get rid of housekeeping tick and AGC reset

zephcore/Kconfig

@@ -370,12 +370,68 @@ endif # ZEPHCORE_ROLE_COMPANION
 
 config ZEPHCORE_HOUSEKEEPING_INTERVAL_MS
 	int "Periodic housekeeping interval (ms)"
+	depends on ZEPHCORE_ROLE_COMPANION || ZEPHCORE_ROLE_ROOM_SERVER
 	default 5000
 	range 1000 60000
 	help
 	  Wake interval for noise floor calibration, battery reads, and UI refresh.
 	  Longer = lower power, less responsive monitoring.
 
+	  Companion and room server only.  Both still run a fixed periodic tick:
+	  the companion's block contains genuine pollers (contact-dump progress,
+	  BLE advertising watchdog) that have no deadline to arm, and the room
+	  server is paced by its own 500 ms push timer regardless.  Repeaters are
+	  deadline-driven instead — see ZEPHCORE_MAINTENANCE_BACKSTOP_MS.
+
+config ZEPHCORE_MAINTENANCE_BACKSTOP_MS
+	int "Repeater maintenance backstop (ms)"
+	depends on ZEPHCORE_ROLE_REPEATER
+	default 60000
+	range 15000 600000
+	help
+	  Hard ceiling on how long a repeater will go without a maintenance pass.
+
+	  main_repeater.cpp arms a one-shot wake at the soonest deadline reported
+	  by msUntilNextMaintenance(), then clamps it to this value.  The clamp is
+	  UNCONDITIONAL, so this is a safety net against a deadline that is missed
+	  or mis-reported — not a value that should ever bind in normal operation.
+
+	  MUST stay above ZEPHCORE_NOISE_FLOOR_INTERVAL_MS and the CAD probe
+	  interval (prefs cad_probe_interval, default 15 s), which are the shortest
+	  recurring deadlines on an idle repeater.  Set at or below them and this
+	  silently becomes the wake period, reproducing the old fixed 5 s tick and
+	  making the deadline scheduling a no-op.  That is exactly the bug the
+	  original 5000 default shipped with.
+
+	  Lowering it costs wakes without improving responsiveness — real work is
+	  already scheduled at its own deadline.  Raising it widens the window in
+	  which a scheduling bug goes unnoticed.
+
+config ZEPHCORE_NOISE_FLOOR_INTERVAL_MS
+	int "Noise floor sampling interval (ms)"
+	default 5000 if ZEPHCORE_ROLE_COMPANION || ZEPHCORE_ROLE_ROOM_SERVER
+	default 15000
+	range 1000 120000
+	help
+	  How often the radio samples ambient RSSI to update the noise floor EMA.
+
+	  This used to be implicit: the sampler ran once per housekeeping tick, so
+	  it inherited that 5 s period.  The sampler lives in shared radio code, so
+	  roles that still tick periodically (companion, room server) keep 5000 —
+	  raising it there would change their RF tracking for no power gain, since
+	  their housekeeping timer wakes them at 5 s regardless.
+
+	  Repeaters use 15 s, matching ZEPHCORE_CAD_PROBE_INTERVAL — the only other
+	  recurring radio deadline — so the two do not interleave into separate
+	  wakes any more often than they have to.
+
+	  On a repeater this is one of the two shortest recurring deadlines, so it
+	  bounds how long the SoC can sleep between wakes.  Raising it stretches
+	  EMA warmup (8 samples) and the every-16th-sample unguarded bypass by the
+	  same factor: at 15000 warmup is ~2 min and the bypass ~4 min apart; at
+	  30000, ~4 min and ~8 min.  Past 15000 the CAD probe becomes the binding
+	  deadline instead, so raise both or neither.
+
 menu "Radio Configuration"
 
 choice ZEPHCORE_RADIO_TYPE

zephcore/Repeater_CLI_commands.md

@@ -198,7 +198,7 @@ All `set uplink.*` changes are saved immediately and only applied after reboot.
 | `get guest.password` | Guest access password |
 | `get owner.info` | Owner/contact info (pipes `\|` display as newlines) |
 | `get int.thresh` | Interference threshold |
-| `get agc.reset.interval` | AGC reset interval in ms (stored in 4 s steps) |
+| `get agc.reset.interval` | Removed - replies `use rxduty instead` |
 | `get multi.acks` | Extra ACK transmit count (`0` or `1`) |
 | `get path.hash.mode` | Path hashing algorithm: `0`, `1`, or `2` |
 | `get loop.detect` | Loop detection level: `off`, `minimal`, `moderate`, or `strict` |
@@ -243,7 +243,7 @@ Changes are persisted immediately unless noted. Some require a reboot.
 | `set guest.password <pwd>` | | Set guest access password |
 | `set owner.info <text>` | Use `\|` for newlines | Owner/contact information |
 | `set int.thresh <value>` | | Interference detection threshold |
-| `set agc.reset.interval <ms>` | Rounded to 4 s | AGC reset interval |
+| `set agc.reset.interval <ms>` | Accepted, ignored | Removed - replies `use rxduty instead` |
 | `set multi.acks <0\|1>` | | Enable extra ACK transmits |
 | `set path.hash.mode <mode>` | 0, 1, or 2 | Path hashing algorithm |
 | `set loop.detect <mode>` | `off`, `minimal`, `moderate`, `strict` | Loop detection sensitivity |

zephcore/adapters/radio/LR1110Radio.cpp

@@ -73,12 +73,6 @@ void LR1110Radio::hwSetRxBoost(bool enable)
 	lr11xx_set_rx_boost(_dev, enable);
 }
 
-void LR1110Radio::hwResetAGC()
-{
-	/* Warm sleep → Calibrate(ALL) → re-calibrate image → re-apply RX boost */
-	lr11xx_reset_agc(_dev);
-}
-
 uint32_t LR1110Radio::hwWakeupTimeUs()
 {
 	return lr11xx_get_wakeup_time_us(_dev);

zephcore/adapters/radio/LR1110Radio.h

@@ -27,7 +27,6 @@ protected:
 	int16_t hwGetCurrentRSSI() override;
 	bool hwIsReceiving() override;
 	void hwSetRxBoost(bool enable) override;
-	void hwResetAGC() override;
 	uint32_t hwWakeupTimeUs() override;
 	int hwCadProbe(int8_t level) override;
 	void hwCadSetPeakOffset(int8_t offset) override;

zephcore/adapters/radio/LR2021Radio.cpp

@@ -73,11 +73,6 @@ void LR2021Radio::hwSetRxBoost(bool enable)
 	lr20xx_set_rx_boost(_dev, enable);
 }
 
-void LR2021Radio::hwResetAGC()
-{
-	lr20xx_reset_agc(_dev);
-}
-
 int LR2021Radio::hwCadProbe(int8_t level)
 {
 	return lr20xx_cad_probe(_dev, level);

zephcore/adapters/radio/LR2021Radio.h

@@ -27,7 +27,6 @@ protected:
 	int16_t hwGetCurrentRSSI() override;
 	bool hwIsReceiving() override;
 	void hwSetRxBoost(bool enable) override;
-	void hwResetAGC() override;
 	int hwCadProbe(int8_t level) override;
 	void hwCadSetPeakOffset(int8_t offset) override;
 	uint8_t hwCadBasePeak() override;

zephcore/adapters/radio/LoRaRadioBase.cpp

@@ -47,9 +47,11 @@ LoRaRadioBase::LoRaRadioBase(const struct device *lora_dev, MainBoard &board,
 	  _last_rssi(0), _last_snr(0),
 	  _rx_head(0), _rx_tail(0),
 	  _noise_floor(DEFAULT_NOISE_FLOOR), _calibration_threshold(0), _ema_unguarded(0),
+	  _noise_floor_next_ms(0),
 	  _cad_auto(false), _cad_offset(0), _cad_probe_interval_s(0),
 	  _cad_busycap_pct(0),
-	  _cad_last_probe_ms(0), _cad_last_decay_ms(0), _cad_probe_rr(0),
+	  _cad_last_probe_ms(0), _cad_last_decay_ms(0), _cad_retry_ms(0),
+	  _cad_probe_rr(0),
 	  _rx_duty_cycle_enabled(IS_ENABLED(CONFIG_ZEPHCORE_LORA_RX_DUTY_CYCLE)),
 	  _rx_boost_enabled(true),
 	  _dc_last_rx_us(0), _dc_last_sleep_us(0),
@@ -786,6 +788,19 @@ void LoRaRadioBase::triggerNoiseFloorCalibrate(int threshold)
 {
 	_calibration_threshold = threshold;
 
