Verified on the board#

What the ESP-Mosaico board showed for OpenCarMind EdgeOS 2.0.0 and 2.1.0. VERIFIED means a board log line shows it. The full table is in the dropdown at the end.

Two board passes ran. The first on OpenCarMind EdgeOS 2.0.0 4aeff43 (SHA-256 a225125d27f3906c...), the second on 06ddf1d, which adds the pairing change, the 16 px text and the model names (merged image SHA-256 4cc402a76a53edc2...). The release image in this branch is rebuilt from its tip, whose Mosaico code equals 06ddf1d apart from the Home card fix in ce58ca7. VERIFIED means a board log line shows it. Some rows ran on a scratch build of the 2.0.0 code with one test switch on, and the row says so. A scratch build is never shipped. Host tests pass for every part: make os (41,635 checks at 4aeff43, 41,719 with the pairing change), make pair, make ppa, make uimetrics and the source audits in tests/host/os_audit.sh.

Of the 28 rows from those two passes, 21 are VERIFIED (6 of those on a scratch build), 3 PARTLY and 4 NOT: not run yet, or switched off in this release. The three figures below show the main results. A third pass is running on build 13757fd, which adds the AI services and pairing safety. Its 8 rows are at the end of the table: 4 VERIFIED so far, 3 being checked, and 1 for the intrusion detector, which is not in the firmware yet. The full table is folded under the figures.

Boot and runlevels on the board. The release boot on 4aeff43, in seconds after power-on: running at 4.3 s with Wi-Fi held, boot health good at 68.0 s with Wi-Fi released and 128.9 KB of internal RAM free, the Wi-Fi driver up at 88.1 s, the clock synced at 98.0 s, the driver down at 106.5 s. If health is bad, on a scratch build forced bad: safe mode, Wi-Fi held, and Sync clock now still starts it by hand. The runlevels seen in the tests: Parked. Driving on a forced drive. Saving when the battery was hot at 34.2 °C, back to Parked when it cooled to 33.7 °C. Safe mode on a forced bad health check. Refused while Driving: the USB door answered ERR busy, and the screen said Not now: a drive is running for the clock sync and Format.

The release boot holds Wi-Fi until the health check says the board is good, then joins, syncs the clock and stops the radio again. The runlevels in the lower half come from forced tests on scratch builds.

Degradation ladder and battery protection, both forced tests on scratch builds. Ladder with extra model load, in seconds after power-on: UI slowed at 110.6 s after 2 misses, learner paused at 122.6 s after 6, model paused at 134.6 s after 10, then one rung back at a time: model back at 214.6 s, learner back at 278.6 s, normal at 348.6 s. Battery protection at test limits: at boot on USB, 34.2 °C sheds load into Saving with Wi-Fi off, the learner paused and the screen dimmed, and 33.7 °C returns to Parked. 4,188 mV from Parked stops the board as a charger fault with the logs flushed and the storage stopped. With the limit lowered to 26 °C, 28 °C stops it as critical, with no power-off. The real limits of 45 °C, 60 °C and 4.25 V were not reached.

The ladder drops the screen rate first and the model last, and climbs back one rung per minute without misses. The battery figures are test limits. The real 45 °C, 60 °C and 4.25 V were not reached.

Internal RAM per phase, the lowest free internal RAM in each of 9 phases on a debug build of 4aeff43, in KB: first minute before Wi-Fi 124.4, first Wi-Fi round with the adapter search 81.0, adapter search alone 110.3, adapter search and Sync clock now together 80.9, Sync clock now alone 81.0, Wi-Fi search 86.7, theme switch 110.2, switch to E.V 100.0, launcher slide 110.0, all above the 60 KB target. Below it, the 28 rows of the board table: 21 verified, 3 partly, 4 not run or off in this release.

Every phase stays above the 60 KB target. The lowest is 80.9 KB, with an adapter search and a clock sync running at once.

