chore(firmware): add release guard against stale-sdkconfig partition mismatch (#1194)

While cutting v0.8.3-esp32, an incremental 8MB build reused a leftover
generated `sdkconfig` and silently linked the 4MB dual-OTA partition layout
(no spiffs, ota_1 @ 0x1F0000) — the would-be released `partition-table.bin`
did not match the 8MB `partitions_display.csv` it claimed.

scripts/firmware-release-guard.sh regenerates the expected partition table
from the CSV the named flash-size variant must use and byte-compares it to the
built `partition-table.bin`, and cross-checks flash size in flasher_args.json.
Fails closed so a release pipeline can't ship a mismatched table.

Usage: scripts/firmware-release-guard.sh <8mb|4mb> <build-dir>


Claude-Session: https://claude.ai/code/session_01AgpTcBLRJ32hUsKWxDXf36

.gitignore

@@ -287,3 +287,7 @@ harness/**/*.tgz
 harness/**/package-lock.json
 harness/**/.claude-flow/
 harness/**/ruvector.db
+
+# ruvector runtime/hook DB — never tracked (any depth)
+ruvector.db
+**/ruvector.db

CHANGELOG.md

@@ -8,6 +8,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## [Unreleased]
 
 ### Fixed
+- **Multistatic fusion never ran on a mixed-mode ESP32 mesh — live bridge fed raw, un-canonicalized per-node CSI to the fuser (#1170).** `node_frame_from_state` (`multistatic_bridge.rs`) wrapped each node's **raw** amplitude vector (HT20 ≈ 64 bins, HT40 ≈ 128/192) into a struct *named* `CanonicalCsiFrame` without ever resampling, so `MultistaticFuser::fuse` tripped `DimensionMismatch` on every cycle, silently fell back to per-node sum/dedup, and spun `total_engine_errors` unbounded. Added `HardwareNormalizer::resample_to_canonical` (resample-only, **no z-score** — preserves the amplitude scale the person-score's `variance/mean²` relies on) and run every node frame through it onto the canonical 56-tone grid before fusion. Heterogeneous meshes now fuse instead of erroring. Pinned by `heterogeneous_node_counts_canonicalize_and_fuse` (mixed 64/192 → fuses), `resample_to_canonical_is_length_only_no_zscore`, and an updated `test_node_frame_conversion`; the pre-existing `engine_bridge::observe_cycle_counts_engine_errors` was retargeted to force a `TimestampMismatch` (its old 56-vs-30 setup now canonicalizes cleanly). `wifi-densepose-signal` 501 / `wifi-densepose-sensing-server` 677 tests, 0 failed.
+- **`csi_fps_ema` reported the CSI frame rate 40–840× too high under bursty UDP delivery (#1180).** `update_csi_fps_ema` only rejected deltas `≤ 0` or `≥ 1 s`, so a 36 µs intra-burst arrival delta yielded `1/dt ≈ 27 kHz` straight into the EMA — the metric measured server arrival jitter, not the node's ~40 fps production rate. Added a `MIN_PLAUSIBLE_CSI_DT_SEC = 0.005` floor (derived from the firmware's 50 fps `CSI_MIN_SEND_INTERVAL_US` ceiling, ×4 slack) and made `observe_csi_frame_arrival` keep its anchor across sub-floor bursts so the next genuine inter-frame gap measures true cadence. Pinned by `subms_burst_delta_rejected`, `burst_interleaved_with_nominal_stays_in_band`, and `observe_csi_frame_arrival_ignores_subms_bursts`.
+- **`stream_sender` ENOMEM backoff starved low-rate control packets under a weak uplink (#1183, follow-up to #1135/#1159).** The global `s_backoff_until_us` gate (triggered by the 50 Hz CSI flood at weak RSSI) also suppressed the ≤48 B, ≤1 Hz `feature_state` / mesh `HEALTH` / sync packets that contribute negligible buffer pressure, so telemetry failed essentially every cycle. Added `stream_sender_send_priority()` — bypasses the backoff gate, reports ENOMEM quietly, and never extends/resets the global streak — and routed `feature_state`, HEALTH/anomaly (`rv_mesh_send`), and sync packets through it. Also fixed the misleading `"HEALTH sent"` log that printed unconditionally even when `rv_mesh_send` returned `ESP_FAIL` (now prints `sent`/`FAILED` from the actual return). Firmware builds clean (ESP-IDF v5.4).
 - **Multistatic fusion guard interval is now operator-configurable — fixes permanent trust demotion with WiFi-synced ESP32 nodes (#1049).** Two independently-clocked ESP32-S3 boards on ESP-NOW sync drift 10–150 ms (typ. ~70 ms) — the 100 ms beacon + WiFi-MAC jitter cannot hold them within the published 60 ms default guard, so the governed-trust cycle permanently demoted to `Restricted`, suppressed all pose output, and spun the error counter to 200k+ with **no escape hatch but a container restart**. Added a **direct `WDP_GUARD_INTERVAL_US` override** (+ optional `WDP_SOFT_GUARD_US`) to `multistatic_guard_config_from_env`, so a deployment can lift the hard guard past its measured spread (e.g. `WDP_GUARD_INTERVAL_US=200000`) without having to know its exact TDM schedule. Precedence is most-specific-wins: a direct override beats the existing `WDP_TDM_SLOTS`+`WDP_TDM_SLOT_US` schedule-derived guard, which beats the 60 ms/20 ms default; the override is applied on top of whichever base is selected, the soft band is always clamped strictly below the hard guard, and a malformed/zero value is ignored (falls back to the base rather than breaking fusion). The effective guard is now logged at startup. Pinned by 6 new tests (`multistatic_guard_config_tests`): direct-override-wins / beats-TDM-derived / soft-clamped-below-hard / lowering-hard-pulls-soft-down / malformed-or-zero-falls-back / default-when-unset. `wifi-densepose-sensing-server` bin tests **449 → 455**, 0 failed; Python proof VERDICT PASS, hash unchanged (off the signal proof path).
 
