fix: ESP32 CSI 0pps (#521), aggregator sibling magics (#517), version.txt (#505) + fix-marker CI guard

[ruvnet]

Three open-issue fixes plus a new CI regression guard. Split into four reviewable commits.

Fixes

fix(esp32) — disable WiFi modem sleep so CSI capture isn't starved — closes #521

csi_collector_init() never called esp_wifi_set_ps(), so the radio stayed on the ESP-IDF STA default WIFI_PS_MIN_MODEM. The modem then sleeps between DTIM beacons; combined with the MGMT-only promiscuous filter (#396) the CSI callback is starved and the per-second yield collapses toward 0 pps — exactly the motion=0.00 presence=0.00 yield=0pps symptom reported on clean multi-node setups. Now forces WIFI_PS_NONE before enabling promiscuous mode (the textbook requirement for reliable CSI capture). New boot line: csi_collector: WiFi modem sleep disabled (WIFI_PS_NONE) for CSI capture. Battery duty-cycling is unaffected — power_mgmt_init() runs after this and re-enables modem sleep when provision.py --duty-cycle <100 is used. Builds clean for esp32s3 (idf.py build, 48% flash free). Needs hardware verification on a real ESP32-S3 before the next v*-esp32 tag.

fix(hardware) — aggregator tolerates sibling RuView UDP packet magics — closes #517

The firmware multiplexes 0xC5110002–0xC5110007 (vitals, feature, fused, compressed, feature-state, temporal) onto the same UDP port as ADR-018 raw CSI frames (0xC5110001). Esp32CsiParser only knew 0xC5110001, so the standalone aggregator binary printed parse error: Invalid magic: expected 0xc5110001, got 0xc5110002 for every vitals packet — noise, not data loss. Adds the sibling-magic constants + ruview_sibling_packet_name(), classifies recognized siblings before the CSI-frame length gate, and returns ParseError::NonCsiPacket { magic, kind } instead of InvalidMagic. The aggregator CLI now skips them quietly (--verbose → [skipped ADR-039 edge vitals packet — not a CSI frame]); the library-level CsiAggregator already dropped them silently. New regression tests cover all seven magics. cargo test -p wifi-densepose-hardware --no-default-features → 116 passed.

fix(firmware) — bump version.txt to 0.6.4 + CI guard for tag↔version — closes #505

version.txt on main was still 0.6.2; CMake reads PROJECT_VER from it, so esp_app_get_description()->version (and the boot log) reported 0.6.2 for any source build, and the v0.6.3-esp32 release binary internally identified as 0.6.2. Bumps version.txt → 0.6.4 and adds a version-guard job to firmware-ci.yml that runs on v*-esp32 tag pushes and fails the run when the tag's X.Y.Z ≠ version.txt.

New: ci — fix-marker regression guard (witness-style)

A fast per-PR gate that asserts previously-shipped fixes are still present in the tree — the CI analogue of the ruflo witness fix-marker system, but self-contained (no plugin dependency, reviewable as plain JSON). Complements the heavier checks (firmware build, deterministic pipeline proof, release witness bundle) by catching the silent-revert class of regression that build+test wouldn't.

• scripts/fix-markers.json — 11 markers: RuView#396, All ESP32-S3 nodes report motion=0.00 presence=0.00 yield=0pps despite correct setup #521, aggregater cannot parse some csi frame from firmware 0.6.4 #517, v0.6.3-esp32 release ships a binary that internally identifies as 0.6.2 #505, Desperately seeking help with ESP32-S3 tracking setup #354, [Bug/Enhancement]: Default fall_thresh (500) causing excessive false positives in high-traffic environments #263, Unable to receive the data from esp32 #266/esp32 node crash #321, 4MB flash support? #265, fix(esp32): use runtime node_id from NVS in outgoing packets #232/Firmware sends node_id=1 for all nodes (relates to #232) #375/Multi-node ESP32-S3 repro on Windows: 6 provisioned nodes transmit, but every UDP packet decodes as node_id = 1 #385/Live UI connects, packets flow, but pose does not move correctly with 6 ESP32-S3 nodes on Windows #386/ESP32-S3: g_nvs_config.node_id clobbered to 1 between main.c:140 and csi_collector_init + LoadProhibited panic loop #390, and the ADR-028 proof + witness-bundle artifacts. Each has files / require (literal substring or /regex/) / optional forbid / rationale / ref.
• scripts/check_fix_markers.py — stdlib-only checker. Exit 0 clean / 1 regression / 2 bad manifest. Modes: --list, --json, --only ID …. Runs on Windows + Linux.
• .github/workflows/fix-regression-guard.yml — runs on PR + push to main/master; gates on the checker, writes the result table into the run summary + an artifact.

