiles a fresh-timestamp proto from the committed JSONL seed under
+    test/fixtures/nodedb/seed_v25_<N>.jsonl each invocation, so the loaded
+    NodeDB always looks "recent" to the connecting phone. Structural data
+    (names, IDs, positions, telemetries) is deterministic per the seed.
+
+    Override the JSONL via `custom_seed_jsonl` to push a hand-edited scenario.
+    """
+    return fixtures.push_fake_nodedb(
+        size=size,
+        target=target,  # type: ignore[arg-type]
+        port=port,
+        portduino_config=portduino_config,
+        backup_existing=backup_existing,
+        confirm=confirm,
+        reboot_after=reboot_after,
+        custom_seed_jsonl=custom_seed_jsonl,
+    )
@@ -0,0 +1,321 @@
+"""USB hub power control via `uhubctl` — hard-recovery for wedged devices +
+deliberate offline-peer simulation for mesh tests.
+
+Why: when a Meshtastic device's serial port wedges (stuck in a boot loop,
+frozen USB CDC, crashed firmware that didn't reboot), the only recovery is
+a physical unplug. uhubctl toggles VBUS per-port on any hub with Per-Port
+Power Switching (PPPS) support — which is most externally-powered hubs
+from the last ~5 years — so the harness can power-cycle a device
+programmatically.
+
+Architecture:
+- `list_hubs()` parses `uhubctl` default output into structured records.
+- `find_port_for_vid(vid)` walks the hubs to find which location+port
+  hosts a given USB VID.
+- `resolve_target(role)` is the public entry for callers that know a role
+  (`nrf52`, `esp32s3`) but not a hub location: env-var pins win, VID
+  auto-detect falls back.
+- `power_on`, `power_off`, `cycle` wrap the corresponding `uhubctl -a`
+  invocations, routed through `hw_tools._run` so they share tee-to-flash-
+  log + timeout handling with esptool / nrfutil / picotool.
+
+Sudo policy: **fail fast**. Modern macOS + most PPPS-capable hubs work
+without root, but Linux without udev rules (or old macOS with specific
+driver quirks) still needs it. We run uhubctl non-root; if stderr
+matches the classic permission pattern we raise `UhubctlError` with an
+install hint pointing at the uhubctl docs. Auto-wrapping with `sudo`
+would prompt in the middle of test runs — bad for CI.
+"""
+
+from __future__ import annotations
+
+import os
+import re
+from typing import Any, Sequence
+
+from . import config, hw_tools
+
+# ---------- Parser ---------------------------------------------------------
+
+# Hub descriptor line:
+#   Current status for hub 1-1.3 [2109:2817 VIA Labs, Inc. USB2.0 Hub, USB 2.10, 4 ports, ppps]
+_HUB_RE = re.compile(
+    r"^Current status for hub (?P<location>\S+)\s+\[(?P<descriptor>.+)\]\s*$"
+)
+
+# Port line:
+#   "  Port 2: 0103 power enable connect [239a:8029 RAKwireless ...]"
+# The bracketed section is absent for empty ports.
+_PORT_RE = re.compile(
+    r"^\s+Port\s+(?P<port>\d+):\s+(?P<status>\S+)\s+(?P<flags>.*?)"
+    r"(?:\s+\[(?P<device_vid>[0-9a-fA-F]{4}):(?P<device_pid>[0-9a-fA-F]{4})(?:\s+(?P<device_desc>.+))?\])?\s*$"
+)
+
+
+class UhubctlError(RuntimeError):
+    """Raised on uhubctl-specific failures: parse errors, permission denied,
+    hub-not-found, or PPPS not supported."""
+
+
+# ---------- Role → VID map -------------------------------------------------
+
+# Mirrors the default hub_profile in `mcp-server/tests/conftest.py:335`.
+# Note: esp32s3 and esp32s3_alt share a logical role — we search both.
+ROLE_VIDS: dict[str, tuple[int, ...]] = {
+    "nrf52": (0x239A,),
+    "esp32s3": (0x303A, 0x10C4),
+}
+
+
+def _normalize_role(role: str) -> str:
+    """Collapse `esp32s3_alt` → `esp32s3` to match the tier conventions."""
+    return role.split("_alt", 1)[0].lower()
+
+
+# ---------- Core subprocess runner -----------------------------------------
+
+
+# If uhubctl hits a permission problem — most commonly Linux without the
+# udev rules, or a macOS variant where the kernel holds the hub driver —
+# it prints something like "Permission denied. Try running as root".
+# Linux error text varies; we match a broad substring rather than exact.
+_PERM_ERROR_PATTERNS = (
+    "permission denied",
+    "operation not permitted",
+    "try running as root",
+    "need root",
+    "requires root",
+)
+
+
+def _run_uhubctl(args: Sequence[str], *, timeout: float = 30.0) -> dict[str, Any]:
+    """Invoke uhubctl with the given args. Returns `hw_tools._run`'s dict.
+
+    Translates permission-denied failures into a `UhubctlError` with the
+    install hint, so callers don't have to match stderr themselves. Other
+    non-zero exits are returned as-is for the caller to interpret.