+	/* Own the sampling cadence rather than inheriting the caller's.  Early
+	 * calls are a no-op, so this is safe to invoke from any wake. */
+	int64_t now = k_uptime_get();
+
+	if (_noise_floor_next_ms != 0 && now < _noise_floor_next_ms) {
+		return;
+	}
+
+	/* Due.  Any bail-out below is a blocked attempt, not a completed one —
+	 * push the deadline out by the short retry so msUntilNextMaintenance()
+	 * cannot report "due now" on a loop. */
+	_noise_floor_next_ms = now + NOISE_FLOOR_RETRY_MS;
+
 	if (!atomic_get(&_in_recv_mode) || atomic_get(&_tx_active)) {
 		return;
 	}
@@ -809,10 +824,15 @@ void LoRaRadioBase::triggerNoiseFloorCalibrate(int threshold)
 	for (int i = 0; i < NOISE_FLOOR_SAMPLES_PER_TICK; i++) {
 		samples[i] = hwGetCurrentRSSI();
 		if (samples[i] == -128) {
-			/* Chip busy or RSSI read contended — retry next tick. */
+			/* Chip busy or RSSI read contended — keep the short
+			 * retry deadline set above and try again shortly. */
 			return;
 		}
 	}
+
+	/* A full sample landed: next one is a full interval away. */
+	_noise_floor_next_ms = now + NOISE_FLOOR_INTERVAL_MS;
+
 	/* Insertion sort — tiny array, branch-friendly on Cortex-M */
 	for (int i = 1; i < NOISE_FLOOR_SAMPLES_PER_TICK; i++) {
 		int16_t key = samples[i];
@@ -873,35 +893,6 @@ void LoRaRadioBase::triggerNoiseFloorCalibrate(int threshold)
 		rssi, _noise_floor, _ema_unguarded - 1);
 }
 
-void LoRaRadioBase::resetAGC()
-{
-	/* Don't reset AGC while transmitting or receiving — warm sleep would
-	 * abort the TX or corrupt the incoming packet.  maintenanceLoop()
-	 * will retry next housekeeping cycle.
-	 * Also skip if the chip is in its duty-cycle sleep phase: hwResetAGC()
-	 * holds the SPI mutex with K_FOREVER and would hang for 3 s. */
-	if (atomic_get(&_tx_active) || isReceiving()) {
-		return;
-	}
-	if (_rx_duty_cycle_enabled && hwIsChipBusy()) {
-		return;
-	}
-
-	hwResetAGC();
-
-	/* Warm sleep + calibrate leaves the radio in STANDBY.
-	 * Restart receive if we were in RX mode. */
-	if (atomic_get(&_in_recv_mode)) {
-		startReceive();
-	}
-
-	/* Reset noise floor so it reconverges from scratch (seed + warmup).
-	 * Without this, a stuck _noise_floor of -120 makes the sampling threshold
-	 * too low to accept normal samples, self-reinforcing the stuck value. */
-	_noise_floor = DEFAULT_NOISE_FLOOR;
-	_ema_unguarded = 0;
-}
-
 bool LoRaRadioBase::isReceiving()
 {
 	if (!atomic_get(&_in_recv_mode) || atomic_get(&_tx_active)) {
@@ -1116,6 +1107,14 @@ void LoRaRadioBase::cadMaintenance()
 	if (now - _cad_last_probe_ms < (int64_t)_cad_probe_interval_s * 1000) {
 		return;
 	}
+	if (now < _cad_retry_ms) {
+		return;
+	}
+
+	/* Due.  _cad_last_probe_ms only advances on a probe that actually runs,
+	 * so hold a short retry deadline across the guards below — otherwise a
+	 * blocked probe reports "due now" to msUntilNextMaintenance() forever. */
+	_cad_retry_ms = now + CAD_PROBE_RETRY_MS;
 
 	/* Same guards as the noise-floor calibrator: only probe from idle
 	 * continuous/duty-cycle RX, never during TX or an active packet,
@@ -1223,6 +1222,57 @@ void LoRaRadioBase::cadMaintenance()
 	}
 }
 
+/* int64 uptime delta → the uint32 "ms from now" the maintenance contract wants.
+ * Already-passed deadlines saturate at 0 (due now), far-future ones at IDLE. */
+static uint32_t clampDeadline(int64_t remaining_ms)
+{
+	if (remaining_ms <= 0) {
+		return 0;
+	}
+	if (remaining_ms >= (int64_t)mesh::MAINTENANCE_IDLE) {
+		return mesh::MAINTENANCE_IDLE;
+	}
+	return (uint32_t)remaining_ms;
+}
+
+/* When does this radio next need a maintenance call?  Two independent items:
+ * the noise floor sampler (always running) and the CAD calibrator (only when
+ * probing is enabled).  Both hold absolute uptime deadlines, so this is a pure
+ * read — it must not touch the chip, since the event loop calls it on every
+ * wake to decide how long it may sleep. */
+uint32_t LoRaRadioBase::msUntilNextMaintenance()
+{
+	int64_t now = k_uptime_get();
+	uint32_t next = mesh::MAINTENANCE_IDLE;
+
+	/* Noise floor.  A zero deadline means "never sampled yet" — due now. */
+	if (_noise_floor_next_ms == 0) {
+		return 0;
+	}
+	next = clampDeadline(_noise_floor_next_ms - now);
+
+	if (_cad_probe_interval_s == 0) {
+		return next;
+	}
+
+	/* CAD probe: whichever is later of the interval since the last probe
+	 * that actually ran and any retry deadline left by a blocked one. */
+	int64_t probe_at = _cad_last_probe_ms + (int64_t)_cad_probe_interval_s * 1000;
+
+	if (probe_at < _cad_retry_ms) {
+		probe_at = _cad_retry_ms;
+	}
+	next = mesh::maintenanceSooner(next, clampDeadline(probe_at - now));
+
+	/* Stats decay. _cad_last_decay_ms == 0 means the first call latches it
+	 * rather than decaying, so treat that as due now. */
+	if (_cad_last_decay_ms == 0) {
+		return 0;
+	}
+	return mesh::maintenanceSooner(
+		next, clampDeadline(_cad_last_decay_ms + (int64_t)CAD_STATS_DECAY_MS - now));
+}
+
 int LoRaRadioBase::formatCadStatus(char *buf, int cap)
 {
 	uint8_t base = hwCadBasePeak();

zephcore/adapters/radio/LoRaRadioBase.h

@@ -68,7 +68,6 @@ public:
 	/* Advanced radio features */
 	int getNoiseFloor() const override;
 	void triggerNoiseFloorCalibrate(int threshold) override;
-	void resetAGC() override;
 	bool isReceiving() override;
 	void recoverRxState() override;
 