Full board results table (all 36 rows, with the build each one ran on)

Part

Result

What the board showed

Build

Boot and name

VERIFIED

OpenCarMind EdgeOS 2.0.0 running at 4.3 s after a power-on, 0 USB drops in 185 s

4aeff43

Boot health, good

VERIFIED

boot health good | model self-test 1, flash FAT 1 (0 repaired), NAND 1, first UI tick 1365 ms, touch ok, 0 misses, 0 stalls, schedulable, runlevel Parked at 68.0 s

4aeff43

Boot health, bad

VERIFIED

boot health bad: safe mode, then Wi-Fi held: boot health bad. Sync clock now still started Wi-Fi by hand

scratch, forced bad

Wi-Fi boot hold

VERIFIED

held at 4.3 s with no driver, released at 68.0 s with 131,975 B of internal RAM free, driver up at 88.1 s and down again at 106.5 s

4aeff43

No addresses in the log

VERIFIED

no access-point address and no board address in the boot and Wi-Fi log

4aeff43

Model admission, admitted

VERIFIED

C 4248 us (self-test 2124 us x 2.0) on core 1, R: ml 4.2/1000 ms, margin 995.8 ms (ml): admitted

4aeff43

Model admission, refused

VERIFIED

the boot load logged REFUSED, and a switch to E.V got ERR CarMind E.V refused: ml would miss its deadline while R.E.V stayed

scratch, safety factor raised from 2 to 2,000

Service contract

VERIFIED

service R.E.V learner: job rev, T 1000 ms, D 30000 ms, stall 60000 ms, CPU 20 per mille in 30 ms bursts, prio 2 core 1

4aeff43

Runlevels and refusals during a drive

VERIFIED

runlevel Driving (active 0x08) and back to runlevel Parked. The USB door answered ERR busy: a drive is running, try again when parked. The screen showed “Not now: a drive is running. Try again later.” for the clock sync and Format

scratch, forced drive

Degradation ladder

VERIFIED

events “UI slowed (2 misses)” at 110.6 s, “learner paused (6 misses)” at 122.6 s, “model paused (10 misses)” at 134.6 s, then one step back at a time to “normal” at 348.6 s. One model load in the whole run

scratch, extra model load

Battery protection

VERIFIED, test limits

“Battery hot (34.2 C on USB): load shed” with Wi-Fi off, the learner paused and the screen dimmed, runlevel Saving, then “Battery cooled (33.7 C)” and Parked, and “Battery over-voltage (4188 mV): charger fault, stopped”. The real limits (45 °C, 60 °C, 4.25 V) were not reached

scratch, test limits

Desktop app

VERIFIED

connect with fw=EdgeOS-2.0.0, a wrong token refused, a model switch both ways, two files read at their listed sizes

4aeff43

Internal RAM above 60 KB

VERIFIED

lowest 82,883 B over 9 phases (see What it costs)

debug build of 4aeff43

PPA and 2D-DMA

PARTLY

on at boot, failures dma 0 ppa 0 in all 18 reports. A clean A/B on the same build: composing one slide frame takes 5.5 ms with the accelerator and 8.4 to 9.0 ms without, but a whole slide runs at 8.4 to 9.0 fps with it and 11.7 to 12.5 fps without, and its slowest frame is about 45 ms longer. Under investigation

06ddf1d

Resets through sys_reset

VERIFIED

the last run ended with sys_reset(REBOOT: door REBOOT) after a door REBOOT, and the last run ended with sys_reset(door: USB door) after a download-door entry

4aeff43, 06ddf1d

Feature gate

PARTLY

the gate starts and Smart swipe’s warm-up runs. No verdict yet

4aeff43

Version on screen

VERIFIED

OpenCarMind EdgeOS 2.0.0 in the boot line, on the Board page and in Settings, and fw=EdgeOS-2.0.0 over USB

4aeff43, 06ddf1d

10-minute soak

VERIFIED

10 min 02 s on the release: 0 restarts, 0 USB drops, 0 UI misses, every job within its deadline. The Wi-Fi round ran on the next boot: clock synced, internal RAM 90,959 B free at its low

06ddf1d

Critical battery stop

VERIFIED, test limit

with the limit lowered to 26 °C: Battery too hot (28 C): stopped and logs flushed, storage stopped, no power-off. The real 60 °C was not reached

scratch, lowered limit

Where the gauge temperature comes from

VERIFIED

the gauge chip’s own sensor: Temperature() equals InternalTemperature() in all 25 readings (28.1 to 31.4 °C). The board has no thermistor on the cell, so this is the temperature next to the cell, not of the cell

debug builds

Board page “lowest” after the fix

PARTLY

the Board page reads 108 KB, the all-internal figure (110,631 B) and no longer 32 KB low. Not shown after a Wi-Fi round

debug build of 06ddf1d

Wi-Fi with no known network in reach

NOT

not run

Going back to mossREV 1.2.0

NOT

not tried

A/B rollback, UPS, GPIO57 power-off

NOT

off in this release

Pairing: a weak adapter is ignored

VERIFIED

over the soak and the next boot: 0 connection tries, 0 pairing questions and 0 Driving spells, automatic adapter search: not paired, strong signal only