+    """
+    binary = config.uhubctl_bin()
+    result = hw_tools._run(binary, args, timeout=timeout)  # noqa: SLF001
+    if result["exit_code"] != 0:
+        combined = (result.get("stderr") or "") + "\n" + (result.get("stdout") or "")
+        lower = combined.lower()
+        if any(pat in lower for pat in _PERM_ERROR_PATTERNS):
+            raise UhubctlError(
+                "uhubctl exited with a permission error. Install the udev "
+                "rules on Linux, or try `sudo` as a fallback: "
+                "https://github.com/mvp/uhubctl#linux-usb-permissions\n"
+                f"stderr: {result.get('stderr_tail')!r}"
+            )
+    return result
+
+
+# ---------- List / parse ---------------------------------------------------
+
+
+def parse_list_output(output: str) -> list[dict[str, Any]]:
+    """Parse the default `uhubctl` stdout into structured hubs.
+
+    Each hub: {
+        "location":   "1-1.3",
+        "descriptor": "2109:2817 VIA Labs ...",
+        "vid":        0x2109,
+        "pid":        0x2817,
+        "ppps":       bool,
+        "ports":      [{"port": int, "status": str, "flags": str,
+                         "device_vid": int | None, "device_pid": int | None,
+                         "device_desc": str | None}, ...],
+    }
+    """
+    hubs: list[dict[str, Any]] = []
+    current: dict[str, Any] | None = None
+
+    for line in output.splitlines():
+        hm = _HUB_RE.match(line)
+        if hm:
+            descriptor = hm.group("descriptor")
+            hub_vid, hub_pid = None, None
+            vid_match = re.match(r"([0-9a-fA-F]{4}):([0-9a-fA-F]{4})", descriptor)
+            if vid_match:
+                hub_vid = int(vid_match.group(1), 16)
+                hub_pid = int(vid_match.group(2), 16)
+            current = {
+                "location": hm.group("location"),
+                "descriptor": descriptor,
+                "vid": hub_vid,
+                "pid": hub_pid,
+                "ppps": ", ppps" in descriptor or descriptor.endswith("ppps"),
+                "ports": [],
+            }
+            hubs.append(current)
+            continue
+
+        pm = _PORT_RE.match(line)
+        if pm and current is not None:
+            device_vid = pm.group("device_vid")
+            device_pid = pm.group("device_pid")
+            current["ports"].append(
+                {
+                    "port": int(pm.group("port")),
+                    "status": pm.group("status"),
+                    "flags": (pm.group("flags") or "").strip(),
+                    "device_vid": int(device_vid, 16) if device_vid else None,
+                    "device_pid": int(device_pid, 16) if device_pid else None,
+                    "device_desc": (pm.group("device_desc") or "").strip() or None,
+                }
+            )
+    return hubs
+
+
+def list_hubs() -> list[dict[str, Any]]:
+    """Enumerate every hub uhubctl can see, with per-port device attachments.
+
+    Pure read — no power state changes. Useful as a pre-flight check before
+    a destructive `power_off` call.
+    """
+    result = _run_uhubctl([], timeout=15.0)
+    if result["exit_code"] != 0:
+        raise UhubctlError(
+            f"uhubctl list failed (exit {result['exit_code']}): {result.get('stderr_tail')!r}"
+        )
+    return parse_list_output(result["stdout"])
+
+
+# ---------- Lookup / resolution -------------------------------------------
+
+
+def find_port_for_vid(
+    vid: int, pid: int | None = None, *, only_ppps: bool = True
+) -> list[tuple[str, int]]:
+    """Return ALL (location, port) matches for a device VID (optionally +PID).
+
+    `only_ppps=True` filters out hubs that don't advertise PPPS — we can't
+    control them anyway. Callers that want to diagnose a missing device can
+    pass `only_ppps=False` to see if the device is on a non-controllable
+    hub (and raise a clearer error).
+    """
+    hubs = list_hubs()
+    matches: list[tuple[str, int]] = []
+    for hub in hubs:
+        if only_ppps and not hub["ppps"]:
+            continue
+        for port in hub["ports"]:
+            if port["device_vid"] != vid:
+                continue
+            if pid is not None and port["device_pid"] != pid:
+                continue
+            matches.append((hub["location"], port["port"]))
+    return matches
+
+
+def resolve_target(role: str) -> tuple[str, int]:
+    """Resolve a Meshtastic role to (hub_location, port_number).
+
+    Priority:
+      1. Env vars `MESHTASTIC_UHUBCTL_LOCATION_<ROLE>` + `_PORT_<ROLE>`
+         (e.g. `MESHTASTIC_UHUBCTL_LOCATION_NRF52=1-1.3`, `_PORT_NRF52=2`).
+      2. VID auto-detect against `ROLE_VIDS[role]`, taking the first PPPS
+         match.
+
+    Raises `UhubctlError` on ambiguity (multiple matches) or no-match. The
+    env-var path exists specifically to disambiguate when two devices share
+    a VID.