$ python scripts/check_fix_markers.py
Fix-marker regression guard - 11 marker(s)
  [PASS] RuView#396 ...  [PASS] RuView#521 ...  ... (all 11) ...
All 11 fix markers present.

If a fix is intentionally removed, you update scripts/fix-markers.json in the same PR with a rationale — the diff becomes the audit trail.

Pre-merge notes

• Hardware-verify All ESP32-S3 nodes report motion=0.00 presence=0.00 yield=0pps despite correct setup #521 (esp_wifi_set_ps(WIFI_PS_NONE)) on a real ESP32-S3 before tagging a release
• CHANGELOG.md entry under [Unreleased] (Added: fix-marker guard; Fixed: All ESP32-S3 nodes report motion=0.00 presence=0.00 yield=0pps despite correct setup #521/aggregater cannot parse some csi frame from firmware 0.6.4 #517/v0.6.3-esp32 release ships a binary that internally identifies as 0.6.2 #505)
• Optional: add a line to the Pre-Merge Checklist in CLAUDE.md referencing python scripts/check_fix_markers.py

🤖 Generated with claude-flow

[ruvnet]

Hardware-verified on a real ESP32-S3 (COM8, d0:cf:13:44:01:84)

Flashed this branch's firmware build (idf.py reconfigure && build && flash) and watched the serial log:

I (253) app_init: App version:      0.6.4
I (257) app_init: Compile time:     May 11 2026 11:09:43
I (389) main: ESP32-S3 CSI Node (ADR-018) — v0.6.4 — Node ID: 1
I (629) wifi:connected with ruv.net, ... channel 5, BW20, ...
I (1809) wifi:Set ps type: 0, coexist: 0
I (1809) csi_collector: WiFi modem sleep disabled (WIFI_PS_NONE) for CSI capture
I (1819) csi_collector: Promiscuous mode enabled (MGMT-only, RuView#396)
I (1829) csi_collector: CSI cb #1: len=128 rssi=-63 ch=5
I (1869) csi_collector: CSI cb #2: len=128 rssi=-51 ch=5
...
I (5159) adaptive_ctrl: medium tick: state=8 yield=0pps  motion=0.00 presence=0.00 rssi=-42   ← partial first 1s window
I (6159) adaptive_ctrl: medium tick: state=5 yield=16pps motion=0.00 presence=0.00 rssi=-41
I (7159) adaptive_ctrl: medium tick: state=5 yield=15pps motion=0.00 presence=0.00 rssi=-38
I (10159) adaptive_ctrl: medium tick: state=6 yield=16pps motion=1.00 presence=11.66 rssi=-38
... steady 13–16 pps thereafter, presence detection live ...

• All ESP32-S3 nodes report motion=0.00 presence=0.00 yield=0pps despite correct setup #521 ✅ — esp_wifi_set_ps(WIFI_PS_NONE) applied (Set ps type: 0), the new boot line is present, the CSI callback fires, and yield holds at ~15 pps after the warm-up window (vs. the reported yield=0pps). Presence/motion detection works.
• v0.6.3-esp32 release ships a binary that internally identifies as 0.6.2 #505 ✅ — boot reports v0.6.4 and a fresh compile timestamp. Side note: the first incremental idf.py build still emitted 0.6.2 because CMake only re-reads PROJECT_VER from version.txt on idf.py reconfigure — a live demo of exactly the failure mode v0.6.3-esp32 release ships a binary that internally identifies as 0.6.2 #505 describes, and why the version-guard job in this PR matters. After reconfigure it bakes in correctly.
• aggregater cannot parse some csi frame from firmware 0.6.4 #517 — Rust-side; covered by the new tests (cargo test -p wifi-densepose-hardware --no-default-features → 116 passed) + a clean cargo check.

Checklist item "Hardware-verify #521" — done.

🤖 Generated with claude-flow