@@ -108,6 +107,7 @@ public:
 	void setCadParams(bool auto_enabled, int8_t offset,
 			  uint16_t probe_interval_s, uint8_t busycap_pct) override;
 	void cadMaintenance() override;
+	uint32_t msUntilNextMaintenance() override;
 	int8_t getCadOffset() const override { return _cad_offset; }
 	void resetCadStats() override;
 	int formatCadStatus(char *buf, int cap) override;
@@ -124,7 +124,6 @@ protected:
 	 * latch + raw IRQ bits, never clears.  Backs LoRaRadioBase::isReceiving(). */
 	virtual bool hwIsReceiving() = 0;
 	virtual void hwSetRxBoost(bool enable) = 0;
-	virtual void hwResetAGC() = 0;
 
 	/** GPIO-only BUSY check (no SPI). Default false for chips without duty-cycle sleep. */
 	virtual bool hwIsChipBusy() { return false; }
@@ -194,6 +193,13 @@ protected:
 	int _noise_floor;
 	int _calibration_threshold;
 	uint8_t _ema_unguarded;         /* tick counter for warmup + periodic bypass */
+	/* Absolute uptime deadline of the next floor sample.  The sampler used
+	 * to run on every housekeeping tick, which pinned its cadence to the
+	 * 5 s timer; owning its own deadline is what lets that timer go away.
+	 * Advanced by NOISE_FLOOR_INTERVAL_MS after a sample lands, and by the
+	 * shorter retry when an attempt is turned away because the radio was
+	 * mid-packet / transmitting / in its duty-cycle sleep window. */
+	int64_t _noise_floor_next_ms;
 
 	/* Adaptive CAD state */
 	struct CadLevelStats {
@@ -209,6 +215,11 @@ protected:
 	uint8_t _cad_busycap_pct;       /* airtime cap: max % TX deferred (0 = off) */
 	int64_t _cad_last_probe_ms;
 	int64_t _cad_last_decay_ms;
+	/* Earliest uptime at which a due-but-blocked probe may be retried.  The
+	 * interval check in cadMaintenance() is against _cad_last_probe_ms,
+	 * which only advances on a probe that actually ran — without this a
+	 * blocked probe would report "due now" forever and spin the wake. */
+	int64_t _cad_retry_ms;
 	uint8_t _cad_probe_rr;          /* round-robin index (sweep) / frontier mix counter */
 
 	int8_t pickCadProbeLevel();

zephcore/adapters/radio/SX126xRadio.cpp

@@ -80,12 +80,6 @@ void SX126xRadio::hwSetRxBoost(bool enable)
 	sx126x_set_rx_boost(_dev, enable);
 }
 
-void SX126xRadio::hwResetAGC()
-{
-	/* Warm sleep → Calibrate(ALL) → re-calibrate image → re-apply RX settings */
-	sx126x_reset_agc(_dev);
-}
-
 bool SX126xRadio::hwIsChipBusy()
 {
 	return sx126x_is_chip_busy(_dev);

zephcore/adapters/radio/SX126xRadio.h

@@ -29,7 +29,6 @@ protected:
 	int16_t hwGetCurrentRSSI() override;
 	bool hwIsReceiving() override;
 	void hwSetRxBoost(bool enable) override;
-	void hwResetAGC() override;
 	bool hwIsChipBusy() override;
 	uint32_t hwWakeupTimeUs() override;
 	int hwCadProbe(int8_t level) override;

zephcore/adapters/radio/SX127xRadio.cpp

@@ -9,7 +9,6 @@
  *   hwGetCurrentRSSI()    — returns -80 dBm sentinel (no hardware path)
  *   hwIsReceiving()       — always false (no preamble/header IRQ exposed)
  *   hwSetRxBoost()        — no-op (SX127x has no RX boost register)
- *   hwResetAGC()          — no-op (loramac-node manages AGC internally)
  *   hwIsChipBusy()        — inherited false (no BUSY pin on SX127x)
  *
  * Everything else (configure, send, receive) uses the standard API.
@@ -96,22 +95,4 @@ void SX127xRadio::hwSetRxBoost(bool enable)
 	ARG_UNUSED(enable);
 }
 
-void SX127xRadio::hwResetAGC()
-{
-	/* The loramac-node SX127x driver manages AGC recalibration internally
-	 * (RadioSetRxConfig re-programs all gain registers on every RX config
-	 * call).  No explicit AGC reset is needed or possible via the standard
-	 * API. */
-}
-
-void SX127xRadio::resetAGC()
-{
-	/* hwResetAGC() is a no-op, so skip the base-class resetAGC() entirely.
-	 * The base class calls startReceive() after hwResetAGC(), but the
-	 * loramac-node modem mutex (STATE_BUSY during async RX) causes
-	 * lora_recv_async() to return -EBUSY, setting _in_recv_mode = 0 and
-	 * corrupting the state machine.  The loramac-node driver self-manages
-	 * AGC, so nothing needs to happen here. */
-}
-
 } /* namespace mesh */

zephcore/adapters/radio/SX127xRadio.h

@@ -49,15 +49,6 @@ protected:
 	/* SX127x has no RX boost / LNA gain switch via standard API. No-op. */
 	void hwSetRxBoost(bool enable) override;
 
-	/* SX127x loramac-node driver manages AGC automatically. No-op. */
-	void hwResetAGC() override;
-
-	/* Override public resetAGC(): hwResetAGC() is a no-op, and the
-	 * loramac-node modem mutex makes the base-class startReceive() call
-	 * fail with -EBUSY (modem STATE_BUSY during async RX), which would
-	 * set _in_recv_mode = 0 and corrupt the state machine. */
-	void resetAGC() override;
-
 	/* SX127x has no BUSY pin. Default (false) from base is correct. */
 	/* bool hwIsChipBusy() — inherited, returns false */
 };

zephcore/adapters/radio/radio_common.h

@@ -12,7 +12,7 @@
 #include <zephyr/drivers/lora.h>
 
 /* --- Noise floor calibration (EMA) ---
- * Median-of-N RSSI reads → EMA.  alpha = 1/8, converges in ~8 ticks (~40s).
+ * Median-of-N RSSI reads → EMA.  alpha = 1/8, converges in ~8 ticks.
  * Samples above floor + SAMPLING_THRESHOLD rejected as interference. */
 #define NOISE_FLOOR_EMA_SHIFT            3   /* alpha = 1/(1<<3) = 1/8 */
 #define NOISE_FLOOR_SAMPLES_PER_TICK     8   /* median of 8 reads per tick */
@@ -20,6 +20,24 @@
 #define NOISE_FLOOR_SAMPLING_THRESHOLD   14  /* dB above floor to reject as interference */
 #define DEFAULT_NOISE_FLOOR              0   /* sentinel: seed from first sample */
 
+/* Sampling cadence.  The sampler used to run once per housekeeping tick and so
+ * inherited that 5 s period; owning an explicit interval is what let the
+ * periodic tick go away (see mesh/Maintenance.h).  Both the 8-sample warmup and
+ * the every-16th-sample unguarded bypass are counted in samples, so they scale
+ * with this value — see the Kconfig help before changing it. */
+#define NOISE_FLOOR_INTERVAL_MS  CONFIG_ZEPHCORE_NOISE_FLOOR_INTERVAL_MS
+/* A due sample that lands while the radio is mid-packet, transmitting, or in
+ * its duty-cycle sleep window is retried on this deadline rather than waiting a
+ * full interval.
+ *
+ * 5000 is not a tuned guess: before the deadline conversion a blocked sample
+ * simply waited for the next 5 s housekeeping tick, so this reproduces the old
+ * retry grid exactly.  It matters under RX duty cycle, where the chip is in its
+ * sleep window a large fraction of the time and blocked attempts are the norm
+ * rather than the exception — a shorter retry there can push the wake rate
+ * ABOVE the fixed tick this conversion replaced, inverting the whole point. */
+#define NOISE_FLOOR_RETRY_MS             5000
+
 /* --- Adaptive CAD (LBT detPeak calibration) ---
  * Housekeeping-tick CAD probes accumulate per-level busy/free statistics;
  * a one-sided staircase converges on the lowest detPeak offset whose
@@ -67,6 +85,11 @@
 #define CAD_BUSY_DEFER_HYST_PERMILLE 100  /* descend only if frontier busy <= cap-10% */
 #define CAD_PROBE_RSSI_GUARD     7     /* dB above floor = channel visibly busy, skip probe */
 #define CAD_STATS_DECAY_MS       (6UL * 3600UL * 1000UL)  /* halve counters every 6 h */
+/* Retry deadline for a due probe turned away by the idle-RX guards or the
+ * RSSI prefilter.  Those paths leave _cad_last_probe_ms untouched (by design —
+ * a skipped probe is not a probe), so the deadline query needs its own marker
+ * to avoid reporting "due now" on every wake. */
+#define CAD_PROBE_RETRY_MS       2000
 