+    """
+    role = _normalize_role(role)
+    env_key_loc = f"MESHTASTIC_UHUBCTL_LOCATION_{role.upper()}"
+    env_key_port = f"MESHTASTIC_UHUBCTL_PORT_{role.upper()}"
+    loc = os.environ.get(env_key_loc)
+    port_str = os.environ.get(env_key_port)
+    if loc and port_str:
+        try:
+            return (loc, int(port_str))
+        except ValueError as exc:
+            raise UhubctlError(
+                f"{env_key_port}={port_str!r} is not a valid integer"
+            ) from exc
+
+    if role not in ROLE_VIDS:
+        raise UhubctlError(
+            f"unknown role {role!r}; known roles: {sorted(ROLE_VIDS)}. "
+            f"Set {env_key_loc} + {env_key_port} to pin manually."
+        )
+
+    matches: list[tuple[str, int]] = []
+    for vid in ROLE_VIDS[role]:
+        matches.extend(find_port_for_vid(vid))
+
+    if not matches:
+        vids = ", ".join(f"0x{v:04x}" for v in ROLE_VIDS[role])
+        raise UhubctlError(
+            f"no controllable hub hosts a device with VID in {{{vids}}} "
+            f"for role={role!r}. Check the device is plugged into a "
+            f"PPPS-capable hub, or pin manually via {env_key_loc} + {env_key_port}."
+        )
+    if len(matches) > 1:
+        shown = ", ".join(f"{loc}:port{p}" for loc, p in matches)
+        raise UhubctlError(
+            f"ambiguous: multiple devices match role={role!r} ({shown}). "
+            f"Pin the target via {env_key_loc} + {env_key_port}."
+        )
+    return matches[0]
+
+
+# ---------- Power actions --------------------------------------------------
+
+
+def _action(
+    action: str,
+    location: str,
+    port: int,
+    *,
+    delay_s: int | None = None,
+    timeout: float = 30.0,
+) -> dict[str, Any]:
+    args: list[str] = ["-a", action, "-l", location, "-p", str(port)]
+    if delay_s is not None:
+        args.extend(["-d", str(delay_s)])
+    # Suppress verbose "before" printout so our parser doesn't have to skip it.
+    args.append("-N")
+    result = _run_uhubctl(args, timeout=timeout)
+    if result["exit_code"] != 0:
+        raise UhubctlError(
+            f"uhubctl -a {action} -l {location} -p {port} failed "
+            f"(exit {result['exit_code']}): {result.get('stderr_tail')!r}"
+        )
+    return {
+        "action": action,
+        "location": location,
+        "port": port,
+        "delay_s": delay_s,
+        "duration_s": result["duration_s"],
+    }
+
+
+def power_on(location: str, port: int) -> dict[str, Any]:
+    """Drive the port VBUS high. Device re-enumerates in 1-3 s on healthy hubs."""
+    return _action("on", location, port)
+
+
+def power_off(location: str, port: int) -> dict[str, Any]:
+    """Drive the port VBUS low. Device disappears from `list_devices` immediately."""
+    return _action("off", location, port)
+
+
+def cycle(location: str, port: int, delay_s: int = 2) -> dict[str, Any]:
+    """Off → wait `delay_s` → on. The common hard-reset pattern."""
+    # uhubctl's own `-a cycle` handles the delay internally; we use a
+    # slightly longer timeout to accommodate delay_s + enumeration.
+    return _action("cycle", location, port, delay_s=delay_s, timeout=30.0 + delay_s * 2)
+
+
+__all__ = [
+    "ROLE_VIDS",
+    "UhubctlError",
+    "cycle",
+    "find_port_for_vid",
+    "list_hubs",
+    "parse_list_output",
+    "power_off",
+    "power_on",
+    "resolve_target",
+]
@@ -393,6 +393,7 @@ def build_testing_profile(
    long_name: str | None = None,
    disable_mqtt: bool = True,
    disable_position: bool = False,
+    enable_ui_log: bool = False,
 ) -> dict[str, Any]:
    """Build a USERPREFS dict for an isolated test-mesh device.

@@ -423,6 +424,10 @@ def build_testing_profile(
            traffic never leaks to a public broker.
        disable_position: if True, disables GPS + position broadcasts — useful
            when test devices sit on a bench without antennas.
+        enable_ui_log: if True, stamps `USERPREFS_UI_TEST_LOG=true` so the
+            firmware emits one `Screen: frame N/M name=... reason=...` log
+            line per frame transition. Test-only; off by default because the
+            log is chatty (multiple times per second during UI interaction).

    """
    if region not in KNOWN_REGIONS:
@@ -475,6 +480,9 @@ def build_testing_profile(
        prefs["USERPREFS_CONFIG_OWNER_LONG_NAME"] = long_name
    if short_name is not None:
        prefs["USERPREFS_CONFIG_OWNER_SHORT_NAME"] = short_name
+    if enable_ui_log:
+        # Consumed by `#ifdef USERPREFS_UI_TEST_LOG` in src/graphics/Screen.cpp.
+        prefs["USERPREFS_UI_TEST_LOG"] = True

    return prefs

@@ -0,0 +1,112 @@
+"""USB hub power control for tests — thin composition of the `uhubctl`
+module + `_port_discovery.resolve_port_by_role`.
+
+Why separate from the production module:
+- `meshtastic_mcp.uhubctl.cycle` returns as soon as uhubctl exits (VBUS is
+  back on, but the device hasn't finished enumerating as a CDC port yet).
