Emergency battery schedule, 28 October 2026

Native checks and nine packaged firmware builds pass, using f4f945324, plus the ESP32 USB recovery fix in be6d636c7. SolarXiao 33S, RAK3401 and Heltec V4.3 passed physical sleep, override, automatic wake and reboot-persistence trials. Current savings and operation with a real battery at 5% were not measured.

The later 1.18.0.2 bench campaign found and fixed nRF52 USB reconnect starvation during automatic emergency sleep. Attached USB hosts now use CPU idle slices of at most 100 ms while LoRa stays asleep. The measurements below retain their original source.

Behavior

Supported infrastructure images enter automatic emergency mode at 5% or less, and recover at 7% or more, using a voltage-derived battery estimate sampled every 30 seconds. Invalid readings neither activate the mode nor clear an existing low-battery state. Existing hard low-voltage shutdown still applies.

Normal work runs during the first 15 minutes of each UTC hour. Outside that window, LoRa sleeps and normal forwarding, sensor and display work pauses. With an implausible clock timestamp, the same cycle follows monotonic uptime from boot. A plausible timestamp still needs synchronization to UTC. Brief maintenance wakes, at intervals of 10 seconds or less, service commands, clock and watchdogs. Active updates and unfinished critical work defer sleep.

get power.emergency
set power.emergency on
set power.awake 30
set power.awake 0
set power.emergency auto

on persists manual emergency mode; auto restores the default automatic policy. Entry allows five seconds for the reply. power.awake grants a runtime override of 0–10080 minutes; zero cancels it and reboot clears it. LoRa commands require an awake window. Local USB can wait for the next maintenance wake. Status separates the desired window from actual radio ownership.

nRF52 retains state through timed RTOS delay and idle sleep; ESP32 uses timer-woken light sleep. Neither path uses a rebooting deep-sleep cycle. ESP32 keeps an attached native USB console usable with short CPU idle waits instead of peripheral-stopping light sleep; detached operation uses timed light sleep. Wireless services restore their previous live subset and preserve manual wireless-off settings. RAK3401 places its amplifier in standby and restores it before radio wake, without cutting the shared supply rail.

Forwarding defaults move from 10/20% to 15/25%. The exact old pair migrates from a v1 record; other pairs, enabled controls and voltage endpoints survive. A v2 record preserves an explicitly selected 10/20 pair. Companion and profiles that omit battery-forwarding controls retain their existing feature scope.

Checks

85 native tests across eleven suites pass: policy/CLI 7, runtime 13, platform backends 3, OTA work 1, battery forwarding 20, build-role policy 8, wireless 4, logging/sleep 3, sleep deadlines 3, CPU clock 9, and bench harness 14. The OTA test compiles three configurations and executes the actual guards for all three roles. Coverage includes rollover, failed persistence, reply holds, active transfers, bounded serving holds, wireless restoration failures and prior Tracker ownership.

Packaged firmware	Capability checks	Available / required RAM, bytes
SolarXiao 33S Repeater Full	11/11	151,348 / 82,784
SolarXiao 33S Repeater Reduced	9/9	152,528 / 80,736
RAK3401 Repeater Full	14/14	80,124 / 75,920
RAK3401 Repeater Reduced	12/12	82,648 / 73,872
Heltec V4 Full MQTT Repeater	16/16	240,112 / 238,732
Heltec V4 Full MQTT Repeater, USB fix	16/16	240,112 / 238,732
SolarXiao 33S Room Server Full	7/7	157,108 / 67,328
SolarXiao 33S Room Server Reduced	5/5	158,264 / 65,280
Heltec V4 Sensor Full	9/9	271,872 / 150,592

Memory admission and packaged update-path checks pass. These are linked capacity estimates, not measurements of peak runtime heap. PlatformIO ran sequentially.

The safe machine-readable report records ELF, package and manifest hashes. Package hashes were independently compared with every corresponding memory-manifest entry.

Integration checks

Eight further normal builds at 8c2ddc9ac pass after the concurrent USB-mode and MediaMesh changes. Constrained profiles use shorter human validation replies and help descriptions; commands, returned values, protocol acknowledgements and destructive-operation warnings remain. Runtime timers and flags are grouped to remove alignment padding. Saved-record layouts, features and filesystem capacity are preserved.

Firmware	Linked flash / limit, bytes	Available / required RAM, bytes
RAK4631 Repeater Full	687,512 / 815,104	83,156 / 83,136
RAK4631 Repeater Reduced	620,020 / 815,104	85,696 / 81,088
Wio E5 Mini Repeater	245,260 / 245,760	43,880 / 32,288
RAK4631 Full Companion	712,092 / 712,704	81,284 / 74,576
RAK3x72 USB Companion	228,924 / 229,376	24,448 / 17,408
Wio E5 Mini USB Companion	228,816 / 229,376	24,424 / 17,408
Wio E5 USB Companion	228,892 / 229,376	24,456 / 17,408
TinyRelay USB Companion	228,992 / 229,376	24,448 / 17,408

Linked flash counts exclude packaging and alignment overhead; normal linker and package checks also pass. The Mini retains its 16 KB filesystem. Native checks compare the reviewed reply substitutions and unchanged C++ control flow. The final RAK Companion ELF passes both RFC8032 vectors with a 1,904-byte signing frame under the 2,048-byte limit, plus four OTA workspace lifetime checks. These Companions do not gain the infrastructure emergency-sleep feature.

Physical trials

Each board entered manual radio sleep, woke for a real one-minute override, returned to sleep when it expired, and retained manual mode after reboot. Two accelerated hourly boundaries then produced queued adverts on an independent RAW receiver before any further USB write to the target.

Target	Independent receiver	Passive capture intervals
SolarXiao 33S	RAK3401	31.586 / 31.587 s
RAK3401	SolarXiao 33S	31.586 / 31.583 s
Heltec V4.3 OLED	RAK3401	21.730 / 21.677 s

The clock was set near the hour boundary; these intervals are evidence of autonomous RF resumption, not a measurement of the CPU's exact wake latency. The V4's initial build lost its USB console during off-window light sleep. The corrected build kept USB responsive while RF slept and passed the whole trial. WiFi/NTP was temporarily stopped so it could not reset the test clock.

All identities, existing settings and present clocks were checked after cleanup. SolarXiao retains the tested Full Repeater in automatic mode with no awake hold. RAK3401 has its original unified Repeater; V4 has its original Full Companion and verified original NVS/filesystem. The Pi collector is active with its original configuration and serial ownership.

Limits

Trials used manual entry and USB power. They do not qualify real 5% battery entry, current draw, or a continuous 45-minute sleep interval. V4 exercised the USB-held CPU idle path; detached light-sleep current remains unmeasured. Solar LTC4162 telemetry was unavailable, and the self ADC is not carrier state of charge. Room and Sensor hooks have native/build coverage, not physical trials. Sustained RF delivery and maximum runtime load need separate evidence.