 ### Security

docs/adr/ADR-183-onboard-led-gamma-stimulus-csi-colormap.md

@@ -0,0 +1,98 @@
+# ADR-183: Onboard LED as a 40 Hz Gamma Stimulus, Colour-Mapped from Live CSI via `ruv-neural-viz`
+
+| Field | Value |
+|-------|-------|
+| **Status** | Accepted — implemented & hardware-confirmed on ESP32-S3 N16R8 (COM8) |
+| **Date** | 2026-06-17 |
+| **Deciders** | ruv |
+| **Codename** | **GAMMA-VIZ** |
+| **Builds on** | `ruv-neural-viz::ColorMap` (now `no_std` — ruvnet/ruv-neural#3 / RuView#1126), the ESP32 edge `motion_energy` metric (`edge_processing.c`), PR #962 (WS2812 on GPIO 48) |
+
+## Context
+
+Two threads converged. (1) `ruv-neural-viz::ColorMap` — the viridis/cool-warm
+palette the rUv-Neural stack uses to render brain-topology graphs — was `std`-only,
+so it couldn't run on the ESP32. (2) The onboard WS2812 on the S3 CSI node was dead
+weight: the firmware only cleared it on boot (and on the wrong pin for N16R8 — GPIO
+38 vs the actual 48, see #962).
+
+The ask: make the LED do something real and honest, using the project's own visual
+capability — not a decorative blink. The natural fit is a **40 Hz gamma stimulus**
+(the GENUS gamma-entrainment frequency from Alzheimer's light-therapy research)
+whose **colour is driven by live sensed motion**, so the node's front panel is both
+a known bio-stimulus waveform and a truthful readout of what the CSI is detecting.
+
+## Decision
+
+### Part A — make `ColorMap` `no_std`
+
+`colormap.rs` is self-contained (no cross-crate deps), so expose it on `no_std`
+targets. The only blockers were two `std`-only `f64` ops:
+
+- `f64::round` / `f64::abs` → replaced with `core`+`alloc`-safe helpers `fround`
+  (round via `f64 as i64` truncation — a `core` cast, no `libm`) and `fabs`.
+- `Vec`/`String`/`format!` → from `alloc`.
+
+The graph-bound modules (`animation`/`ascii`/`export`/`layout`) and their heavy deps
+move behind a default `std` feature; `--no-default-features` builds the crate `no_std`
+and exposes only `colormap`. Output is **byte-identical** (8/8 colormap tests pass with
+the same RGB values), so this is a pure portability change.
+
+### Part B — the LED stimulus (firmware)
+
+`firmware/esp32-csi-node/main/main.c`, on boot:
+
+- WS2812 on **GPIO 48** (N16R8 / DevKitC-1 v1.1; GPIO 8 on C6).
+- An `esp_timer` periodic at **12 500 µs toggles a square wave → 40 Hz, 50 % duty**
+  (full-on / full-off — a *perceptible* gamma flicker, not a colour drift).
+- **ON-phase colour = live CSI motion.** Each ON phase reads `edge_get_vitals().motion_energy`,
+  normalises it (`/ LED_MOTION_FULLSCALE`, clamped `[0,1]`), and indexes a **60-step
+  viridis LUT generated from `ColorMap::viridis().map()`** — still = dark purple,
+  strong motion = yellow.
+
+The LUT is baked from the real crate (Part A makes the same `ColorMap` embeddable
+for a future direct FFI path once the ESP Rust toolchain is in CI). The colours are
+therefore provably `ruv-neural-viz`'s, and the motion is provably real.
+
+## Honesty (what it is and is not)
+
+- **40 Hz is a real square-wave stimulus** (12.5 ms on / 12.5 ms off), not a label on
+  a colour sweep. It is *not* tied to any measured 40 Hz brain rhythm — it is an
+  *output* stimulus at the gamma frequency, not a readout of neural gamma.
+- **Colour is a real CSI readout** — `motion_energy` is the on-device phase-variance
+  motion metric the node already computes; no fabrication. At rest the LED sits at the
+  purple (low) end and flickers there.
+- No therapeutic claim is made. 40 Hz GENUS entrainment is cited as the *origin of the
+  frequency choice*, not as a validated medical effect of this device.
+
+## Consequences
+
+**Positive**
+- The LED is now an honest front-panel: gamma-frequency flicker + a live motion readout.
+- `ColorMap` is embeddable (`no_std`), unblocking on-device use of the rUv-Neural
+  palette beyond this LED.
+- Confirms #962's GPIO-48 fix visually (the LED lights on N16R8).
+
+**Negative / risks**
+- Changes the *default* firmware behaviour: the onboard LED animates instead of staying
+  off. Now **gated by `CONFIG_LED_GAMMA_VIZ`** (default `y`); set it `n` for a dark,
+  lower-power boot (the LED is just cleared) — no source change needed.
+- A 40 Hz flicker can be an issue for photosensitive users; document on the enclosure
+  and disable `CONFIG_LED_GAMMA_VIZ` in those deployments.
+- The saturation point is now `CONFIG_LED_MOTION_FULLSCALE_MILLI` (default 250 = 0.25),
+  operator-tunable; still not auto-calibrated per-environment.
+- The colour uses a baked LUT, not the live Rust `ColorMap` (FFI path deferred — needs
+  the ESP Rust/xtensa toolchain, not yet in CI).
+
+## Validation
+
+- `ruv-neural-viz`: `cargo build` (std) ✓, `cargo test colormap` 8/8 ✓ (identical RGB),
+  `cargo build --no-default-features` compiles `no_std` ✓.
+- Firmware: built (1.13 MB), flashed to ESP32-S3 N16R8 (COM8). Boot log:
+  `Onboard WS2812: 40 Hz gamma flicker (GENUS), colour=CSI motion via ruv-neural-viz, GPIO 48`;
+  CSI continues (27–38 pps), `motion=0.00` at rest → purple flicker as designed.
+- Full on-device (xtensa) Rust build of `ColorMap` not run — ESP Rust toolchain absent.
+
+## References
+- ruvnet/ruv-neural#3 (ColorMap no_std), RuView#1126 (submodule bump), #962 (GPIO 48).
+- Singer/Tsai GENUS 40 Hz gamma entrainment (origin of the frequency, not a device claim).

firmware/esp32-csi-node/main/Kconfig.projbuild

@@ -468,3 +468,29 @@ menu "Mock CSI (QEMU Testing)"
         depends on CSI_MOCK_ENABLED
         default n
 endmenu
+
+menu "Onboard LED (ADR-183)"
+
+    config LED_GAMMA_VIZ
+        bool "Onboard WS2812: 40 Hz gamma flicker + CSI-motion colour"
+        default y
+        help
+            Drive the onboard WS2812 as a GENUS-style 40 Hz gamma square wave
+            (12.5 ms on / 12.5 ms off, 50% duty). The ON-phase colour is live
+            CSI motion (edge motion_energy) mapped through the ruv-neural-viz
+            viridis colormap (still=purple, moving=yellow).
+
+            Disable to leave the LED off at boot — lower power, no flicker.
+            NOTE: a 40 Hz flicker can affect photosensitive users; disable or
+            shield the LED in those environments. Not a medical device.
+
+    config LED_MOTION_FULLSCALE_MILLI
+        int "Motion value (x1000) that saturates the colormap to yellow"
+        depends on LED_GAMMA_VIZ
+        default 250
+        range 1 100000
+        help
+            edge motion_energy that maps to the top (yellow) of the viridis
+            colormap, in milli-units (250 = 0.25). Lower = more sensitive
+            (reaches yellow with less motion).
+endmenu

firmware/esp32-csi-node/main/adaptive_controller.c

@@ -319,7 +319,9 @@ static void emit_feature_state(void)
                               (uint64_t)esp_timer_get_time(),
                               profile);
 
-    int sent = stream_sender_send((const uint8_t *)&pkt, sizeof(pkt));
+    /* feature_state is ~1 Hz and small — priority path so the CSI ENOMEM
+     * backoff can't starve it (#1183). */
+    int sent = stream_sender_send_priority((const uint8_t *)&pkt, sizeof(pkt));
     if (sent < 0) {
         ESP_LOGW(TAG, "feature_state emit failed");
     }
@@ -333,11 +335,14 @@ static void slow_loop_cb(TimerHandle_t t)
      * detect sync-error drift. */
     uint8_t nid[8];
     node_id_bytes(nid);
-    rv_mesh_send_health(s_role, s_mesh_epoch, nid);
+    /* #1183: report the actual send result — the old log printed "HEALTH sent"
+     * unconditionally even when rv_mesh_send returned ESP_FAIL. */
+    esp_err_t health_rc = rv_mesh_send_health(s_role, s_mesh_epoch, nid);
 
-    ESP_LOGI(TAG, "slow tick (state=%u, feature_state_seq=%u, role=%u, epoch=%u) HEALTH sent",
+    ESP_LOGI(TAG, "slow tick (state=%u, feature_state_seq=%u, role=%u, epoch=%u) HEALTH %s",
              (unsigned)s_state, (unsigned)s_feature_state_seq,
-             (unsigned)s_role, (unsigned)s_mesh_epoch);
+             (unsigned)s_role, (unsigned)s_mesh_epoch,
+             health_rc == ESP_OK ? "sent" : "FAILED");
 }
 
