bench kit: reproduce the LifeTrac radio bench from GitHub (setup, deploy, legs, guards) by dorkmo · Pull Request #142 · OpenSourceEcology/LifeTrac

dorkmo · 2026-10-10T19:52:53Z

Stacked on #140. Merge #140 first; GitHub then retargets this PR to main.

A reproducibility audit (2026-10-10) found that someone with a fresh clone of main could rebuild the radio firmware byte-identically and pass every test, but could not set up two boards and fly a leg the way RS-12/RS-13 did. This kit closes that gap. Docs and scripts only: no firmware or product-code change, and nothing in it was run against the boards (radios kept asleep).

Entry point: bench_tools/BENCH_QUICKSTART.md.

What's in it

Entry point and PC setup.
- BENCH_QUICKSTART.md takes a newcomer from PC setup through bench.env, boards, firmware, deploy, a session, a leg, analysis, parking and board capture.
- PC_SETUP.md lists the exact toolchain (Arm GNU 12.2.MPACBTI-Rel1 gives byte-identical L072 builds), Git core.longpaths, the PowerShell execution policy, adb, uuu 1.5.243 and requirements-pc.txt.
- The repo and DESIGN-CONTROLLER READMEs link to the quickstart.
Board setup.
- BENCH_SETUP.md takes two new boards to bench-ready.
- provision_bench_board.sh <base|tractor> [--check] is idempotent and does:
  - install the sudoers drop-in;
  - mask the base's compose-apps units (otherwise /var/lib/docker is wiped every 60 s);
  - optionally pin the NIC;
  - turn the tractor's WiFi off (the disable-wifi.service unit from the capture, the wpa mask, and a root-only backup of the WiFi profiles before deleting them);
  - run camera provisioning;
  - preload the offline tractor's images.
- T3a pins the OS image by URL and sha256 (934.tar.gz; 674.tar.gz is hosted too) and fixes the mislabeled 05-13 row.
Deploy and images.
- deploy_base.sh does the git archive, upload, extract and DEPLOYED_FROM.txt, creates secrets only when they are missing, builds lifetrac-v25, and starts only the broker. It never starts lora_bridge.
- build_tractor_image.sh builds on the base, saves to the archive, loads on the tractor, and keeps the old image as :previous.
- The lock files hold the flown wheel versions, behind an opt-in LOCKED=1 build argument; default builds and CI are unchanged.
- DEPLOY.md documents both scripts.
Settings and leg scripts.
- bench.env.example, lib/bench_env.sh and lib/env_file.sh define one set of names with one precedence order: a non-empty environment variable, then bench.env, then the defaults.
- bench_tools/legs/ holds parameterised copies of the RS-13 scripts. There are no hardcoded paths, serials or passwords. leg_prep refuses a DTS leg without today's carrier and needs PC_HOST. The scripts have cleanup traps, and there is a generic power_up_guard.sh (reads and parks the radios, never transmits).
- The RS-13 record's copies stay unchanged as history.
Harness, flash and firmware.
- run_live_radio_monitor.ps1 now:
  - refuses a profile-2 leg on the 915.000 MHz default unless given -ForceFrfHz or -AllowDefaultCarrier;
  - rejects misspelled parameters ([CmdletBinding()]);
  - starts the PC broker with py -3;
  - gives a clear "adb not on PATH" error.
- The flash wrapper defaults to REVIVE_MODE=reboot and checks its staging before it runs. flash_l072.sh stages, checks and flashes on --go.
- FLASH_RUNBOOK gains a "first flash from a factory board" section.
- murata_l072/README gets a Building section with the expected md5s.
- A non-blocking CI job (l072-bench-md5.yml) checks the bench and production md5s with the pinned toolchain.
Hygiene.
- .gitattributes makes board-side files check out with LF line endings and marks binaries; no committed file changes.
- The WiFi SSID and password are out of the seven repo-root scripts (now env vars), and the VS Code tasks that carried the password are removed.
- The doc errors the audits found are fixed (board labels, a renamed script, the handheld hardware, an x8h7 overwrite warning, the misnamed L072 binaries).

Review

Six implementers in isolated worktrees, one owner per file. All six merged cleanly; afterwards I reconciled the bench.env names.
Three reviewers: a newcomer walkthrough, script safety, and mechanical checks. They found no blockers, 13 majors and about 30 minors.
A docs pass and a scripts pass fixed them. A verifier merged both and confirmed all 13 majors fixed.
Checks on the final tree: bash -n (18 scripts plus the 7 WiFi scripts), the PowerShell parser (3 files), strict JSON for tasks.json, a markdown link and anchor check (0 broken), and base_station tests at 1537 passed, 2 skipped.