+- Tests that want to immediately issue a `connect(port=...)` need the NEW
+  `/dev/cu.*` path, which can differ from the pre-cycle path on nRF52
+  boards (CDC re-enumeration assigns a fresh `cu.usbmodemNNNN`).
+- `resolve_port_by_role` already handles that wait + path-resolution for
+  the `factory_reset` flow. Composing the two gives a one-call helper.
+
+Also exposes `is_uhubctl_available()` so fixtures can skip cleanly when
+uhubctl isn't installed — we never want "no uhubctl" to look like a test
+failure.
+"""
+
+from __future__ import annotations
+
+import time
+from typing import Any
+
+from meshtastic_mcp import config as config_mod
+from meshtastic_mcp import uhubctl as uhubctl_mod
+
+from ._port_discovery import resolve_port_by_role
+
+
+def is_uhubctl_available() -> bool:
+    """Return True iff `config.uhubctl_bin()` resolves AND the binary is callable.
+
+    Soft-fails silently — fixtures use this to `pytest.skip` with an
+    actionable message when the operator hasn't installed uhubctl.
+    """
+    try:
+        config_mod.uhubctl_bin()
+    except Exception:  # noqa: BLE001
+        return False
+    # Do NOT actually invoke uhubctl here — on macOS a non-sudo run would
+    # fail, which is a config issue, not a tool-missing issue. That gets
+    # surfaced to the user when they actually run a recovery action.
+    return True
+
+
+def power_on(role: str) -> dict[str, Any]:
+    """Power on the hub port hosting `role`. Does NOT wait for re-enumeration.
+    Use `power_cycle` or follow with `resolve_port_by_role` to block on readiness.
+    """
+    loc, port = uhubctl_mod.resolve_target(role)
+    return uhubctl_mod.power_on(loc, port)
+
+
+def power_off(role: str) -> dict[str, Any]:
+    """Power off the hub port hosting `role`. The device disappears from
+    `list_devices` immediately.
+    """
+    loc, port = uhubctl_mod.resolve_target(role)
+    return uhubctl_mod.power_off(loc, port)
+
+
+def power_cycle(
+    role: str,
+    *,
+    delay_s: int = 2,
+    rediscover_timeout_s: float = 30.0,
+) -> str:
+    """Cycle the port hosting `role`, wait for re-enumeration, return the
+    new port path.
+
+    On nRF52 the post-cycle path typically matches the pre-cycle path, but
+    macOS may assign a different `/dev/cu.usbmodemNNNN` if the previous
+    CDC endpoint hasn't been fully released. `resolve_port_by_role`
+    handles that transparently.
+    """
+    loc, port = uhubctl_mod.resolve_target(role)
+    uhubctl_mod.cycle(loc, port, delay_s=delay_s)
+    # After uhubctl exits, VBUS is on but the device may still be in
+    # bootloader init. Give it ~500 ms head-start before polling so we
+    # don't spam list_devices pointlessly.
+    time.sleep(0.5)
+    return resolve_port_by_role(role, timeout_s=rediscover_timeout_s)
+
+
+def wait_for_absence(role: str, *, timeout_s: float = 10.0) -> None:
+    """Block until a device matching `role` is NOT in `list_devices`.
+
+    Used by the recovery tier to assert power_off actually took effect.
+    Raises TimeoutError on failure.
+    """
+    from meshtastic_mcp import devices as devices_mod
+
+    from ._port_discovery import _ROLE_VIDS, _coerce_vid  # type: ignore[attr-defined]
+
+    if role not in _ROLE_VIDS:
+        raise ValueError(f"unknown role {role!r}")
+    wanted = _ROLE_VIDS[role]
+    deadline = time.monotonic() + timeout_s
+    while time.monotonic() < deadline:
+        found = devices_mod.list_devices(include_unknown=True)
+        if not any(_coerce_vid(d.get("vid")) in wanted for d in found):
+            return
+        time.sleep(0.3)
+    raise TimeoutError(f"role {role!r} still visible after {timeout_s}s of power_off")
+
+
+__all__ = [
+    "is_uhubctl_available",
+    "power_cycle",
+    "power_off",
+    "power_on",
+    "wait_for_absence",
+]
@@ -123,15 +123,24 @@ def pytest_collection_modifyitems(
            return (2, item.nodeid)
        if "/monitor/" in path or "tests/monitor" in path:
            return (3, item.nodeid)
-        if "/fleet/" in path or "tests/fleet" in path:
+        # Recovery tier: explicitly cycles device power via uhubctl. Slots
+        # between monitor (read-only) and ui (state-preserving) so any tier
+        # after it starts from a known re-enumerated + re-verified state.
+        if "/recovery/" in path or "tests/recovery" in path:
            return (4, item.nodeid)
+        # UI tier slots here — read-only w.r.t. mesh state, only mutates
+        # the on-screen UI (BACK×5 guard restores home before each test).