 /* ---- Public API ---- */

firmware/esp32-csi-node/main/csi_collector.c

@@ -341,7 +341,9 @@ static void wifi_csi_callback(void *ctx, wifi_csi_info_t *info)
             memcpy(&sync[24], &s_sequence, 4);    /* high-water seq for pairing */
             uint32_t zero32 = 0;
             memcpy(&sync[28], &zero32, 4);        /* reserved (room for leader_id low32) */
-            int sr = stream_sender_send(sync, sizeof(sync));
+            /* Sync packets are 32 B at ~0.5 Hz — priority path so the CSI
+             * ENOMEM backoff can't starve cross-node time alignment (#1183). */
+            int sr = stream_sender_send_priority(sync, sizeof(sync));
             static uint32_t s_sync_count = 0;
             s_sync_count++;
             if (s_sync_count <= 3 || (s_sync_count % 60) == 0) {

firmware/esp32-csi-node/main/main.c

@@ -144,6 +144,54 @@ static void wifi_init_sta(void)
     }
 }
 
+#if CONFIG_LED_GAMMA_VIZ
+/* Viridis colormap (60 steps), generated from ruv-neural-viz::ColorMap::viridis()
+ * — the rUv-Neural brain-topology colormap, now no_std (ruvnet/ruv-neural#3 /
+ * RuView#1126). Used as the ON-phase colour of the 40 Hz gamma flicker below:
+ * dark-purple (still) -> teal -> green -> yellow (strong motion). */
+static const uint8_t VIRIDIS_LUT[60][3] = {
+    { 68,  1, 84},{ 67,  6, 88},{ 67, 12, 91},{ 66, 17, 95},{ 66, 23, 99},
+    { 65, 28,103},{ 64, 34,106},{ 64, 39,110},{ 63, 45,114},{ 63, 50,118},
+    { 62, 56,121},{ 61, 61,125},{ 61, 67,129},{ 60, 72,132},{ 59, 78,136},
+    { 59, 83,139},{ 57, 87,139},{ 55, 92,139},{ 53, 96,139},{ 52,100,139},
+    { 50,104,139},{ 48,109,139},{ 46,113,139},{ 44,117,140},{ 43,122,140},
+    { 41,126,140},{ 39,130,140},{ 37,134,140},{ 36,139,140},{ 34,143,140},
+    { 35,147,139},{ 39,151,136},{ 43,154,133},{ 47,158,130},{ 52,162,127},
+    { 56,166,124},{ 60,170,121},{ 64,173,119},{ 68,177,116},{ 72,181,113},
+    { 76,185,110},{ 81,189,107},{ 85,192,104},{ 89,196,102},{ 93,200, 99},
+    {102,203, 95},{113,205, 91},{124,207, 87},{134,209, 82},{145,211, 78},
+    {156,213, 74},{167,215, 70},{178,217, 66},{188,219, 62},{199,221, 58},
+    {210,223, 54},{221,225, 49},{231,227, 45},{242,229, 41},{253,231, 37},
+};
+static led_strip_handle_t s_viz_led;
+
+/* motion_energy that saturates the colormap to yellow (CONFIG, milli-units). */
+#define LED_MOTION_FULLSCALE ((float)CONFIG_LED_MOTION_FULLSCALE_MILLI / 1000.0f)
+
+/* GENUS-style 40 Hz gamma flicker: full on/off square wave, 50% duty (toggled
+ * every 12.5 ms → 40 Hz). The ON colour is live CSI motion (edge motion_energy)
+ * mapped through the ruv-neural-viz viridis LUT — still=purple, moving=yellow.
+ * So the LED is a real 40 Hz gamma stimulus whose hue tracks sensed motion. */
+static void led_gamma_40hz_cb(void *arg)
+{
+    static bool on = false;
+    on = !on;
+    if (on) {
+        edge_vitals_pkt_t v;
+        float m = edge_get_vitals(&v) ? v.motion_energy : 0.0f;
+        float norm = m / LED_MOTION_FULLSCALE;
+        if (norm < 0.0f) norm = 0.0f;
+        if (norm > 1.0f) norm = 1.0f;
+        int idx = (int)(norm * 59.0f + 0.5f);
+        const uint8_t *c = VIRIDIS_LUT[idx];
+        led_strip_set_pixel(s_viz_led, 0, c[0], c[1], c[2]); /* R,G,B (driver maps to GRB) */
+    } else {
+        led_strip_set_pixel(s_viz_led, 0, 0, 0, 0);          /* off phase */
+    }
+    led_strip_refresh(s_viz_led);
+}
+#endif /* CONFIG_LED_GAMMA_VIZ */
+
 void app_main(void)
 {
     /* Initialize NVS */
@@ -173,15 +221,16 @@ void app_main(void)
     ESP_LOGI(TAG, "%s CSI Node (ADR-018 / ADR-110) — v%s — Node ID: %d",
              target_name, app_desc->version, g_nvs_config.node_id);
 
-    /* Turn off onboard WS2812 LED.
-     * S3 dev boards put the LED on GPIO 38; C6 dev boards on GPIO 8.
-     * On C6, GPIO 38 doesn't exist (only 0-30) — gate the init by target. */
+    /* Onboard WS2812. C6 wires the LED to GPIO 8; S3 to GPIO 38 (DevKitC-1 v1.0)
+     * or GPIO 48 (DevKitC-1 v1.1 / N16R8 — see #962). On S3 we drive 48 (the
+     * common module). On C6, GPIO 38/48 don't exist (only 0-30) — gate by target.
+     * Behaviour is set by CONFIG_LED_GAMMA_VIZ (ADR-183): on = 40 Hz gamma flicker
+     * coloured by CSI motion; off = clear the LED at boot. */
 #if defined(CONFIG_IDF_TARGET_ESP32C6)
     const int led_gpio = 8;
 #else
-    const int led_gpio = 38;
+    const int led_gpio = 48;
 #endif
-    led_strip_handle_t led_strip;
     led_strip_config_t strip_config = {
         .strip_gpio_num = led_gpio,
         .max_leds = 1,
@@ -193,9 +242,26 @@ void app_main(void)
         .resolution_hz = 10 * 1000 * 1000, // 10MHz
         .flags.with_dma = false,
     };
+#if CONFIG_LED_GAMMA_VIZ
+    if (led_strip_new_rmt_device(&strip_config, &rmt_config, &s_viz_led) == ESP_OK) {
+        const esp_timer_create_args_t viz_args = {
+            .callback = &led_gamma_40hz_cb,
+            .name = "led_gamma_40hz",
+        };
+        esp_timer_handle_t viz_timer;
+        if (esp_timer_create(&viz_args, &viz_timer) == ESP_OK) {
+            esp_timer_start_periodic(viz_timer, 12500); // 12.5 ms toggle → 40 Hz square wave
+            ESP_LOGI(TAG, "Onboard WS2812: 40 Hz gamma flicker (GENUS), colour=CSI motion via ruv-neural-viz, GPIO %d", led_gpio);
+        }
+    }
+#else
+    /* Viz disabled — clear the onboard LED at boot and release the RMT channel. */
+    led_strip_handle_t led_strip;
     if (led_strip_new_rmt_device(&strip_config, &rmt_config, &led_strip) == ESP_OK) {
         led_strip_clear(led_strip);
+        led_strip_del(led_strip);
     }
+#endif /* CONFIG_LED_GAMMA_VIZ */
 