Not verified on hardware (by rule — radios stayed asleep)

provision, deploy, build and flash against real boards;
the ssh transport;
sudo --preserve-env on LmP;
the new cleanup traps;
leg_prep's PC_HOST auto-detection;
the WiFi-profile backup.

All of these pass syntax checks and code review only. The first real use should be a supervised dry run.

Open, outside this PR

Change the WiFi passphrase on the access point. The old value is still in public git history; the docs now say so.
Licence: the repo has no LICENSE file and redistributes Arduino's OOTB image (foundries_python.tar). OSE needs to decide.
Remaining known minors:
- the EVIDENCE_DIR default;
- the harness's chmod 0777 on the staging folders;
- the sudo password passed on command lines;
- widening the carrier guard to profile 0.

🤖 Generated with Claude Code

Board scripts, systemd/udev units, mosquitto/openocd configs, Dockerfiles, compose/YAML/TOML, Python, *.env(.example) and requirements*.txt now check out with LF on every OS, so a Windows clone (core.autocrlf=true) and `git archive` no longer hand the X8 boards CRLF files. *.ps1/*.bat check out CRLF (text=auto). Firmware images, archives, wheels, objects, images and PDFs are marked binary. Renormalize check (`git add --renormalize .`, not committed): - against the existing autocrlf working tree: 108 tracked files would change (60 *.pdf, 43 *.hex, 3 bench-evidence *.bin, 1 *.zip, bom.csv). These are NOT index CRLF: autocrlf had guessed those binaries were text and mangled them on checkout, and re-adding the mangled copies would have corrupted the index. Not committed. - after rewriting the working tree under the new rules: 1 file (DESIGN-STRUCTURAL/drawings/generated/bom.csv, which is stored with CRLF but is not covered by these rules; it only shows up because core.autocrlf=true). 0 covered files renormalize: the index already stores every covered file with LF, so this commit changes checkout behaviour only. The refresh recipe is in the file header. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…elled params run_live_radio_monitor.ps1 (bench kit, reproducibility audit): - A profile-2 (DTS) leg with -ForceFrfHz 0 is refused before anything touches a board; the profile default is 915.000 MHz, the RS-11.6 emitter channel every RS-13.1 DTS leg flew on by accident. New -AllowDefaultCarrier switch for a deliberate control leg (recorded in params.txt). Legs that pass -ForceFrfHz are unchanged. - [CmdletBinding()]: a misspelled parameter (e.g. -DurationSeconds) is now a ParameterBindingException instead of being silently ignored. Verified by binding the parsed param block in a synthetic scriptblock (no catch-all ValueFromRemainingArguments exists). - The PC MQTT broker and the -TxFeed host publisher start with -PythonExe (default `py`, run as `py -3`) instead of a hardcoded per-user interpreter path; the amqtt dependency is documented and an advisory re-probe warns when the broker did not come up. - Warn when -HostIp is not one of this PC's addresses (default unchanged). - Make /tmp/lifetrac_p0c writable before pushing 08_boot_user_app.cfg and warn if it did not land (the tractor SWD reset needs it). - Fix the "$HostIp:1883" messages, which printed empty under StrictMode. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Pins the Python packages the bench PC's own tools import (amqtt for the harness's PC broker, paho-mqtt, numpy, opencv-python-headless, Pillow, cryptography) at the versions installed on the PC that flew RS-12/RS-13, with a per-package note of which scripts need it. amqtt was declared nowhere before. All six pins match the bench PC's installed versions. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

The seven repo-root WiFi helpers (connect_wpa.sh, do_wifi.sh, setup_5star_wifi.sh, wifi_connect.sh, wifi_connect_fast.sh, wifi_persist_and_reboot.sh, wifi_rescan_connect.sh) carried the bench WiFi SSID and password as literals. They now require LIFETRAC_WIFI_SSID and LIFETRAC_WIFI_PSK from the environment (POSIX ${VAR:?} guard: the script stops before touching the radio if either is unset) and carry a header saying credentials must never be committed. Behaviour is otherwise unchanged. connect_wpa.sh writes /tmp/wpa.conf under umask 077 because the file holds the passphrase. .vscode/tasks.json had the same password inline in "Rescan then connect"; it and the five tasks that push and run these scripts now take ${env:LIFETRAC_WIFI_SSID} / ${env:LIFETRAC_WIFI_PSK} from the PC environment, passed through `sudo -S env ...` (sudo drops the caller's environment), using the quoting the SSID already used there. Not run: no adb on the bench during this change. The old values remain in git history. The operator is rotating that password; after rotation the stored '5star' NetworkManager profile on the tractor should be deleted (BENCH_BOARDS.md follow-ups). The SSID alone still appears in a board-state capture (bench-evidence/board_state_2026-10-10/tractor/20_network.txt), left as evidence. Checks: bash -n and sh -n on all seven; each exits 1 with the guard message and no output when the variables are unset; the connect_wpa.sh heredoc was exercised in a sandbox copy with fake values containing ', $ and backticks; .vscode/tasks.json parses as strict JSON; git grep finds the password in no tracked file. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