 /* --- RX ring buffer --- */
 #define RX_RING_SIZE 8  /* ~2 KB; buffers burst arrivals at SF7/BW500 */

zephcore/app/RepeaterMesh.cpp

@@ -1396,6 +1396,48 @@ void RepeaterMesh::loop() {
     last_millis = now;
 }
 
+uint32_t RepeaterMesh::msUntilNextMaintenance() {
+    uint32_t now = (uint32_t)_ms->getMillis();
+    uint32_t next = mesh::Mesh::msUntilNextMaintenance();
+
+    /* Advert timers.  0 means "disabled" for both — flood defaults to 47 h,
+     * periodic local advert is off unless configured. */
+    if (next_flood_advert) {
+        next = mesh::maintenanceSooner(next, mesh::maintenanceUntil(now, next_flood_advert));
+    }
+    if (next_local_advert) {
+        next = mesh::maintenanceSooner(next, mesh::maintenanceUntil(now, next_local_advert));
+    }
+
+    /* Temporary radio params: apply, then revert. Both CLI-armed, both 0 when idle. */
+    if (set_radio_at) {
+        next = mesh::maintenanceSooner(next, mesh::maintenanceUntil(now, set_radio_at));
+    }
+    if (revert_radio_at) {
+        next = mesh::maintenanceSooner(next, mesh::maintenanceUntil(now, revert_radio_at));
+    }
+
+    /* Deferred ACL write-back. */
+    if (dirty_contacts_expiry) {
+        next = mesh::maintenanceSooner(next, mesh::maintenanceUntil(now, dirty_contacts_expiry));
+    }
+
+#if IS_ENABLED(CONFIG_ZEPHCORE_REPEATER_UPLINK) && IS_ENABLED(CONFIG_MQTT_LIB)
+    if (_uplink_next_status_at) {
+        next = mesh::maintenanceSooner(next,
+                                       mesh::maintenanceUntil(now, _uplink_next_status_at));
+    }
+#endif
+
+    /* Mesh time sync evaluates at most every 15 min, and only when enabled. */
+    if (_prefs.meshtimesync) {
+        next = mesh::maintenanceSooner(
+            next, _timesync.msUntilNextEval((uint32_t)(k_uptime_get() / 1000)));
+    }
+
+    return next;
+}
+
 void RepeaterMesh::timeSyncTick() {
     if (!_prefs.meshtimesync) return;
     if (!_timesync.runTick(*getRTCClock())) return;

zephcore/app/RepeaterMesh.h

@@ -169,12 +169,6 @@ protected:
     int getInterferenceThreshold() const override {
         return _prefs.interference_threshold;
     }
-    int getAGCResetInterval() const override {
-        if (_prefs.rx_duty_cycle) {
-            return 0;
-        }
-        return ((int)_prefs.agc_reset_interval) * 4000;
-    }
     uint8_t getExtraAckTransmitCount() const override {
         return _prefs.multi_acks;
     }
@@ -288,5 +282,12 @@ public:
     void handleCommand(uint32_t sender_timestamp, char* command, char* reply);
     void loop();
 
+    /* Folds this role's own time-based work in loop() — advert timers,
+     * tempradio apply/revert, ACL flush, uplink status, mesh time sync — into
+     * the base maintenance deadline, so the event loop can arm one wake that
+     * covers both loop() and maintenanceLoop().  Every item here is an
+     * absolute deadline already; none of them was ever a poll. */
+    uint32_t msUntilNextMaintenance() override;
+
     bool hasPendingWork() const;
 };

zephcore/app/RoomServerMesh.h

@@ -138,12 +138,6 @@ protected:
     int getInterferenceThreshold() const override {
         return _prefs.interference_threshold;
     }
-    int getAGCResetInterval() const override {
-        if (_prefs.rx_duty_cycle) {
-            return 0;
-        }
-        return ((int)_prefs.agc_reset_interval) * 4000;
-    }
     uint8_t getExtraAckTransmitCount() const override {
         return _prefs.multi_acks;
     }

zephcore/helpers/CommonCLI.cpp

@@ -458,7 +458,7 @@ void CommonCLI::handleCommand(uint32_t sender_timestamp, const char* command, ch
         } else if (memcmp(config, "int.thresh", 10) == 0) {
             snprintf(reply, CLI_REPLY_SIZE, "> %u", (uint32_t)_prefs->interference_threshold);
         } else if (memcmp(config, "agc.reset.interval", 18) == 0) {
-            snprintf(reply, CLI_REPLY_SIZE, "> %u", ((uint32_t)_prefs->agc_reset_interval) * 4);
+            strcpy(reply, "Removed - use rxduty instead");
         } else if (memcmp(config, "multi.acks", 10) == 0) {
             snprintf(reply, CLI_REPLY_SIZE, "> %u", (uint32_t)_prefs->multi_acks);
         } else if (memcmp(config, "allow.read.only", 15) == 0) {
@@ -625,15 +625,14 @@ void CommonCLI::handleCommand(uint32_t sender_timestamp, const char* command, ch
                 strcpy(reply, "OK");
             }
         } else if (memcmp(config, "agc.reset.interval ", 19) == 0) {
-            /* Companion runtime never reads this (getAGCResetInterval is only
-             * overridden in Repeater/RoomServer). */
-            if (strcmp(_callbacks->getRole(), "companion") == 0) {
-                strcpy(reply, "Error: not supported on companion");
-            } else {
-                _prefs->agc_reset_interval = atoi(&config[19]) / 4;
-                savePrefs();
-                snprintf(reply, CLI_REPLY_SIZE, "OK - interval rounded to %u", ((uint32_t)_prefs->agc_reset_interval) * 4);
-            }
+            /* Periodic AGC recalibration was removed: it reset the noise floor
+             * to its unseeded sentinel on every fire, forcing a fresh seed and
+             * a full EMA warmup, and it was already forced off under RX duty
+             * cycle.  RX duty cycle is the supported way to cut RX current.
+             * The prefs BYTE is retained (read/written, never acted on) — the
+             * on-disk layout is byte-exact and shifting it would corrupt every
+             * existing node's prefs. */
+            strcpy(reply, "Removed - use rxduty instead");
         } else if (memcmp(config, "cad.auto ", 9) == 0) {
             if (memcmp(&config[9], "on", 2) == 0 || memcmp(&config[9], "off", 3) == 0) {
                 _prefs->cad_auto = (config[9] == 'o' && config[10] == 'n') ? 1 : 0;

zephcore/helpers/MeshTimeSync.h

@@ -135,6 +135,18 @@ public:
 	 * gates on its own enable pref. */
 	bool runTick(mesh::RTCClock &rtc);
 