     /* ADR-110 P4: 802.15.4 mesh time-sync (C6 only).
      * Initialized BEFORE WiFi so it's available even when WiFi STA can't

firmware/esp32-csi-node/main/mmwave_detect.h

@@ -0,0 +1,37 @@
+/**
+ * @file mmwave_detect.h
+ * @brief Pure (host-testable) mmWave frame-validation predicates for probe-time
+ *        sensor detection. No ESP-IDF deps — safe to #include in a host unit test.
+ *
+ * Detection must validate a *full* frame, never a bare header byte/pattern: a
+ * floating UART with no sensor reads line noise that can contain header-looking
+ * bytes, which the old loose checks mistook for a real sensor (#1107 MR60,
+ * #1135 LD2410). These predicates are the validate-before-trust gate.
+ */
+#ifndef MMWAVE_DETECT_H
+#define MMWAVE_DETECT_H
+
+#include <stdint.h>
+#include <stdbool.h>
+
+/**
+ * True iff buf[i..] begins a *validated* LD2410 report frame within [0,len):
+ *   F4 F3 F2 F1 | len(LE,2) | data[len] | F8 F7 F6 F5
+ * Requires the head magic, a sane intra-frame length, AND the matching tail at
+ * head+6+len. Pure noise that merely contains 0xF4F3F2F1 fails the tail check.
+ */
+static inline bool mmwave_ld2410_valid_at(const uint8_t *buf, int i, int len)
+{
+    if (i < 0 || i + 5 >= len) return false;
+    if (!(buf[i] == 0xF4 && buf[i+1] == 0xF3 && buf[i+2] == 0xF2 && buf[i+3] == 0xF1))
+        return false;
+    uint16_t flen = (uint16_t)buf[i+4] | ((uint16_t)buf[i+5] << 8);
+    /* Real LD2410 report frames are small (basic=13, engineering=35). */
+    if (flen < 1 || flen > 64) return false;
+    int tail = i + 6 + (int)flen;
+    if (tail + 3 >= len) return false;
+    return buf[tail] == 0xF8 && buf[tail+1] == 0xF7
+        && buf[tail+2] == 0xF6 && buf[tail+3] == 0xF5;
+}
+
+#endif /* MMWAVE_DETECT_H */

firmware/esp32-csi-node/main/mmwave_sensor.c

@@ -26,6 +26,7 @@
  */
 
 #include "mmwave_sensor.h"
+#include "mmwave_detect.h"
 
 #include <string.h>
 #include <math.h>
@@ -401,10 +402,12 @@ static mmwave_type_t probe_at_baud(uint32_t baud)
                     }
                 }
             }
-            /* LD2410: 4-byte header 0xF4F3F2F1 (already specific enough). */
-            if (i + 3 < len && buf[i] == 0xF4 && buf[i+1] == 0xF3
-                && buf[i+2] == 0xF2 && buf[i+3] == 0xF1
-                && baud == MMWAVE_LD2410_BAUD) {
+            /* LD2410: require a *full validated* report frame, not just the
+             * 4-byte head. A floating UART1 at 256000 baud can emit the head
+             * pattern 0xF4F3F2F1 from line noise (#1135 bug #2). The shared
+             * predicate (host-unit-tested in mmwave_detect.h) demands a sane
+             * intra-frame length AND the matching tail 0xF8F7F6F5. */
+            if (baud == MMWAVE_LD2410_BAUD && mmwave_ld2410_valid_at(buf, i, len)) {
                 ld2410_header_seen++;
             }
         }

firmware/esp32-csi-node/main/rv_mesh.c

@@ -188,7 +188,9 @@ size_t rv_mesh_encode_calibration_start(uint8_t sender_role,
 esp_err_t rv_mesh_send(const uint8_t *frame, size_t len)
 {
     if (frame == NULL || len == 0) return ESP_ERR_INVALID_ARG;
-    int sent = stream_sender_send(frame, len);
+    /* Mesh control packets (HEALTH, anomaly) are low-rate and tiny — send them
+     * on the priority path so the CSI ENOMEM backoff can't starve them (#1183). */
+    int sent = stream_sender_send_priority(frame, len);
     if (sent < 0) {
         ESP_LOGW(TAG, "rv_mesh_send: stream_sender failed (len=%u)",
                  (unsigned)len);

firmware/esp32-csi-node/main/stream_sender.c

@@ -26,9 +26,16 @@ static struct sockaddr_in s_dest_addr;
  * rapid-fire CSI callbacks can exhaust the pbuf pool and crash the device.
  */
 static int64_t s_backoff_until_us = 0;       /* esp_timer timestamp to resume */
-#define ENOMEM_COOLDOWN_MS  100              /* suppress sends for 100 ms */
+#define ENOMEM_COOLDOWN_MS  100              /* base backoff; doubles per streak */
+#define ENOMEM_COOLDOWN_MAX_MS 2000          /* cap on the exponential backoff */
 #define ENOMEM_LOG_INTERVAL 50               /* log every Nth suppressed send */
 static uint32_t s_enomem_suppressed = 0;
+/* Consecutive ENOMEM episodes without an intervening successful send. A fixed
+ * 100 ms backoff is too short to drain sustained lwIP/WiFi buffer pressure
+ * (#1135 bug #1: tier-2 + concurrent TX keeps the node stuck), so the backoff
+ * grows 100→200→400→…→2000 ms per streak and resets on the first send that
+ * succeeds. */
+static uint32_t s_enomem_streak = 0;
 
 static int sender_init_internal(const char *ip, uint16_t port)
 {
@@ -93,16 +100,52 @@ int stream_sender_send(const uint8_t *data, size_t len)
                       (struct sockaddr *)&s_dest_addr, sizeof(s_dest_addr));
     if (sent < 0) {
         if (errno == ENOMEM) {
-            /* Start backoff to let lwIP reclaim buffers */
-            s_backoff_until_us = esp_timer_get_time() +
-                                 (int64_t)ENOMEM_COOLDOWN_MS * 1000;
-            ESP_LOGW(TAG, "sendto ENOMEM — backing off for %d ms", ENOMEM_COOLDOWN_MS);
+            /* Exponential backoff: double the cooldown each consecutive ENOMEM
+             * (capped) so sustained buffer pressure actually drains instead of
+             * the node re-failing every 100 ms forever (#1135 bug #1). */
+            uint32_t shift = s_enomem_streak < 5 ? s_enomem_streak : 5;
+            uint32_t cooldown = ENOMEM_COOLDOWN_MS << shift;
+            if (cooldown > ENOMEM_COOLDOWN_MAX_MS) cooldown = ENOMEM_COOLDOWN_MAX_MS;
+            s_enomem_streak++;
+            s_backoff_until_us = esp_timer_get_time() + (int64_t)cooldown * 1000;
+            ESP_LOGW(TAG, "sendto ENOMEM — backing off for %lu ms (streak %lu)",
+                     (unsigned long)cooldown, (unsigned long)s_enomem_streak);
         } else {
             ESP_LOGW(TAG, "sendto failed: errno %d", errno);
         }
         return -1;
     }
 