BENCH_QUICKSTART.md is the single map a newcomer follows to reproduce the RS-12/RS-13 radio bench: hardware, PC setup, boards, L072 firmware (expected md5 0c1bb0a9), board software, bench.env, the per-session power-up guard, staging and DTS carrier spot-check, the RS-13 harness line, analysis, park and power-down, board-state capture, a symptom table, and the GO rule. PC_SETUP.md pins the Windows 11 bench PC: Git for Windows with core.longpaths, py -3 + requirements-pc.txt / requirements-dev.txt, the exact Arm GNU Toolchain 12.2.MPACBTI-Rel1, WinLibs mingw32-make (PowerShell only), adb with MSYS_NO_PATHCONV, uuu 1.5.243, optional arduino-cli / J-Link, the Firefox scene rule (never --kiosk) and the local test command with CI's dummy fleet key. Links to the other kit files use the shared contract names (BENCH_SETUP.md, DEPLOY.md, bench.env.example, legs/README.md, requirements-pc.txt, ...). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

BENCH_RUNBOOK.md: point newcomers at BENCH_QUICKSTART.md and the other kit docs near the top; link PC_SETUP.md from the toolchain fact and legs/stage_boards.sh from prep step 1; add the bench web UI on the base recipe (lifetrac-v25:latest from /tmp/lifetrac_strict, uvicorn on 8090, bench-only PIN, LIFETRAC_MQTT_HOST=127.0.0.1, MQTT tap because web_ui logs nothing, Auto arm/disarm trap, teardown), as recorded in the F10/F11 sessions and leg V; add the harness parameter-name trap (misspelled names used to fall into $args; now refused; read params.txt). README.md and DESIGN-CONTROLLER/README.md: one short pointer each to the radio bench quickstart. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…ash_l072.sh - full_flash_pipeline.sh and bench_tools/run_flash_bench.sh now default REVIVE_MODE to `reboot` (was `full`). Every flash since 2026-09-12 passed reboot by hand; the full path re-inserts x8h7 and OOPSes the base's 6.1.24 kernel. `REVIVE_MODE=full` stays selectable. revive_bridge.sh accepts `reboot` as an alias of reset_run_only (a standalone call with the wrapper's environment can no longer fall into the module reload) and warns before a module reload on a 6.x kernel. - run_flash_bench.sh: password, home dir from BENCH_SUDO_PW / BENCH_HOME (defaults fio, /home/fio); FLASH_VERIFY_ONLY passes through sudo; a preflight refuses, before any hardware is touched, when a pipeline file or the image is missing/empty or a .sh/.cfg has CRLF (tmpfiles ages /tmp at 5 d); FLASH_PREFLIGHT_ONLY=1 stops after that check. - bench_tools/flash_l072.sh (new, PC side): FLASH_RUNBOOK sections 1-2 as a script. Serials, base host/user/key, sudo password and REVIVE_MODE come from bench_tools/bench.env (or the environment) with the current bench values as defaults. Stages LF-clean and runs the board preflight; flashes only with --go (a flash is a radio-on event). Tested with stub adb/ssh/scp only; never run against a board. - bench_tools/wdog_regs.py: the board-only WDOG1 register reader recovered from the 2026-10-04/10 captures, code unchanged, with a header (what it reads, read-only, run inside the docker image with --privileged -v /dev/mem:/dev/mem). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…udit - setup_x8_mqtt_link.ps1 (comment only) and ADB_TIPS_AND_TRICKS.md 3.1: the 2026-05-24 `adb reverse` failure on LmP 4.0.11-934-91 was on the base 2D0A, not 2E2C. 934-91 has only ever run on 2D0A (2E2C runs the as-shipped 4.0.3-674-88), and the 05-24 Phase C plan issued `adb -s $base reverse` with $base = 2D0A (uname 6.1.24 = 934-91). - HC-02_linux_bridge_full.md: board1_healthcheck.sh was renamed to x8_max_carrier_healthcheck.sh in 97accdc (2026-05-15); the procedure and helper link now use the existing file. - FIRMWARE_UPDATES.md: the handheld is an Arduino MKR WAN 1310 (SAMD21, firmware/handheld_mkr, FQBN arduino:samd:mkrwan1310), not a Portenta H747. Heading, artifact name and bootloader entry (double-tap RESET) corrected. tools/flash_handheld.py does not exist yet, so it is now marked as planned, with arduino-cli as the stopgap. - BUILD-CONTROLLER/05_firmware_installation.md Step 2: a warning box. Uploading tractor_h7 as "Portenta H7 (M7 core)" overwrites the X8's stock x8h7 bridge firmware (as happened on 2026-05-04), and the current bench does not use tractor_h7. The step itself is unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