+	/* Milliseconds until runTick() would next actually evaluate (it no-ops
+	 * in between).  Lets a deadline-driven event loop size its sleep instead
+	 * of calling this on a fixed cadence.  Callers still gate on their own
+	 * enable pref — a disabled time sync has no deadline at all. */
+	uint32_t msUntilNextEval(uint32_t uptime_secs) const
+	{
+		if (_next_eval_uptime == 0 || uptime_secs >= _next_eval_uptime) {
+			return 0;
+		}
+		return (_next_eval_uptime - uptime_secs) * 1000U;
+	}
+
 	/* Manual clock set (CLI time/clock sync, app time set, SNTP): arms the
 	 * 7-day suppression window AND drift-envelope pedigree. Suppression
 	 * gates bootstrap too. */

zephcore/helpers/NodePrefs.h

@@ -62,7 +62,10 @@ struct NodePrefs {
 	uint8_t flood_max_unscoped;     // hop limit for un-scoped (ROUTE_TYPE_FLOOD) floods
 	uint8_t flood_max_advert;       // hop limit for ADVERT floods (curbs advert churn)
 	uint8_t interference_threshold;
-	uint8_t agc_reset_interval;     // stored as secs / 4
+	uint8_t agc_reset_interval;     // RETIRED: read/written for on-disk layout
+	                                // compatibility only, never acted on.
+	                                // "set agc.reset.interval" replies
+	                                // "use rxduty instead".
 	// Power saving
 	uint8_t powersaving_enabled;
 	// GPS settings

zephcore/include/mesh/ContentionTracker.h

@@ -9,6 +9,7 @@
 #pragma once
 
 #include <stdint.h>
+#include <mesh/Maintenance.h>
 
 namespace mesh {
 
@@ -43,6 +44,13 @@ public:
 	/* Finalize expired entries into EMA. */
 	void tick(uint32_t now_ms);
 
+	/* Milliseconds until tick() next has work to do, or MAINTENANCE_IDLE
+	 * when nothing is pending (no tracked entries, EMA already at rest).
+	 * Lets the event loop schedule a wake at the deadline instead of
+	 * polling: an idle repeater has no tracked retransmits and a decayed
+	 * EMA, so this returns MAINTENANCE_IDLE and costs no wakes at all. */
+	uint32_t msUntilNextTick(uint32_t now_ms) const;
+
 	float getContentionEstimate() const;
 
 	/* Saturating curve: MIN + (MAX-MIN) * est/(est+HALFPOINT).
@@ -65,6 +73,15 @@ private:
 	static constexpr float DEFAULT_BACKOFF_MULT = 0.2f; /* airtime*0.2 per dupe heard */
 	static constexpr uint32_t REACTIVE_HARD_CAP_MS = 2000; /* max cumulative reactive extension */
 	static constexpr uint32_t STALE_MS = 300000;      /* 5 min: reset EMA if no traffic */
+	/* Wall-clock period of one 1/8 stale-decay step.  The decay used to be
+	 * "one step per tick()", which silently coupled it to the 5 s
+	 * housekeeping cadence; pinning it to 5000 ms keeps exactly that rate
+	 * while making it independent of how often tick() is actually called. */
+	static constexpr uint32_t DECAY_PERIOD_MS = 5000;
+	/* Bound the catch-up loop when tick() is called much later than one
+	 * period.  8 steps takes the EMA down by ~66%; anything beyond that is
+	 * indistinguishable from rest given the WARMUP/read paths. */
+	static constexpr int MAX_DECAY_CATCHUP = 8;
 
 	struct Entry {
 		uint32_t hash32;
@@ -79,6 +96,7 @@ private:
 	uint32_t _ema_x256;
 	int _finalized_count;
 	uint32_t _last_retransmit_ms;
+	uint32_t _last_decay_ms;
 	float _backoff_multiplier;
 
 	void finalizeEntry(int idx);

zephcore/include/mesh/Dispatcher.h

@@ -11,6 +11,7 @@
 #include <mesh/Utils.h>
 #include <mesh/Radio.h>
 #include <mesh/Clock.h>
+#include <mesh/Maintenance.h>
 #include <string.h>
 
 namespace mesh {
@@ -29,8 +30,6 @@ public:
 	virtual uint32_t getOutboundSchedule(int i) const = 0;
 	virtual bool rescheduleOutbound(int i, uint32_t new_scheduled_for) = 0;
 	virtual uint8_t peekNextOutboundPriority(uint32_t now) const = 0;
-	virtual void queueInbound(Packet *packet, uint32_t scheduled_for) = 0;
-	virtual Packet *getNextInbound(uint32_t now) = 0;
 };
 
 /* Notifies event loop of pending TX so it can schedule a wake. */
@@ -43,6 +42,9 @@ typedef uint32_t DispatcherAction;
 #define ACTION_RETRANSMIT(pri)   (((uint32_t)1 + (pri))<<24)
 #define ACTION_RETRANSMIT_DELAYED(pri, _delay)  ((((uint32_t)1 + (pri))<<24) | (_delay))
 
+/* Radio considered stalled after this long neither receiving nor sending. */
+#define RADIO_STALL_THRESHOLD_MS    8000
+
 #define ERR_EVENT_FULL              (1 << 0)
 #define ERR_EVENT_CAD_TIMEOUT       (1 << 1)
 #define ERR_EVENT_STARTRX_TIMEOUT   (1 << 2)
@@ -57,7 +59,6 @@ class Dispatcher {
 	uint32_t last_budget_update;
 	uint32_t duty_cycle_window_ms;
 	uint32_t radio_nonrx_start;
-	uint32_t next_agc_reset_time;
 	bool prev_isrecv_mode;
 	int8_t cad_offset_shadow;
 	bool cad_offset_shadow_valid;
@@ -89,7 +90,6 @@ protected:
 	virtual uint32_t getCADFailRetryDelay() const;
 	virtual uint32_t getCADFailMaxDuration() const;
 	virtual int getInterferenceThreshold() const { return 0; }
-	virtual int getAGCResetInterval() const { return 0; }
 	virtual uint32_t getDutyCycleWindowMs() const { return 3600000UL; } /* 1h default */
 	/* Adaptive CAD: called when the auto staircase moved the operating
 	 * detPeak offset — subclasses persist it to prefs. */
@@ -99,6 +99,13 @@ public:
 	void begin();
 	void loop();
 	void maintenanceLoop();
+
+	/* Milliseconds until maintenanceLoop() next has work, or
+	 * MAINTENANCE_IDLE if nothing is pending.  Cheap and side-effect free:
+	 * the event loop calls it after every pass to size its next sleep.
+	 * Subclasses that add their own time-based work in loop() override this
+	 * and fold their deadlines in — see RepeaterMesh. */
+	virtual uint32_t msUntilNextMaintenance();
 	Packet *obtainNewPacket();
 	void releasePacket(Packet *packet);
 	void sendPacket(Packet *packet, uint8_t priority, uint32_t delay_millis = 0);