+        if "/ui/" in path or "tests/ui" in path:
+            return (5, item.nodeid)
+        if "/fleet/" in path or "tests/fleet" in path:
+            return (6, item.nodeid)
        # State-mutating tiers run last.
        if "/admin/" in path or "tests/admin" in path:
-            return (5, item.nodeid)
+            return (7, item.nodeid)
        if "/provisioning/" in path or "tests/provisioning" in path:
-            return (6, item.nodeid)
+            return (8, item.nodeid)
        # Top-level + anything else falls between.
-        return (7, item.nodeid)
+        return (9, item.nodeid)

    items.sort(key=sort_key)

@@ -156,13 +165,20 @@ def session_seed(request: pytest.FixtureRequest) -> str:

@pytest.fixture(scope="session")
 def test_profile(session_seed: str) -> dict[str, Any]:
-    """The canonical isolated-mesh test profile for this session."""
+    """The canonical isolated-mesh test profile for this session.
+
+    `enable_ui_log=True` stamps `USERPREFS_UI_TEST_LOG` so the firmware
+    emits `Screen: frame N/M name=... reason=...` log lines per UI
+    transition — consumed by the `tests/ui/` tier. Harmless on boards
+    without a screen (the `#ifdef` sits behind `HAS_SCREEN`).
+    """
    return userprefs.build_testing_profile(
        psk_seed=session_seed,
        channel_name="McpTest",
        channel_num=88,
        region="US",
        modem_preset="LONG_FAST",
+        enable_ui_log=True,
    )


@@ -654,6 +670,7 @@ def pytest_generate_tests(metafunc: pytest.Metafunc) -> None:
 def baked_single(
    baked_mesh: dict[str, Any],
    baked_single_role: str,
+    hub_devices: dict[str, str],
 ) -> dict[str, Any]:
    """Function-scoped: a single verified baked device.

@@ -662,10 +679,75 @@ def baked_single(
    (e.g. `test_owner_survives_reboot[nrf52]` +
    `test_owner_survives_reboot[esp32s3]`). Tests never hardcode a role
    and never skip a device that happens to be connected.
+
+    Auto-recovery: if the baked device fails a pre-test `device_info` probe
+    AND uhubctl is available, power-cycle the port once and retry. Without
+    uhubctl, surface the wedge as a clear skip. This catches "device got
+    stuck between tests" without masking persistent regressions (a second
+    wedge after cycling still skips).
    """
    if baked_single_role not in baked_mesh:
        pytest.skip(f"role {baked_single_role!r} not present on the hub")
-    return {"role": baked_single_role, **baked_mesh[baked_single_role]}
+
+    entry = baked_mesh[baked_single_role]
+    port = entry.get("port")
+    if port:
+        try:
+            _run_with_timeout(lambda: info.device_info(port=port, timeout_s=3.0), 5.0)
+        except Exception:
+            # Device didn't respond. Try a power-cycle recovery if uhubctl
+            # is installed; otherwise surface a skip that names the root
+            # cause clearly.
+            from tests import _power
+
+            if not _power.is_uhubctl_available():
+                pytest.skip(
+                    f"device {baked_single_role!r} unresponsive on {port}; "
+                    "install uhubctl (`brew install uhubctl` / `apt install "
+                    "uhubctl`) for auto power-cycle recovery"
+                )
+            try:
+                new_port = _power.power_cycle(baked_single_role, delay_s=2)
+            except Exception as exc:  # noqa: BLE001
+                pytest.skip(
+                    f"device {baked_single_role!r} wedged and power-cycle "
+                    f"failed: {exc}"
+                )
+            # Mutate both the session-scoped `hub_devices` map AND the
+            # baked_mesh entry so downstream fixtures see the recovered port.
+            hub_devices[baked_single_role] = new_port
+            baked_mesh[baked_single_role]["port"] = new_port
+            entry = baked_mesh[baked_single_role]
+    return {"role": baked_single_role, **entry}
+
+
+@pytest.fixture
+def power_cycle(
+    hub_devices: dict[str, str],
+) -> Callable[..., str]:
+    """Return a callable `(role, delay_s=2) -> new_port` that hard-resets the
+    hub port hosting `role`. Skips the test cleanly when uhubctl isn't
+    installed — never want "no uhubctl" to look like a test failure.
+
+    The callable mutates `hub_devices[role]` in place so subsequent fixture
+    lookups pick up the post-cycle port (mirrors the pattern in
+    provisioning/test_userprefs_survive_factory_reset.py).
+    """
+    from tests import _power
+
+    if not _power.is_uhubctl_available():
+        pytest.skip(
+            "uhubctl not installed; this test needs it for power control. "
+            "Install via `brew install uhubctl` (macOS) or `apt install "
+            "uhubctl` (Debian/Ubuntu)."
+        )
+
+    def _cycle(role: str, delay_s: int = 2) -> str:
+        new_port = _power.power_cycle(role, delay_s=delay_s)
+        hub_devices[role] = new_port
+        return new_port
+
+    return _cycle


 _DEFAULT_ROLE_ENVS = {
@@ -960,6 +1042,45 @@ def _run_with_timeout(fn: Callable[[], Any], timeout: float) -> Any:
            raise TimeoutError(f"operation did not complete within {timeout}s") from exc


+def _attach_ui_captures(item: pytest.Item, report: Any) -> None:
+    """Embed per-step UI captures (PNG + OCR) into the pytest-html extras.