06ddf1d

Pairing: offering and saving your adapter

NOT

needs the owner’s adapter next to the board

Model names and 16 px text on screen

VERIFIED

37 board captures: “active-estimator” and “ev-frozen” on the model pages, no text under 16 px, nothing in a 60 px corner. One overlap on Home (fixed in ce58ca7, host-checked)

06ddf1d

The download door in every state

VERIFIED

the 1200-baud door entered the ROM download mode on the release, in Saving, during a forced drive and after a battery safety stop

4aeff43, 06ddf1d, scratch

Boot health, with the AI services

VERIFIED

rated good at 68.9 s

13757fd

Wi-Fi boot hold, with the AI services

VERIFIED

Wi-Fi released after the good verdict with 130,623 B of internal RAM free

13757fd

AI worker admission

VERIFIED

admitted on core 1 with R 2.00 ms against its 1,000 ms deadline, a margin of 998 ms

13757fd

Battery estimator self-test

VERIFIED

570 µs against its 2 ms budget

13757fd

Battery SoC and SoH on the Board page

BEING CHECKED

built, being checked on the board. The parameters are placeholders until the cell’s curve is fitted

13757fd

Ladder rung “AI services slowed”

BEING CHECKED

built, being checked on the board

13757fd

Pairing safety: fingerprint, no silent re-pair, clock jump guard

BEING CHECKED

built, being checked on the board

13757fd

Network intrusion detector

BEING CHECKED

in the firmware since 2.1.0 as the network guard, see the 2.1.0 rows

73db232

2.1.0: boot health and AI admission

VERIFIED

health good at 68.8 s; bat, banom and ids admitted on core 1, R 5.00 ms against 1,000 ms, margin 995 ms; self-tests (fastest of 5) 229, 19 and 58 µs, the IDS self-test replays 64 vectors

73db232

2.1.0: 10-minute soak

VERIFIED

640 s, 0 restarts, 0 USB drops, 0 UI misses in 64 windows, 0 scheduler misses; bat 536, banom 537 and ids 108 runs, 0 deadline misses, worst response 12.9, 13.0 and 11.9 ms

73db232

2.1.0: network guard in its shadow

VERIFIED

scan score 0 to 34 of 256 over 112 windows (22,585 adverts), 0 blocks, 0 “would block”; the first Wi-Fi scan (15 networks) scored 256, at the threshold, not over it; no pairing prompt; the clock sync was accepted

73db232

2.1.0: internal RAM, lowest per phase

VERIFIED

debug build of 73db232 with the RAM monitor: 122,795 B at the boot verdict, 109,643 B as the AI services and the guard start, 78,227 B over a Wi-Fi round with the Health app open (13757fd round: 81,531 B), 108,327 B on Board and Settings; all above 60 KB. The AI worker’s stack never used 2,648 of its 4,096 B

73db232

2.1.0: Health app on the board

VERIFIED

3 pages in both themes captured from the board, no overlap, 16 px text

73db232

2.1.0: Health in plain words

VERIFIED

real states on the board (battery learning, guard learning this place, AI on time) and, on a debug build that puts a state into the Health view only, battery good, worn, three anomaly reasons, guard watching, blocking and adapter mismatch, AI slowed, in both themes. The banner with Open Health opened the right page, a swipe closed it. The first Wi-Fi scan after a start is no longer scored (0 scans, “no data”, 73db232 showed 290 of 256)

81017a5

2.1.0: battery anomaly detector

BEING CHECKED

admitted, in its 24 h warm-up

73db232

2.1.0: SoH regressor, counted SoH and DCIR

NOT

the cell stayed full at 0 mA, so no charge cycle, no capacity sample and no DCIR sample yet (0 of 10 reference cycles)

73db232