zephcore/include/mesh/Maintenance.h

@@ -0,0 +1,50 @@
+/*
+ * SPDX-License-Identifier: MIT
+ * Deadline-driven maintenance scheduling.
+ *
+ * maintenanceLoop() used to be driven by a fixed periodic tick (the 5 s
+ * "housekeeping" timer).  Every item inside it is really either a deadline
+ * (AGC reset, CAD probe, advert timers, tempradio revert) or a sampler with
+ * its own interval (noise floor) — nothing needs a poll.  The msUntilNext*()
+ * queries below let the event loop ask "when does maintenance next have work?"
+ * and arm a single one-shot wake for exactly that moment, so an otherwise idle
+ * node sleeps until its soonest real deadline instead of waking on a fixed
+ * cadence.
+ *
+ * Contract: every msUntilNext*() returns milliseconds from now, 0 when work is
+ * due right now, and MAINTENANCE_IDLE when the item has nothing pending at all.
+ * Callers combine them with maintenanceSooner().
+ */
+
+#pragma once
+
+#include <stdint.h>
+
+namespace mesh {
+
+/* "No deadline pending." Deliberately not UINT32_MAX so that callers can add
+ * a modest slack to a returned deadline without wrapping. */
+static const uint32_t MAINTENANCE_IDLE = 0x7FFFFFFFu;
+
+/* Fold one deadline into a running minimum. */
+static inline uint32_t maintenanceSooner(uint32_t a, uint32_t b)
+{
+	return (a < b) ? a : b;
+}
+
+/* Milliseconds from now until `deadline` (a millisecond-clock timestamp),
+ * saturating at 0 when it has already passed.  Wrap-safe: the subtraction is
+ * done in unsigned arithmetic and the sign is taken from the wrapped result,
+ * matching Dispatcher::millisHasNowPassed(). */
+static inline uint32_t maintenanceUntil(uint32_t now_ms, uint32_t deadline_ms)
+{
+	int32_t remaining = (int32_t)(deadline_ms - now_ms);
+
+	if (remaining <= 0) {
+		return 0;
+	}
+	return ((uint32_t)remaining < MAINTENANCE_IDLE) ? (uint32_t)remaining
+						       : MAINTENANCE_IDLE;
+}
+
+} /* namespace mesh */

zephcore/include/mesh/Mesh.h

@@ -80,6 +80,7 @@ public:
 	void begin();
 	void loop();
 	void maintenanceLoop();
+	uint32_t msUntilNextMaintenance() override;
 
 	LocalIdentity self_id;
 

zephcore/include/mesh/Radio.h

@@ -6,6 +6,7 @@
 #pragma once
 
 #include <stdint.h>
+#include <mesh/Maintenance.h>
 
 namespace mesh {
 
@@ -22,7 +23,6 @@ public:
 
 	virtual int getNoiseFloor() const { return 0; }
 	virtual void triggerNoiseFloorCalibrate(int threshold) { (void)threshold; }
-	virtual void resetAGC() {}
 
 	virtual bool isInRecvMode() const = 0;
 	virtual bool isReceiving() { return false; }
@@ -49,6 +49,11 @@ public:
 	/* One housekeeping tick of the CAD calibrator: maybe run a probe,
 	 * update stats, maybe step the staircase (auto mode). */
 	virtual void cadMaintenance() {}
+
+	/* Milliseconds until this radio's periodic work (noise floor sampling,
+	 * CAD probing/decay) next needs a call, or MAINTENANCE_IDLE when it has
+	 * nothing pending.  Radios with no periodic work keep the default. */
+	virtual uint32_t msUntilNextMaintenance() { return MAINTENANCE_IDLE; }
 	virtual int8_t getCadOffset() const { return 0; }
 	virtual void resetCadStats() {}
 	/* Writes a human-readable status block; returns chars written (0 = not

zephcore/include/mesh/StaticPoolPacketManager.h

@@ -23,8 +23,6 @@ public:
 	uint32_t getOutboundSchedule(int i) const override;
 	bool rescheduleOutbound(int i, uint32_t new_scheduled_for) override;
 	uint8_t peekNextOutboundPriority(uint32_t now) const override;
-	void queueInbound(Packet *packet, uint32_t scheduled_for) override;
-	Packet *getNextInbound(uint32_t now) override;
 };
 
 } /* namespace mesh */

zephcore/src/ContentionTracker.cpp

@@ -11,7 +11,8 @@ namespace mesh {
 
 ContentionTracker::ContentionTracker()
 	: _next_idx(0), _ema_x256(0), _finalized_count(0),
-	  _last_retransmit_ms(0), _backoff_multiplier(DEFAULT_BACKOFF_MULT)
+	  _last_retransmit_ms(0), _last_decay_ms(0),
+	  _backoff_multiplier(DEFAULT_BACKOFF_MULT)
 {
 	memset(_ring, 0, sizeof(_ring));
 }
@@ -144,12 +145,72 @@ void ContentionTracker::tick(uint32_t now_ms)
 		}
 	}
 
-	/* Decay EMA toward 0 if no retransmit in STALE_MS */
-	if (_last_retransmit_ms != 0 && now_ms - _last_retransmit_ms > STALE_MS) {
-		if (_ema_x256 > 0) {
+	/* Decay EMA toward 0 if no retransmit in STALE_MS.  Paced by wall clock
+	 * (one 1/8 step per DECAY_PERIOD_MS) rather than one step per call, so
+	 * the decay rate no longer depends on how often the event loop happens
+	 * to call us — see DECAY_PERIOD_MS. */
+	if (_last_retransmit_ms != 0 && now_ms - _last_retransmit_ms > STALE_MS &&
+	    _ema_x256 > 0) {
+		if (_last_decay_ms == 0) {
+			/* Arm one full period in the PAST, so the loop below
+			 * applies a step on this very call.  The old code decayed
+			 * immediately on the first tick that observed staleness;
+			 * arming at now_ms instead would make the first check
+			 * (0 < DECAY_PERIOD_MS) break without decaying, deferring
+			 * onset by a whole period on every stale transition — and
+			 * would let the reset branch below starve decay entirely
+			 * for traffic that goes stale and un-stale repeatedly.
+			 * Unsigned wraparound makes this exact even near zero. */
+			_last_decay_ms = now_ms - DECAY_PERIOD_MS;
+		}
+		for (int n = 0; n < MAX_DECAY_CATCHUP; n++) {
+			if (now_ms - _last_decay_ms < DECAY_PERIOD_MS ||
+			    _ema_x256 == 0) {
+				break;
+			}
 			_ema_x256 -= _ema_x256 >> EMA_SHIFT;
+			_last_decay_ms += DECAY_PERIOD_MS;
+		}
+	} else {
+		/* Not decaying — restart the phase next time we are. */
+		_last_decay_ms = 0;
+	}
+}
+
+uint32_t ContentionTracker::msUntilNextTick(uint32_t now_ms) const
+{
+	uint32_t next = MAINTENANCE_IDLE;
+
+	/* Soonest ring entry to fall out of the observation window. */
+	for (int i = 0; i < RING_SIZE; i++) {
+		if (!_ring[i].active) {
+			continue;
+		}
+		next = maintenanceSooner(
+			next, maintenanceUntil(now_ms,
+					       _ring[i].first_seen_ms + WINDOW_MS + 1));
+	}
+
+	/* Stale-decay step, only while there is EMA left to decay. */
+	if (_last_retransmit_ms != 0 && _ema_x256 > 0) {
+		uint32_t stale_at = _last_retransmit_ms + STALE_MS + 1;
+
+		if (now_ms - _last_retransmit_ms > STALE_MS) {
+			/* Already stale: next step is one period after the last
+			 * one (or immediately, if we have not started yet). */
+			next = maintenanceSooner(
+				next, _last_decay_ms == 0
+					      ? 0
+					      : maintenanceUntil(now_ms,
+								 _last_decay_ms +
+									 DECAY_PERIOD_MS));
+		} else {
+			next = maintenanceSooner(next,
+						 maintenanceUntil(now_ms, stale_at));
 		}
 	}
+
+	return next;
 }
 
 float ContentionTracker::getContentionEstimate() const

zephcore/src/Dispatcher.cpp

@@ -34,7 +34,6 @@ Dispatcher::Dispatcher(Radio &radio, MillisecondClock &ms, PacketManager &mgr)
 	tx_budget_ms = 0;
 	last_budget_update = 0;
 	duty_cycle_window_ms = 0;
-	next_agc_reset_time = 0;
 	_err_flags = 0;
 	radio_nonrx_start = 0;
 	prev_isrecv_mode = true;
@@ -148,15 +147,8 @@ void Dispatcher::loop()
 		} else {
 			return;
 		}
-		next_agc_reset_time = futureMillis(getAGCResetInterval());
 	}
 