+    /* A send got through — buffer pressure cleared; reset the backoff streak. */
+    s_enomem_streak = 0;
+    return sent;
+}
+
+int stream_sender_send_priority(const uint8_t *data, size_t len)
+{
+    if (s_sock < 0) {
+        return -1;
+    }
+
+    /* Priority path (#1183): low-rate control packets (feature_state, HEALTH,
+     * mesh sync) bypass the global ENOMEM backoff gate so the high-rate CSI
+     * stream cannot starve them. These are ≤48 B at ≤1 Hz — negligible pbuf
+     * pressure, so they won't re-trigger the crash cascade that the backoff
+     * (driven by the 50 Hz CSI flood) exists to prevent.
+     *
+     * Crucially, an ENOMEM here is reported quietly and does NOT extend the
+     * global streak/backoff: a tiny control packet failing is a symptom of
+     * the bulk-stream pressure, not a cause, so it must not feed the cooldown
+     * that suppresses the next CSI frame. Likewise a success does not reset
+     * the streak — the bulk path owns that signal. */
+    int sent = sendto(s_sock, data, len, 0,
+                      (struct sockaddr *)&s_dest_addr, sizeof(s_dest_addr));
+    if (sent < 0) {
+        if (errno != ENOMEM) {
+            ESP_LOGW(TAG, "priority sendto failed: errno %d", errno);
+        }
+        return -1;
+    }
     return sent;
 }
 

firmware/esp32-csi-node/main/stream_sender.h

@@ -36,6 +36,20 @@ int stream_sender_init_with(const char *ip, uint16_t port);
  */
 int stream_sender_send(const uint8_t *data, size_t len);
 
+/**
+ * Send a low-rate control packet, bypassing the ENOMEM backoff gate (#1183).
+ *
+ * Intended for ≤48 B, ≤1 Hz control traffic (feature_state, HEALTH, mesh
+ * sync) that must not be starved by the global backoff the high-rate CSI
+ * stream triggers. An ENOMEM on this path is reported quietly and does NOT
+ * extend or reset the global backoff streak.
+ *
+ * @param data Frame data buffer.
+ * @param len  Length of data to send.
+ * @return Number of bytes sent, or -1 on error.
+ */
+int stream_sender_send_priority(const uint8_t *data, size_t len);
+
 /**
  * Close the UDP sender socket.
  */

firmware/esp32-csi-node/test/Makefile

@@ -44,9 +44,9 @@ FUZZ_DURATION ?= 30
 FUZZ_JOBS     ?= 1
 
 .PHONY: all clean run_serialize run_edge run_nvs run_all test_adr110 run_adr110 \
-        test_vitals run_vitals host_tests
+        test_vitals run_vitals test_mmwave_detect run_mmwave_detect host_tests
 
-all: fuzz_serialize fuzz_edge fuzz_nvs test_adr110 test_vitals
+all: fuzz_serialize fuzz_edge fuzz_nvs test_adr110 test_vitals test_mmwave_detect
 
 # --- ADR-110 encoding unit tests ---
 # Host-side, no libFuzzer needed — plain C99 deterministic table tests
@@ -69,8 +69,19 @@ test_vitals: test_vitals_count_presence.c $(MAIN_DIR)/edge_processing.h
 run_vitals: test_vitals
 	./test_vitals
 
-host_tests: run_adr110 run_vitals
-	@echo "Host tests passed (ADR-110 + vitals #998/#996)"
+# --- mmWave LD2410 detection predicate (#1135 bug #2) ---
+# Host-side, no libFuzzer. Proves a floating-UART head pattern (0xF4F3F2F1)
+# without a valid frame length+tail is REJECTED, so a phantom LD2410 is never
+# detected on a node with no sensor wired. Tests the real predicate the
+# firmware uses (../main/mmwave_detect.h) — test and firmware can't disagree.
+test_mmwave_detect: test_mmwave_detect.c $(MAIN_DIR)/mmwave_detect.h
+	cc -std=c99 -Wall -Wextra -I$(MAIN_DIR) -o $@ $<
+
+run_mmwave_detect: test_mmwave_detect
+	./test_mmwave_detect
+
+host_tests: run_adr110 run_vitals run_mmwave_detect
+	@echo "Host tests passed (ADR-110 + vitals #998/#996 + mmwave detect #1135)"
 
 # --- Serialize fuzzer ---
 # Tests csi_serialize_frame() with random wifi_csi_info_t inputs.

firmware/esp32-csi-node/test/test_mmwave_detect.c

@@ -0,0 +1,80 @@
+/**
+ * @file test_mmwave_detect.c
+ * @brief Host-side unit tests for the LD2410 frame-validation predicate (#1135).
+ *
+ * Proves the phantom-detection fix: a floating UART can emit the 4-byte head
+ * 0xF4F3F2F1, but the predicate rejects it unless a sane length + matching tail
+ * 0xF8F7F6F5 are also present. Tests the REAL predicate from mmwave_detect.h
+ * (the same code the firmware's probe_at_baud calls).
+ *
+ *   cc -std=c99 -Wall -I../main -o test_mmwave_detect test_mmwave_detect.c && ./test_mmwave_detect
+ *
+ * Exits 0 on all-pass; prints the failing case otherwise.
+ */
+#include <stdint.h>
+#include <stdio.h>
+#include <string.h>
+#include "mmwave_detect.h"
+
+static int failures = 0;
+#define CHECK(cond, msg) do { \
+    if (!(cond)) { printf("FAIL: %s\n", msg); failures++; } \
+    else        { printf("ok:   %s\n", msg); } \
+} while (0)
+
+/* Build a valid LD2410 report frame: F4F3F2F1 | len(LE) | data[len] | F8F7F6F5 */
+static int make_frame(uint8_t *out, uint16_t dlen)
+{
+    int n = 0;
+    out[n++] = 0xF4; out[n++] = 0xF3; out[n++] = 0xF2; out[n++] = 0xF1;
+    out[n++] = (uint8_t)(dlen & 0xFF); out[n++] = (uint8_t)(dlen >> 8);
+    for (uint16_t k = 0; k < dlen; k++) out[n++] = (uint8_t)(0xAA ^ k);
+    out[n++] = 0xF8; out[n++] = 0xF7; out[n++] = 0xF6; out[n++] = 0xF5;
+    return n;
+}
+
+int main(void)
+{
+    uint8_t buf[256];
+
+    /* 1. A real basic-report frame (data len 13) validates. */
+    int n = make_frame(buf, 13);
+    CHECK(mmwave_ld2410_valid_at(buf, 0, n), "valid basic frame (len=13) accepted");
+
+    /* 2. A real engineering-report frame (data len 35) validates. */
+    n = make_frame(buf, 35);
+    CHECK(mmwave_ld2410_valid_at(buf, 0, n), "valid engineering frame (len=35) accepted");
+
+    /* 3. Head magic present but NO valid tail — the #1135 phantom case. */
+    memset(buf, 0x00, sizeof(buf));
+    buf[0]=0xF4; buf[1]=0xF3; buf[2]=0xF2; buf[3]=0xF1; buf[4]=13; buf[5]=0;
+    /* data present but tail is zeros, not F8F7F6F5 */
+    CHECK(!mmwave_ld2410_valid_at(buf, 0, 64), "head magic without valid tail REJECTED (#1135)");
+
+    /* 4. Head magic with insane length is rejected. */
+    memset(buf, 0xFF, sizeof(buf));
+    buf[0]=0xF4; buf[1]=0xF3; buf[2]=0xF2; buf[3]=0xF1; buf[4]=0xFF; buf[5]=0xFF; /* len=65535 */
+    CHECK(!mmwave_ld2410_valid_at(buf, 0, 200), "head magic with oversized length REJECTED");
+
+    /* 5. Pure noise (no head) is rejected. */
+    for (int k = 0; k < 64; k++) buf[k] = (uint8_t)(0x5A + k);
+    CHECK(!mmwave_ld2410_valid_at(buf, 0, 64), "non-header noise REJECTED");
+
+    /* 6. Truncated frame (tail would run past the buffer) is rejected. */
+    n = make_frame(buf, 13);
+    CHECK(!mmwave_ld2410_valid_at(buf, 0, n - 2), "truncated frame (tail past buffer) REJECTED");
+
+    /* 7. Valid frame at a non-zero offset still validates. */
+    memset(buf, 0x00, sizeof(buf));
+    n = make_frame(buf + 7, 13);
+    CHECK(mmwave_ld2410_valid_at(buf, 7, 7 + n), "valid frame at offset 7 accepted");
+
+    /* 8. Repeated head bytes without a frame (worst-case noise) rejected. */
+    for (int k = 0; k + 3 < 64; k += 4) {
+        buf[k]=0xF4; buf[k+1]=0xF3; buf[k+2]=0xF2; buf[k+3]=0xF1;
+    }
+    CHECK(!mmwave_ld2410_valid_at(buf, 0, 64), "repeated bare head bytes REJECTED");
+
+    printf("\n%s (%d failures)\n", failures ? "FAILED" : "ALL PASS", failures);
+    return failures ? 1 : 0;
+}