+
+    Runs for every UI-tier test on BOTH pass and fail so the HTML report
+    always shows the image strip + OCR transcript. Silently no-ops if
+    pytest-html isn't installed or the test didn't use `frame_capture`.
+    """
+    captures = getattr(item, "_ui_captures", None)
+    if not captures:
+        return
+    try:
+        from pytest_html import extras as html_extras  # type: ignore[import-untyped]
+    except ImportError:
+        return
+
+    existing = getattr(report, "extras", None) or []
+    extras_list = list(existing)
+    for cap in captures:
+        png_path = cap.get("png_path")
+        label = f"{cap.get('step', '?')}: {cap.get('label', '')}"
+        frame = cap.get("frame") or {}
+        frame_str = (
+            f" — frame {frame.get('idx')} {frame.get('name')!r}" if frame else ""
+        )
+        if png_path:
+            try:
+                with open(png_path, "rb") as fh:
+                    import base64
+
+                    b64 = base64.b64encode(fh.read()).decode("ascii")
+                extras_list.append(html_extras.png(b64, name=f"{label}{frame_str}"))
+            except OSError:
+                pass
+        ocr = (cap.get("ocr_text") or "").strip()
+        if ocr:
+            extras_list.append(html_extras.text(ocr, name=f"OCR: {label}{frame_str}"))
+    report.extras = extras_list  # type: ignore[attr-defined]
+
+
@pytest.hookimpl(hookwrapper=True)
 def pytest_runtest_makereport(item: pytest.Item, call: pytest.CallInfo[Any]) -> Any:
    """On test failure, attach serial capture + device state as report artifacts.
@@ -967,10 +1088,20 @@ def pytest_runtest_makereport(item: pytest.Item, call: pytest.CallInfo[Any]) ->
    Hard-bounded by `_run_with_timeout` — if the device is unreachable (stuck
    port, unbaked firmware, dead board), the dump is skipped rather than
    hanging the session.
+
+    For UI-tier tests, also embeds per-step camera captures + OCR on every
+    test (pass or fail) so the HTML report shows visual evidence of what
+    the device did.
    """
    outcome = yield
    report = outcome.get_result()

+    # Attach UI captures on any outcome (pass + fail) — these are the whole
+    # point of the UI tier. Do this before the failure-only branch below so
+    # passing tests still get their image strip.
+    if report.when == "call":
+        _attach_ui_captures(item, report)
+
    if report.when != "call" or report.outcome != "failed":
        return

@@ -0,0 +1,155 @@
+"""Isolation test for peer-offline-then-back mid-conversation.
+
+Verifies the mesh stack's behavior when a peer is physically powered
+off mid-send via uhubctl, then powered back on.
+
+Flow (parametrized over every directed mesh_pair):
+  1. Bilateral PKI warmup (same pattern as test_direct_with_ack).
+  2. TX sends a broadcast text "msg-1" — RX confirms receipt via pubsub.
+  3. Power OFF RX via uhubctl. The RX device disappears from the OS.
+  4. TX sends a directed text "msg-2" with wantAck=True. Firmware retries
+     internally for ~30s before giving up. Assertion: the packet object
+     was accepted by the TX stack (non-None) — we don't assert an ACK
+     since there's no peer to send one.
+  5. Power ON RX. Wait for re-enumeration + boot.
+  6. Bilateral PKI re-nudge — RX's in-RAM PKI cache was wiped on reboot,
+     so the first directed send may err=35 without a fresh NodeInfo ping.
+  7. TX sends a directed "msg-3" — RX receives it via pubsub, confirming
+     the mesh recovered.
+
+Skips cleanly if uhubctl isn't installed (via the `power_cycle` fixture's
+auto-skip). Skips for pair directions where RX isn't power-controllable
+(e.g. a USB-IF hub that doesn't support PPPS for its port).
+"""
+
+from __future__ import annotations
+
+import time
+from typing import Any
+
+import pytest
+from meshtastic_mcp.connection import connect
+from tests import _power
+from tests._port_discovery import resolve_port_by_role
+
+from ._receive import ReceiveCollector, nudge_nodeinfo
+
+
+@pytest.mark.timeout(360)
+def test_peer_offline_then_recovers(
+    mesh_pair: dict[str, Any],
+    power_cycle,  # noqa: ARG001 — forces uhubctl-availability skip
+    hub_devices: dict[str, str],
+) -> None:
+    tx_port = mesh_pair["tx"]["port"]
+    rx_node_num = mesh_pair["rx"]["my_node_num"]
+    tx_role = mesh_pair["tx_role"]
+    rx_role = mesh_pair["rx_role"]
+
+    unique_pre = f"peer-offline-pre-{tx_role}-to-{rx_role}-{int(time.time())}"
+    unique_post = f"peer-offline-post-{tx_role}-to-{rx_role}-{int(time.time())}"
+
+    # Step 1 + 2: warm up + confirm baseline delivery works before the test.
+    with ReceiveCollector(
+        mesh_pair["rx"]["port"], topic="meshtastic.receive.text"
+    ) as rx:
+        rx.broadcast_nodeinfo_ping()
+        with connect(port=tx_port) as tx_iface:
+            nudge_nodeinfo(tx_iface)
+            # Wait for bilateral PKI (RX pubkey in TX's nodesByNum).