-	{
-		Packet *pkt = _mgr->getNextInbound((uint32_t)_ms->getMillis());
-		if (pkt) {
-			processRecvPacket(pkt);
-		}
-	}
 	checkRecv();
 	checkSend();
 }
@@ -165,8 +157,12 @@ void Dispatcher::maintenanceLoop()
 {
 	_radio->triggerNoiseFloorCalibrate(getInterferenceThreshold());
 
-	/* RX mode watchdog: TX counts as "active" to avoid false triggers
-	 * when the 5s housekeeping timer misses brief RX windows. */
+	/* RX mode watchdog: TX counts as "active" to avoid false triggers when
+	 * a maintenance pass lands between brief RX windows.  Diagnostic only —
+	 * it raises a status bit and recovers nothing — but that bit is surfaced
+	 * on every role: the repeater/room-server "stats" CLI reply and binary
+	 * telemetry read _err_flags directly, the MQTT uplink publishes it, and
+	 * the companion returns it in its BLE device-status response. */
 	bool is_active = _radio->isInRecvMode() || !_radio->isSendComplete();
 	if (is_active != prev_isrecv_mode) {
 		prev_isrecv_mode = is_active;
@@ -174,16 +170,11 @@ void Dispatcher::maintenanceLoop()
 			radio_nonrx_start = (uint32_t)_ms->getMillis();
 		}
 	}
-	if (!is_active && (uint32_t)_ms->getMillis() - radio_nonrx_start > 8000) { /* 8s stall threshold */
+	if (!is_active &&
+	    (uint32_t)_ms->getMillis() - radio_nonrx_start > RADIO_STALL_THRESHOLD_MS) {
 		_err_flags |= ERR_EVENT_STARTRX_TIMEOUT;
 	}
 
-	/* Periodic AGC recalibration */
-	if (getAGCResetInterval() > 0 && millisHasNowPassed(next_agc_reset_time)) {
-		_radio->resetAGC();
-		next_agc_reset_time = futureMillis(getAGCResetInterval());
-	}
-
 	/* Adaptive CAD: probe scheduling + staircase live in the radio;
 	 * we only surface offset changes so the app layer can persist them. */
 	_radio->cadMaintenance();
@@ -199,6 +190,31 @@ void Dispatcher::maintenanceLoop()
 	}
 }
 
+uint32_t Dispatcher::msUntilNextMaintenance()
+{
+	uint32_t now = (uint32_t)_ms->getMillis();
+	/* Noise floor sampling + CAD probing/decay both live in the radio and
+	 * carry their own deadlines. */
+	uint32_t next = _radio->msUntilNextMaintenance();
+
+	/* Radio stall watchdog.  Only pending while the radio is known to be
+	 * neither receiving nor transmitting as of the last pass — the
+	 * transition into that state is itself event-driven (TX start, RX done,
+	 * CAD), so there is nothing to poll for while the radio is active.
+	 *
+	 * The already-flagged check is load-bearing: the verdict is a latched
+	 * status bit, so once raised its deadline sits permanently in the past.
+	 * Without this the query would return 0 on every call and the event loop
+	 * would re-arm at its minimum interval forever. */
+	if (!prev_isrecv_mode && !(_err_flags & ERR_EVENT_STARTRX_TIMEOUT)) {
+		next = maintenanceSooner(
+			next, maintenanceUntil(now, radio_nonrx_start +
+						       RADIO_STALL_THRESHOLD_MS));
+	}
+
+	return next;
+}
+
 bool Dispatcher::tryParsePacket(Packet *pkt, const uint8_t *raw, int len)
 {
 	int i = 0;

zephcore/src/Mesh.cpp

@@ -34,6 +34,14 @@ void Mesh::maintenanceLoop()
 	_contention.tick(now);
 }
 
+uint32_t Mesh::msUntilNextMaintenance()
+{
+	uint32_t now = (uint32_t)_ms->getMillis();
+
+	return maintenanceSooner(Dispatcher::msUntilNextMaintenance(),
+				 _contention.msUntilNextTick(now));
+}
+
 void Mesh::extendPendingRetransmit(uint32_t hash32)
 {
 	uint32_t now = (uint32_t)_ms->getMillis();

zephcore/src/StaticPoolPacketManager.cpp

@@ -113,7 +113,6 @@ struct PacketQueue {
 static Packet _packet_pool[POOL_SIZE];
 static PacketQueue _unused;
 static PacketQueue _send_queue;
-static PacketQueue _rx_queue;
 static bool _initialized = false;
 
 static void init_pool() {
@@ -202,18 +201,4 @@ uint8_t StaticPoolPacketManager::peekNextOutboundPriority(uint32_t now) const
 	return _send_queue.peekPriority(now);
 }
 
-void StaticPoolPacketManager::queueInbound(Packet *packet, uint32_t scheduled_for)
-{
-	if (!_rx_queue.add(packet, 0, scheduled_for)) {
-		LOG_WRN("queueInbound: FULL (%d entries) — dropping type=%d",
-			_rx_queue.count(), packet->getPayloadType());
-		free(packet);
-	}
-}
-
-Packet *StaticPoolPacketManager::getNextInbound(uint32_t now)
-{
-	return _rx_queue.get(now);
-}
-
 } /* namespace mesh */

zephcore/src/main_repeater.cpp

@@ -57,6 +57,19 @@ extern "C" void bt_ctlr_assert_handle(char *file, uint32_t line)
 /* UI subsystem (display, buttons, buzzer) */
 #include "ui_task.h"
 
+/* Headless repeaters link the weak no-op ui_* stubs (ui_headless_stubs.c), so
+ * the periodic UI refresh in the maintenance pass is pure work for nothing on
+ * them.  Guard the hot path on a real ui_* implementation being linked; the
+ * one-shot calls at init and in the CLI reply path stay unguarded, matching
+ * the rest of the file.
+ *
+ * This MUST mirror the CMake condition that selects the stubs, not the display
+ * devicetree node: ZEPHCORE_UI_DESIGN_BUTTON is enabled by BUTTONS *or*
+ * DISPLAY *or* BUZZER, so a board with buttons/buzzer and no panel still links
+ * the real UI and still needs these updates. */
+#define ZEPHCORE_HAS_UI (IS_ENABLED(CONFIG_ZEPHCORE_UI_DESIGN_BUTTON) || \
+			 IS_ENABLED(CONFIG_ZEPHCORE_UI_DESIGN_JOYSTICK))
+
 /* Radio + mesh includes (shared header selects LR1110 or SX126x) */
 #include <mesh/RadioIncludes.h>
 
@@ -85,15 +98,26 @@ static const struct gpio_dt_spec led1 = GPIO_DT_SPEC_GET(LED1_NODE, gpios);
 #define MESH_EVENT_LORA_RX       BIT(0)  /* LoRa packet received */
 #define MESH_EVENT_LORA_TX_DONE  BIT(1)  /* LoRa TX complete */
 #define MESH_EVENT_CLI_RX        BIT(2)  /* CLI command received */
-#define MESH_EVENT_HOUSEKEEPING  BIT(3)  /* Periodic housekeeping (noise floor, etc.) */
+#define MESH_EVENT_MAINTENANCE   BIT(3)  /* A maintenance deadline came due */
 #define MESH_EVENT_GPS_ACTION    BIT(4)  /* GPS state change (must run on main thread!) */
 #define MESH_EVENT_TX_DRAIN      BIT(5)  /* Outbound packet delay expired, run checkSend */
 #define MESH_EVENT_RTC_SAVE      BIT(6)  /* Hardware-RTC write requested off-main */
 #define MESH_EVENT_INIT_ADVERT   BIT(7)  /* Deferred boot advert — send on main thread */
-#define MESH_EVENT_ALL           (MESH_EVENT_LORA_RX | MESH_EVENT_LORA_TX_DONE | MESH_EVENT_CLI_RX | MESH_EVENT_HOUSEKEEPING | MESH_EVENT_GPS_ACTION | MESH_EVENT_TX_DRAIN | MESH_EVENT_RTC_SAVE | MESH_EVENT_INIT_ADVERT)
+#define MESH_EVENT_ALL           (MESH_EVENT_LORA_RX | MESH_EVENT_LORA_TX_DONE | MESH_EVENT_CLI_RX | MESH_EVENT_MAINTENANCE | MESH_EVENT_GPS_ACTION | MESH_EVENT_TX_DRAIN | MESH_EVENT_RTC_SAVE | MESH_EVENT_INIT_ADVERT)
 