scripts/firmware-release-guard.sh

@@ -0,0 +1,94 @@
+#!/usr/bin/env bash
+#
+# firmware-release-guard.sh — guard against shipping firmware built from a
+# stale generated `sdkconfig` (the v0.8.3-esp32 release bug).
+#
+# Symptom it catches: an incremental build reuses a leftover `sdkconfig`
+# instead of `sdkconfig.defaults`, so an "8MB" build silently links the 4MB
+# dual-OTA partition layout (no spiffs, ota_1 @ 0x1F0000) and the released
+# `partition-table.bin` does not match the flash-size variant it claims to be.
+#
+# What it does: for the named flash-size variant, regenerate the EXPECTED
+# partition table from the partition CSV that variant must use, and byte-compare
+# it against the freshly built `partition-table.bin`. Also cross-checks the
+# flash size recorded in the build's `flasher_args.json`. Exits non-zero on any
+# mismatch so a release pipeline fails closed.
+#
+# Usage:
+#   scripts/firmware-release-guard.sh <8mb|4mb> <build-dir>
+#
+# Example:
+#   scripts/firmware-release-guard.sh 8mb firmware/esp32-csi-node/build
+#
+set -euo pipefail
+
+VARIANT="${1:-}"
+BUILD_DIR="${2:-}"
+
+if [[ -z "$VARIANT" || -z "$BUILD_DIR" ]]; then
+  echo "usage: $0 <8mb|4mb> <build-dir>" >&2
+  exit 2
+fi
+
+# Firmware project root (this script lives in <repo>/scripts).
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+FW_DIR="$SCRIPT_DIR/../firmware/esp32-csi-node"
+
+case "$VARIANT" in
+  8mb) EXPECT_CSV="partitions_display.csv"; EXPECT_FLASH="8MB" ;;
+  4mb) EXPECT_CSV="partitions_4mb.csv";     EXPECT_FLASH="4MB" ;;
+  *)   echo "ERROR: unknown variant '$VARIANT' (want 8mb|4mb)" >&2; exit 2 ;;
+esac
+
+BUILT_PT="$BUILD_DIR/partition_table/partition-table.bin"
+CSV_PATH="$FW_DIR/$EXPECT_CSV"
+
+[[ -f "$BUILT_PT" ]]  || { echo "ERROR: built partition table not found: $BUILT_PT" >&2; exit 1; }
+[[ -f "$CSV_PATH" ]]  || { echo "ERROR: expected CSV not found: $CSV_PATH" >&2; exit 1; }
+
+# Locate the ESP-IDF partition table generator.
+GEN="${IDF_PATH:-}/components/partition_table/gen_esp32part.py"
+if [[ ! -f "$GEN" ]]; then
+  GEN="C:/Users/ruv/esp/v5.4/esp-idf/components/partition_table/gen_esp32part.py"
+fi
+[[ -f "$GEN" ]] || { echo "ERROR: gen_esp32part.py not found (set IDF_PATH)" >&2; exit 1; }
+
+PY="${PYTHON:-python}"
+command -v "$PY" >/dev/null 2>&1 || PY="C:/Espressif/tools/python/v5.4/venv/Scripts/python.exe"
+
+TMP="$(mktemp -d)"
+trap 'rm -rf "$TMP"' EXIT
+EXPECT_PT="$TMP/expected-partition-table.bin"
+
+# Regenerate the expected table from the CSV this variant must use.
+"$PY" "$GEN" --quiet "$CSV_PATH" "$EXPECT_PT"
+
+fail=0
+
+if ! cmp -s "$EXPECT_PT" "$BUILT_PT"; then
+  echo "FAIL: built partition table does not match $EXPECT_CSV for the $VARIANT variant." >&2
+  echo "      The build likely reused a stale sdkconfig. Decoded built table:" >&2
+  "$PY" "$GEN" "$BUILT_PT" 2>/dev/null | grep -vE '^#|^Parsing|^Verifying' | sed 's/^/        /' >&2
+  fail=1
+fi
+
+# Cross-check the flash size the build actually targeted.
+FA="$BUILD_DIR/flasher_args.json"
+if [[ -f "$FA" ]]; then
+  GOT_FLASH="$("$PY" - "$FA" <<'PYEOF'
+import json,sys
+with open(sys.argv[1]) as f: d=json.load(f)
+print(d.get("flash_settings",{}).get("flash_size",""))
+PYEOF
+)"
+  if [[ "$GOT_FLASH" != "$EXPECT_FLASH" ]]; then
+    echo "FAIL: flasher_args.json flash_size='$GOT_FLASH', expected '$EXPECT_FLASH'." >&2
+    fail=1
+  fi
+fi
+
+if [[ "$fail" -ne 0 ]]; then
+  exit 1
+fi
+
+echo "OK: $VARIANT firmware build matches $EXPECT_CSV (flash_size=$EXPECT_FLASH)."

v2/crates/wifi-densepose-sensing-server/src/engine_bridge.rs

@@ -402,17 +402,36 @@ mod tests {
         assert!(!bridge.suppress_raw_outputs());
     }
 
-    /// Error wiring (review finding 1a): two live nodes with mismatched
-    /// subcarrier counts make fusion return a `DimensionMismatch` →
-    /// `EngineError` — previously dropped by `if let Some(Ok(..))` at the
+    /// Error wiring (review finding 1a): a live cycle that fails fusion yields
+    /// an `EngineError` — previously dropped by `if let Some(Ok(..))` at the
     /// call sites. The counter must increment and the last good trust state
     /// must survive a later failure.
+    ///
+    /// Originally this forced the failure with a 56-vs-30 subcarrier mismatch
+    /// (`DimensionMismatch`). Since #1170 the live bridge canonicalizes every
+    /// node onto the 56-tone grid, so heterogeneous counts now fuse cleanly —
+    /// a frame-timestamp spread wider than the fuser's 60 ms guard interval is
+    /// the remaining deterministic way to provoke a fusion error here.
     #[test]
     fn observe_cycle_counts_engine_errors() {
+        // Both nodes are 56-subcarrier (canonicalization-clean), but their
+        // frame timestamps are 500 ms apart — far beyond the 60 ms guard —
+        // so the fuser rejects the cycle with TimestampMismatch. Future
+        // offsets keep both instants safely after the bridge's lazy EPOCH.
+        fn mismatched_states() -> HashMap<u8, NodeState> {
+            let now = Instant::now();
+            let mut a = node_state_with_history(1.0, 56);
+            a.last_frame_time = Some(now + std::time::Duration::from_millis(600));
+            let mut b = node_state_with_history(1.05, 56);
+            b.last_frame_time = Some(now + std::time::Duration::from_millis(100));
+            let mut m = HashMap::new();
+            m.insert(0u8, a);
+            m.insert(1u8, b);
+            m
+        }
+
         let mut bridge = EngineBridge::new(PrivacyMode::PrivateHome, 1, "r", "R");
-        let mut mismatched = HashMap::new();
-        mismatched.insert(0u8, node_state_with_history(1.0, 56));
-        mismatched.insert(1u8, node_state_with_history(1.05, 30)); // 30 ≠ 56 subcarriers
+        let mismatched = mismatched_states();
 
         assert!(bridge.observe_cycle(&mismatched, 1_000).is_none());
         assert_eq!(bridge.engine_error_count(), 1);

v2/crates/wifi-densepose-sensing-server/src/main.rs

@@ -518,17 +518,31 @@ const NOVELTY_HISTORY_CAPACITY: usize = 64;
 /// subcarrier ordering / normalisation so banks reject stale data.
 const NOVELTY_SKETCH_VERSION: u16 = 1;
 