+            deadline = time.monotonic() + 45.0
+            last_nudge = time.monotonic()
+            while time.monotonic() < deadline:
+                rec = (tx_iface.nodesByNum or {}).get(rx_node_num, {})
+                if rec.get("user", {}).get("publicKey"):
+                    break
+                if time.monotonic() - last_nudge > 15.0:
+                    rx.broadcast_nodeinfo_ping()
+                    nudge_nodeinfo(tx_iface)
+                    last_nudge = time.monotonic()
+                time.sleep(1.0)
+            else:
+                pytest.skip(
+                    f"bilateral PKI never completed ({tx_role}→{rx_role}); "
+                    "can't run the offline test without a warm baseline"
+                )
+
+            tx_iface.sendText(unique_pre, destinationId=rx_node_num, wantAck=True)
+            got = rx.wait_for(
+                lambda pkt: pkt.get("decoded", {}).get("text") == unique_pre,
+                timeout=30,
+            )
+            assert got is not None, (
+                f"baseline directed send ({tx_role}→{rx_role}) didn't land — "
+                "skipping offline test to avoid false positive"
+            )
+
+    # Step 3: power off RX. uhubctl skips the test with a clear message if
+    # the RX role isn't on a controllable hub.
+    try:
+        _power.power_off(rx_role)
+    except Exception as exc:  # noqa: BLE001
+        pytest.skip(f"can't power-control {rx_role!r}: {exc}")
+
+    try:
+        _power.wait_for_absence(rx_role, timeout_s=10.0)
+    except TimeoutError:
+        _power.power_on(rx_role)  # restore hub state before failing
+        resolve_port_by_role(rx_role, timeout_s=30.0)
+        pytest.fail(f"{rx_role!r} didn't disappear after power_off")
+
+    # Step 4: send to a peer that isn't there. Firmware will retry
+    # internally. We don't wait for an ACK (there won't be one); we just
+    # confirm TX's stack accepts the packet without crashing.
+    try:
+        with connect(port=tx_port) as tx_iface:
+            packet = tx_iface.sendText(
+                f"while-offline-{rx_role}",
+                destinationId=rx_node_num,
+                wantAck=True,
+            )
+            assert packet is not None
+            # Give firmware a moment to do a retry or two while RX is down.
+            time.sleep(5.0)
+    except Exception as exc:  # noqa: BLE001 — TX should survive the peer being gone
+        # Restore RX before reraising so the bench state is sane.
+        _power.power_on(rx_role)
+        resolve_port_by_role(rx_role, timeout_s=30.0)
+        raise AssertionError(f"TX crashed when sending to offline peer: {exc}") from exc
+
+    # Step 5: power RX back on + rediscover.
+    _power.power_on(rx_role)
+    time.sleep(0.5)
+    new_rx_port = resolve_port_by_role(rx_role, timeout_s=30.0)
+    hub_devices[rx_role] = new_rx_port
+
+    # Step 6 + 7: bilateral re-warmup + directed send that should now work.
+    with ReceiveCollector(new_rx_port, topic="meshtastic.receive.text") as rx:
+        # RX rebooted → its PKI cache is gone. Re-warm.
+        rx.broadcast_nodeinfo_ping()
+        with connect(port=tx_port) as tx_iface:
+            nudge_nodeinfo(tx_iface)
+            time.sleep(3.0)
+
+            got = None
+            for _attempt in range(3):
+                packet = tx_iface.sendText(
+                    unique_post,
+                    destinationId=rx_node_num,
+                    wantAck=True,
+                )
+                assert packet is not None
+                got = rx.wait_for(
+                    lambda pkt: pkt.get("decoded", {}).get("text") == unique_post,
+                    timeout=30,
+                )
+                if got is not None:
+                    break
+                rx.broadcast_nodeinfo_ping()
+                nudge_nodeinfo(tx_iface)
+                time.sleep(5.0)
+
+    assert got is not None, (
+        f"post-recovery directed send {unique_post!r} ({tx_role}→{rx_role}) "
+        "never landed — recovery path may be broken"
+    )
@@ -0,0 +1,6 @@
+"""Recovery tier — exercises `uhubctl` power control end-to-end.
+
+Requires `uhubctl` installed AND at least one connected device on a
+PPPS-capable hub. The whole tier skips cleanly via
+`tests/recovery/conftest.py::_recovery_tier_guard` when either is missing.
+"""
@@ -0,0 +1,44 @@
+"""Recovery-tier gating + shared helpers.
+
+Session-scoped guard skips the whole tier when uhubctl isn't installed.
+Tests under this directory assume uhubctl is callable AND that at least
+one hub role is detected on a PPPS-capable port.
+"""
+
+from __future__ import annotations
+
+import pytest
+
+
+@pytest.fixture(scope="session", autouse=True)
+def _recovery_tier_guard() -> None:
+    """Skip the tier when uhubctl is unavailable OR no device is on a
+    PPPS-capable hub. Prints the specific reason so operators know what
+    to fix."""