-/* Housekeeping interval - infrequent to preserve power savings */
-#define HOUSEKEEPING_INTERVAL_MS CONFIG_ZEPHCORE_HOUSEKEEPING_INTERVAL_MS
+/* Maintenance is deadline-driven, not periodic: after every pass the loop asks
+ * the mesh when its soonest pending deadline is (msUntilNextMaintenance) and
+ * arms a single one-shot wake for exactly that moment.  An idle repeater with
+ * nothing scheduled therefore sleeps until its next real deadline instead of
+ * waking on a fixed cadence.
+ *
+ * MAINTENANCE_BACKSTOP_MS bounds that: it caps how long we will go without a
+ * pass even when everything reports idle, so a deadline that is missed or
+ * mis-reported degrades to the old behaviour instead of wedging.
+ * MAINTENANCE_MIN_MS floors it, so an item that is due-but-blocked (radio
+ * mid-packet, duty-cycle sleep window) re-arms shortly rather than spinning. */
+#define MAINTENANCE_BACKSTOP_MS CONFIG_ZEPHCORE_MAINTENANCE_BACKSTOP_MS
+#define MAINTENANCE_MIN_MS      50
 
 /* Event object for mesh loop */
 static struct k_event mesh_events;
@@ -128,15 +152,16 @@ K_MSGQ_DEFINE(cli_cmd_queue, sizeof(struct cli_cmd_line), 4, 4);
 
 /* Work items for event-driven processing */
 static void cli_rx_work_fn(struct k_work *work);
-static void housekeeping_timer_fn(struct k_timer *timer);
+static void maintenance_timer_fn(struct k_timer *timer);
 static void tx_drain_work_fn(struct k_work *work);
 static void initial_advert_work_fn(struct k_work *work);
 K_WORK_DEFINE(cli_rx_work, cli_rx_work_fn);
 K_WORK_DELAYABLE_DEFINE(tx_drain_work, tx_drain_work_fn);
 K_WORK_DELAYABLE_DEFINE(initial_advert_work, initial_advert_work_fn);
 
-/* Housekeeping timer for periodic tasks (noise floor calibration, etc.) */
-K_TIMER_DEFINE(housekeeping_timer, housekeeping_timer_fn, NULL);
+/* One-shot maintenance wake, re-armed after every loop pass (see
+ * arm_maintenance_wake).  Not periodic — the period is the deadline. */
+K_TIMER_DEFINE(maintenance_timer, maintenance_timer_fn, NULL);
 
 /* Forward declarations */
 #ifdef ZEPHCORE_LORA
@@ -244,11 +269,43 @@ static void process_cli_commands(void)
 }
 #endif
 
-/* Housekeeping timer callback - signals event to wake mesh loop periodically */
-static void housekeeping_timer_fn(struct k_timer *timer)
+/* Maintenance deadline expired — wake the mesh loop to run the pass. */
+static void maintenance_timer_fn(struct k_timer *timer)
 {
 	ARG_UNUSED(timer);
-	k_event_post(&mesh_events, MESH_EVENT_HOUSEKEEPING);
+	k_event_post(&mesh_events, MESH_EVENT_MAINTENANCE);
+}
+
+/* Ask the mesh for its soonest pending deadline and arm the one-shot for it.
+ * Called after every loop pass, whatever woke us: any event may have created
+ * or cleared a deadline (a queued advert, a CLI tempradio command, a CAD probe
+ * that just ran), so the schedule is recomputed from scratch each time rather
+ * than tracked incrementally.
+ *
+ * The backstop below is an unconditional CEILING on the wait, not a fallback
+ * used only when everything reports idle.  That means it must stay well above
+ * the shortest legitimate recurring deadline (the noise-floor sampler, and the
+ * CAD probe) or it becomes the effective period and the deadline scheduling
+ * buys nothing — see ZEPHCORE_MAINTENANCE_BACKSTOP_MS. */
+static void arm_maintenance_wake(void)
+{
+	uint32_t delay = MAINTENANCE_BACKSTOP_MS;
+
+#ifdef ZEPHCORE_LORA
+	if (repeater_mesh_ptr) {
+		uint32_t next = repeater_mesh_ptr->msUntilNextMaintenance();
+
+		if (next < delay) {
+			delay = next;
+		}
+	}
+#endif
+
+	if (delay < MAINTENANCE_MIN_MS) {
+		delay = MAINTENANCE_MIN_MS;
+	}
+
+	k_timer_start(&maintenance_timer, K_MSEC(delay), K_NO_WAIT);
 }
 
 #ifdef ZEPHCORE_LORA
@@ -398,9 +455,8 @@ static void repeater_event_loop(void)
 	/* Print startup banner (no prompt - Arduino style) */
 	cli_print("\r\n=== ZephCore Repeater ===\r\n");
 
-	/* Start housekeeping timer for periodic maintenance tasks */
-	k_timer_start(&housekeeping_timer, K_MSEC(HOUSEKEEPING_INTERVAL_MS),
-		      K_MSEC(HOUSEKEEPING_INTERVAL_MS));
+	/* Arm the first maintenance wake; every pass below re-arms it. */
+	arm_maintenance_wake();
 
 	for (;;) {
 		/* Wait for any mesh event - blocks until signaled */
@@ -436,8 +492,8 @@ static void repeater_event_loop(void)
 		}
 #endif
 
-		/* Periodic housekeeping — maintenance + display refresh */
-		if (events & MESH_EVENT_HOUSEKEEPING) {
+		/* A maintenance deadline came due — run the pass. */
+		if (events & MESH_EVENT_MAINTENANCE) {
 #ifdef ZEPHCORE_LORA
 			/* Radio maintenance: noise floor calibration, AGC reset,
 			 * RX watchdog.  Separated from loop() so these never run
@@ -452,16 +508,21 @@ static void repeater_event_loop(void)
 			}
 #endif
 
+#if ZEPHCORE_HAS_UI
 			ui_set_clock(rtc_clock.getCurrentTime());
 
 #ifdef ZEPHCORE_LORA
 			/* Refresh live radio state (noise floor, TX power
-			 * reduction, RX/TX mode, packet counters). */
+			 * reduction, RX/TX mode, packet counters).  Headless
+			 * repeaters compile this out entirely — every ui_set_*
+			 * below it is a weak no-op there (ui_headless_stubs.c),
+			 * so the whole block was pure work for nothing. */
 			refresh_repeater_ui_radio_state();
 
 			/* Battery is now refreshed lazily from ui_pages_render() with
 			 * a 30 s freshness guard — no periodic ADC fire here. */
 #endif
+#endif /* ZEPHCORE_HAS_UI */
 		}
 
 		/* Off-main RTC write request (gps_fix_callback runs in modem_chat
@@ -476,6 +537,11 @@ static void repeater_event_loop(void)
 			}
 #endif
 		}
+
+		/* Re-arm for the soonest deadline this pass left behind.  Done for
+		 * every wake, not just maintenance ones: a CLI command, an inbound
+		 * packet or a GPS fix can all create or clear a deadline. */
+		arm_maintenance_wake();
 	}
 }