+/// Lower plausibility floor (seconds) for a CSI inter-frame delta.
+///
+/// The firmware caps CSI sends at `CSI_MIN_SEND_INTERVAL_US = 20 ms`
+/// (`csi_collector.c`), so a single node cannot physically produce frames
+/// faster than 50 fps. UDP/OS buffering, however, delivers frames in tight
+/// bursts whose intra-burst arrival deltas are tens of microseconds apart —
+/// a 36 µs delta yields `1/dt ≈ 27 kHz`, which the old `< 1 s` guard let
+/// straight into the EMA and inflated `csi_fps_ema` by 1–3 orders of
+/// magnitude (issue #1180). We reject any delta implying more than 200 fps
+/// (4× the physical ceiling, leaving slack for benign arrival jitter); such
+/// deltas are burst artifacts, not distinct production intervals.
+pub(crate) const MIN_PLAUSIBLE_CSI_DT_SEC: f64 = 0.005;
+
 /// ADR-110 iter 18 — EMA update for per-node CSI fps tracking.
 ///
 /// Returns the new EMA value, or `None` if the delta is implausible
-/// (≤ 0, or > 1 second — likely a connection gap, not a real frame
-/// rate sample). α = 1/8 fixed shift, ~8-sample effective window,
-/// matching the firmware-side ESP-NOW offset smoother in §A0.10.
+/// (below [`MIN_PLAUSIBLE_CSI_DT_SEC`] — a sub-ms burst artifact, see
+/// issue #1180 — or `> 1 second`, likely a connection gap rather than a
+/// real frame-rate sample). α = 1/8 fixed shift, ~8-sample effective
+/// window, matching the firmware-side ESP-NOW offset smoother in §A0.10.
 ///
 /// Free function for testability — every transformation that doesn't
 /// touch the rest of `NodeState` lives outside the `impl` block.
 pub(crate) fn update_csi_fps_ema(prev_fps: f64, dt_sec: f64) -> Option<f64> {
-    if !(dt_sec > 0.0 && dt_sec < 1.0) {
+    if !(dt_sec >= MIN_PLAUSIBLE_CSI_DT_SEC && dt_sec < 1.0) {
         return None;
     }
     let instantaneous = 1.0 / dt_sec;
@@ -569,6 +583,35 @@ mod fps_ema_tests {
     fn long_gap_rejected_as_implausible() {
         assert!(update_csi_fps_ema(20.0, 2.0).is_none());
     }
+
+    #[test]
+    fn subms_burst_delta_rejected() {
+        // Issue #1180: a 36 µs intra-burst delta implies ~27 kHz and must
+        // not enter the EMA. Anything below the 5 ms floor is rejected.
+        assert!(update_csi_fps_ema(40.0, 0.000_036).is_none());
+        assert!(update_csi_fps_ema(40.0, 0.001).is_none());
+        // Just above the floor is accepted.
+        assert!(update_csi_fps_ema(40.0, 0.005).is_some());
+    }
+
+    #[test]
+    fn burst_interleaved_with_nominal_stays_in_band() {
+        // A true ~40 fps node whose frames arrive in sub-ms bursts: feeding
+        // only the plausible (nominal-cadence) deltas keeps the EMA near the
+        // ground truth instead of blowing up. Burst deltas are rejected by
+        // the caller (see NodeState::observe_csi_frame_arrival), so the EMA
+        // only ever sees the ~25 ms inter-group gaps.
+        let mut fps = 40.0;
+        for _ in 0..40 {
+            // nominal 25 ms gap (40 fps); intervening sub-ms bursts skipped
+            fps = update_csi_fps_ema(fps, 0.025).unwrap();
+            assert!(update_csi_fps_ema(fps, 0.000_040).is_none());
+        }
+        assert!(
+            (fps - 40.0).abs() < 1.0,
+            "EMA should stay within ~1 Hz of the 40 fps ground truth, got {fps}"
+        );
+    }
 }
 
 impl NodeState {
@@ -653,6 +696,15 @@ impl NodeState {
     pub(crate) fn observe_csi_frame_arrival(&mut self, now: std::time::Instant) {
         if let Some(prev) = self.last_frame_time {
             let dt = now.duration_since(prev).as_secs_f64();
+            // Burst arrivals (sub-floor dt, issue #1180): do NOT re-anchor on
+            // them. Keeping the previous anchor means the next genuine
+            // inter-frame gap measures the true cadence across the whole
+            // burst instead of intra-burst jitter — so a 50 fps node whose
+            // frames arrive in 36 µs bursts every 25 ms still reads ~40 fps,
+            // not 27 kHz.
+            if dt < MIN_PLAUSIBLE_CSI_DT_SEC {
+                return;
+            }
             if let Some(new_ema) = update_csi_fps_ema(self.csi_fps_ema, dt) {
                 self.csi_fps_ema = new_ema;
                 self.csi_fps_samples = self.csi_fps_samples.saturating_add(1);
@@ -8037,6 +8089,36 @@ mod sync_snapshot_helper_tests {
         assert_eq!(snap.csi_fps_samples, 42);
     }
 
+    #[test]
+    fn observe_csi_frame_arrival_ignores_subms_bursts() {
+        // Issue #1180 regression: a ~40 fps node whose frames are delivered
+        // in tight UDP bursts (sub-ms intra-burst deltas) must still report
+        // ~40 fps, not tens of kHz. Synthesize the arrival stream by adding
+        // Durations to a base Instant.
+        use std::time::Duration;
+        let base = std::time::Instant::now();
+        let mut ns = NodeState::new();
+        ns.csi_fps_ema = 40.0; // pretend already warmed up
+        ns.csi_fps_samples = 10;
+
+        // 30 nominal 25 ms groups, each preceded by a 3-frame sub-ms burst.
+        for g in 0..30u64 {
+            let group_t = base + Duration::from_millis(25 * g);
+            ns.observe_csi_frame_arrival(group_t);
+            // burst: two extra arrivals 40 µs and 80 µs later — must be
+            // ignored for rate purposes (anchor must not advance to them).
+            ns.observe_csi_frame_arrival(group_t + Duration::from_micros(40));
+            ns.observe_csi_frame_arrival(group_t + Duration::from_micros(80));
+        }
+
+        assert!(
+            (ns.csi_fps_ema - 40.0).abs() < 2.0,
+            "csi_fps_ema must stay near the 40 fps ground truth despite \
+             sub-ms bursts, got {}",
+            ns.csi_fps_ema
+        );
+    }
+
     #[test]
     fn apply_sync_packet_populates_a_fresh_node() {
         // Mirrors what udp_receiver_task does on the very first sync

v2/crates/wifi-densepose-sensing-server/src/multistatic_bridge.rs

@@ -10,7 +10,7 @@ use std::collections::HashMap;
 use std::sync::LazyLock;
 use std::time::{Duration, Instant};
 