+    from tests import _power
+
+    if not _power.is_uhubctl_available():
+        pytest.skip(
+            "uhubctl not installed; recovery tier needs it. "
+            "Install via `brew install uhubctl` or `apt install uhubctl`.",
+            allow_module_level=True,
+        )
+
+    # Probe: can we even list hubs? (A macOS user without sudo gets a
+    # permission error here — we'd rather find out once at tier-start than
+    # 6 tests later.)
+    from meshtastic_mcp import uhubctl
+
+    try:
+        hubs = uhubctl.list_hubs()
+    except uhubctl.UhubctlError as exc:
+        pytest.skip(
+            f"uhubctl list failed: {exc}. Try the udev rules or `sudo` as a fallback.",
+            allow_module_level=True,
+        )
+
+    if not any(h["ppps"] for h in hubs):
+        pytest.skip(
+            "no PPPS-capable hubs detected — recovery tier has nothing to exercise.",
+            allow_module_level=True,
+        )
@@ -0,0 +1,43 @@
+"""Smoke test: `uhubctl_list` returns a well-formed structure.
+
+No destructive action. Runs first in the tier as a sanity check that the
+tier's dependencies (uhubctl binary + permissions) are actually satisfied.
+"""
+
+from __future__ import annotations
+
+import pytest
+from meshtastic_mcp import uhubctl
+
+
+@pytest.mark.timeout(30)
+def test_list_hubs_returns_at_least_one_ppps_hub() -> None:
+    hubs = uhubctl.list_hubs()
+    assert hubs, "uhubctl found no hubs at all — is a USB hub connected?"
+    assert any(h["ppps"] for h in hubs), (
+        "no PPPS-capable hubs detected; power control won't work. "
+        "Check that the hub supports Per-Port Power Switching."
+    )
+
+
+@pytest.mark.timeout(30)
+def test_list_hubs_structure(hub_devices: dict[str, str]) -> None:
+    hubs = uhubctl.list_hubs()
+    for hub in hubs:
+        assert "location" in hub and hub["location"]
+        assert "ports" in hub and isinstance(hub["ports"], list)
+        for port in hub["ports"]:
+            assert "port" in port and isinstance(port["port"], int)
+            assert "status" in port
+
+    # At least one of the detected Meshtastic roles should show up in some
+    # port's device_vid — otherwise the recovery tier can't drive them.
+    seen_vids = {
+        p["device_vid"] for h in hubs for p in h["ports"] if p["device_vid"] is not None
+    }
+    expected_any = {0x239A, 0x303A, 0x10C4} & seen_vids
+    assert expected_any or not hub_devices, (
+        f"hub_devices detected roles {list(hub_devices)} but uhubctl sees "
+        f"VIDs {sorted(hex(v) for v in seen_vids)} — the devices may be on "
+        "a hub that uhubctl can't see (e.g. built-in laptop ports)."
+    )
@@ -0,0 +1,60 @@
+"""Hard reset via uhubctl must NOT wipe NVS. Verify the test profile's
+region + channel survive a power-cycle.
+
+Guards against a regression where a firmware change treats unexpected
+power loss as a factory-reset trigger (e.g. bad EEPROM wear-leveling,
+erase-on-boot-for-safety). Such a regression would be catastrophic for
+field deployments.
+"""
+
+from __future__ import annotations
+
+import time
+
+import pytest
+from meshtastic_mcp import admin, info
+from tests import _power
+from tests._port_discovery import resolve_port_by_role
+
+
+@pytest.mark.timeout(180)
+def test_lora_config_survives_power_cycle(
+    baked_single: dict[str, object],
+    test_profile: dict[str, object],
+) -> None:
+    role = baked_single["role"]
+    pre_port = baked_single["port"]
+
+    pre_config = admin.get_config(section="lora", port=pre_port)["config"]["lora"]
+    pre_region = pre_config.get("region")
+    pre_preset = pre_config.get("modem_preset")
+    assert pre_region, f"lora.region not set pre-cycle on {role}"
+
+    # Hard power-cycle.
+    _power.power_cycle(role, delay_s=2)
+    time.sleep(0.5)
+    new_port = resolve_port_by_role(role, timeout_s=30.0)
+    # Let the firmware complete boot before admin reads.
+    time.sleep(2.0)
+    # Quick readiness probe.
+    probe = info.device_info(port=new_port, timeout_s=10.0)
+    assert (
+        probe.get("my_node_num") is not None
+    ), f"device {role!r} didn't respond after power-cycle"
+
+    post_config = admin.get_config(section="lora", port=new_port)["config"]["lora"]
+    post_region = post_config.get("region")
+    post_preset = post_config.get("modem_preset")
+
+    assert post_region == pre_region, (
+        f"lora.region wiped by power-cycle on {role}: "
+        f"pre={pre_region!r} post={post_region!r}"
+    )
+    assert post_preset == pre_preset, (
+        f"lora.modem_preset wiped by power-cycle on {role}: "
+        f"pre={pre_preset!r} post={post_preset!r}"
+    )
+
+    # Channel-0 name should also match the test profile.
+    pri_ch = admin.get_channel_url(port=new_port)
+    assert pri_ch.get("url"), f"channel URL empty after power-cycle on {role}"
--- a/Show More
+++ b/Show More