-use wifi_densepose_signal::hardware_norm::{CanonicalCsiFrame, HardwareType};
+use wifi_densepose_signal::hardware_norm::{CanonicalCsiFrame, HardwareNormalizer, HardwareType};
 use wifi_densepose_signal::ruvsense::multiband::MultiBandCsiFrame;
 use wifi_densepose_signal::ruvsense::multistatic::{FusedSensingFrame, MultistaticFuser};
 
@@ -26,6 +26,11 @@ const DEFAULT_FREQ_MHZ: u32 = 2437; // Channel 6
 /// are relative to this instant, avoiding wall-clock/monotonic mixing issues.
 static EPOCH: LazyLock<Instant> = LazyLock::new(Instant::now);
 
+/// Shared length-only canonicalizer (issue #1170). The default 56-tone grid
+/// matches what `MultistaticFuser` (ADR-154) expects. Stateless and immutable,
+/// so a single process-wide instance is safe to share across nodes.
+static NORMALIZER: LazyLock<HardwareNormalizer> = LazyLock::new(HardwareNormalizer::new);
+
 /// Convert a single `NodeState` into a `MultiBandCsiFrame` suitable for
 /// multistatic fusion.
 ///
@@ -38,7 +43,14 @@ pub fn node_frame_from_state(node_id: u8, ns: &NodeState) -> Option<MultiBandCsi
         return None;
     }
 
-    let amplitude: Vec<f32> = latest.iter().map(|&v| v as f32).collect();
+    // Issue #1170: resample the raw amplitude onto the canonical 56-tone grid
+    // BEFORE fusion. ESP32 nodes in mixed HT20/HT40 capture modes report
+    // different subcarrier counts (64 / 128 / 192); feeding those raw into
+    // `MultistaticFuser::fuse` tripped `DimensionMismatch` on every cycle and
+    // silently disabled real multistatic fusion. Length-only canonicalization
+    // (no z-score) keeps the amplitude scale the person-score relies on.
+    let canonical_amp = NORMALIZER.resample_to_canonical(latest);
+    let amplitude: Vec<f32> = canonical_amp.iter().map(|&v| v as f32).collect();
     let n_sub = amplitude.len();
     let phase = vec![0.0_f32; n_sub];
 
@@ -201,15 +213,58 @@ mod tests {
         assert_eq!(frame.channel_frames.len(), 1);
 
         let ch = &frame.channel_frames[0];
-        assert_eq!(ch.amplitude.len(), 3);
-        assert!((ch.amplitude[0] - 10.0_f32).abs() < f32::EPSILON);
-        assert!((ch.amplitude[1] - 20.0_f32).abs() < f32::EPSILON);
-        assert!((ch.amplitude[2] - 30.5_f32).abs() < f32::EPSILON);
+        // Issue #1170: amplitude is now resampled onto the canonical 56-tone
+        // grid regardless of the raw count.
+        assert_eq!(ch.amplitude.len(), 56);
+        // resample_cubic preserves the endpoints (no z-scoring), so the scale
+        // the person-score relies on is intact.
+        assert!((ch.amplitude[0] - 10.0_f32).abs() < 1e-3);
+        assert!((ch.amplitude[55] - 30.5_f32).abs() < 1e-3);
         // Phase should be all zeros
         assert!(ch.phase.iter().all(|&p| p == 0.0));
         assert_eq!(ch.hardware_type, HardwareType::Esp32S3);
     }
 
+    #[test]
+    fn heterogeneous_node_counts_canonicalize_and_fuse() {
+        // Issue #1170 regression: a mixed mesh with HT20 (64-bin) and HT40
+        // (192-bin) nodes must canonicalize to a uniform 56 tones and fuse,
+        // instead of tripping DimensionMismatch on every cycle.
+        let mut states: HashMap<u8, NodeState> = HashMap::new();
+
+        let mut h64 = VecDeque::new();
+        h64.push_back((0..64).map(|i| 1.0 + 0.1 * i as f64).collect::<Vec<f64>>());
+        states.insert(1, make_node_state(h64, Some(Instant::now()), 1));
+
+        let mut h192 = VecDeque::new();
+        h192.push_back((0..192).map(|i| 2.0 + 0.05 * i as f64).collect::<Vec<f64>>());
+        states.insert(3, make_node_state(h192, Some(Instant::now()), 1));
+
+        let frames = node_frames_from_states(&states);
+        assert_eq!(frames.len(), 2, "both nodes should produce frames");
+        for f in &frames {
+            assert_eq!(
+                f.channel_frames[0].amplitude.len(),
+                56,
+                "every node must present the canonical 56-tone dimension"
+            );
+        }
+
+        // The fuser must now accept the cycle (no DimensionMismatch).
+        let fuser = MultistaticFuser::new();
+        let result = fuser.fuse(&frames);
+        assert!(
+            result.is_ok(),
+            "heterogeneous mesh should fuse after canonicalization, got {result:?}"
+        );
+
+        // And the higher-level fallback path returns the fused frame, not the
+        // sum/dedup fallback.
+        let (fused, fallback) = fuse_or_fallback(&fuser, &states, 3.0);
+        assert!(fused.is_some(), "fusion should succeed");
+        assert!(fallback.is_none(), "no fallback when fusion succeeds");
+    }
+
     #[test]
     fn test_stale_node_excluded() {
         let mut states: HashMap<u8, NodeState> = HashMap::new();

v2/crates/wifi-densepose-signal/src/hardware_norm.rs

@@ -167,6 +167,22 @@ impl HardwareNormalizer {
             hardware_type: hw,
         })
     }
+
+    /// Resample a raw 1-D CSI vector onto the canonical subcarrier grid
+    /// **without** z-score normalization (length-only canonicalization).
+    ///
+    /// Used by the live multistatic bridge (issue #1170): heterogeneous
+    /// ESP32 capture modes report different subcarrier counts (HT20 ≈ 64,
+    /// HT40 ≈ 128/192), and [`MultistaticFuser`] requires every node frame
+    /// to share one dimension. Full [`Self::normalize`] would z-score the
+    /// amplitude (mean → 0), which saturates the downstream person-score
+    /// (a squared coefficient of variation `variance / mean²`); resampling
+    /// alone makes frames fusable while preserving amplitude scale.
+    ///
+    /// [`MultistaticFuser`]: crate::ruvsense::multistatic::MultistaticFuser
+    pub fn resample_to_canonical(&self, raw: &[f64]) -> Vec<f64> {
+        resample_cubic(raw, self.canonical_subcarriers)
+    }
 }
 
 impl Default for HardwareNormalizer {
@@ -344,6 +360,25 @@ mod tests {
         }
     }
 
+    #[test]
+    fn resample_to_canonical_is_length_only_no_zscore() {
+        // Issue #1170: resample_to_canonical must change length to 56 but
+        // NOT z-score (mean must be preserved, not driven to ~0). A raw
+        // amplitude vector with a large positive mean keeps that mean.
+        let norm = HardwareNormalizer::new();
+        let raw: Vec<f64> = (0..192).map(|i| 50.0 + 0.1 * i as f64).collect();
+        let out = norm.resample_to_canonical(&raw);
+        assert_eq!(out.len(), 56, "must resample onto the 56-tone grid");
+        let mean = out.iter().sum::<f64>() / out.len() as f64;
+        assert!(
+            mean > 40.0,
+            "resample-only must preserve amplitude scale (mean ~60), got {mean}"
+        );
+        // Endpoints preserved.
+        assert!((out[0] - raw[0]).abs() < 1e-6);
+        assert!((out[55] - raw[191]).abs() < 0.5);
+    }
+
     #[test]
     fn zscore_produces_zero_mean_unit_std() {
         let data: Vec<f64> = (0..100)