x8_lora_bootloader_helper/mlm32l07x01.bin carries the stock Murata/Arduino file name, but it is the LifeTrac "hello v0.2" test build. It is byte-identical to murata_l072/build_hello/hello_world.bin; md5 and strings are in the README. The new README says so and also identifies: - hello.bin: identical to murata_l072_hello/hello.bin; - board2_l072_flash_dump.bin: a 16,440-byte readback of L072 flash taken with dump_l072_flash.py in May 2026. It matches no tracked build and is not the factory image; - LifeTrac-v25/tools/mlm32l07x01.bin: the real stock MKRWAN AT firmware (ARD-078 1.2.3, 83,032 B); - the actual bench firmware: `mingw32-make bench`, md5 0c1bb0a9. The files are not renamed, because old scripts default to that path (run_flash_l072.sh). The README also points to FLASH_RUNBOOK.md and the bench_tools docs. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…defaults - Section 0: the exact toolchain for byte-identical builds (Arm GNU Toolchain 12.2.MPACBTI-Rel1, gcc 12.2.1 20230214) and the expected md5s (bench 0c1bb0a9..., production 589c1203..., both 24 860 B at bb4a207); other compilers build working firmware with different bytes. - Sections 1-2: the scripted form bench_tools/flash_l072.sh (bench.env driven, stages + preflights, flashes only with --go), the wrapper's new staging preflight, and REVIVE_MODE now defaulting to reboot (full still selectable). - Section 3: wdog_regs.py now lives at bench_tools/wdog_regs.py, with the docker command and the expected WCR=0x773d. - New section 5, first flash from a factory board: what the stock MKRWAN board carries and how the X8 reaches the ROM (PA11/PF4 via the X8's openocd, AN3155 19200 8E1); a read-only --verify-only first contact against tools/mlm32l07x01.bin; option-byte / readout-protection rules (never 0x73/0x82/0x92 or writes to 0x1FF80000; the 2026-05-14 wunprot brick); the 2026-05 lessons (H7-side failures are not L072 bricks, the falsified PWM bypass, J-Link on the DNP CN2 header as the last resort, the 05-14/05-17 board disagreement). Cites the AI NOTES and T2/T4/HC-03. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Replaces the short "Build" section: the Arm GNU Toolchain 12.2.MPACBTI-Rel1 (gcc 12.2.1 20230214) needed for byte-identical builds, with the Linux tarball URL and Arm's published sha256; mingw32-make from PowerShell only; the all/bench/check/size/clean targets (and the bench-after-clean gotcha); the expected md5s at bb4a207 (bench 0c1bb0a9..., production 589c1203...); that Arduino's bundled gcc (build.ps1 picks it first) and Ubuntu's gcc-arm-none-eabi build working firmware with different bytes; and the two CI jobs (blocking size gate, non-blocking md5 check). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…el1) New .github/workflows/l072-bench-md5.yml, on pushes/PRs touching firmware/murata_l072 (and workflow_dispatch). It downloads the Linux x86_64 Arm GNU Toolchain 12.2.MPACBTI-Rel1 (URL and sha256 pinned, verified against Arm's published .sha256asc; cached), runs `make bench` and `make`, and prints the md5s of firmware_bench_diag.bin and firmware.bin next to the murata_l072/README.md reference values and the committed firmware.bin, in the step summary and as ::warning annotations on mismatch. Job-level continue-on-error: true and the compare step always exits 0, so it never blocks. Existing workflows untouched. YAML checked with PyYAML; every run: block passes bash -n. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…unit provision_bench_board.sh <base|tractor> [--check] runs on the PC and applies, idempotently, every board-side customization the original bench pair carries (captures board_state_2026-10-04/base, 2026-10-10/tractor): - both: sudoers drop-in via the existing install_lifetrac_nopasswd.sh, docker running, probe image from the repo-root foundries_python.tar - base: compose-apps-early-start{,-recovery}.service masked (the 934 recovery unit wipes /var/lib/docker every ~60 s), PC ssh key, optional eth0 100BASE-TX full pin (--nic-100, autoneg kept on as measured), eclipse-mosquitto:2 - tractor: units/disable-wifi.service, wpa_supplicant masked, nmcli radio off, stored WiFi profiles deleted (names only printed), camera USB provisioning by running provision_x8.sh unchanged (guarded against restarting an installed camera compose app), eclipse-mosquitto:2 from the base or a tarball, camera unit + container left stopped It never opens /dev/ttymxc3, never starts lifetrac-base* or lifetrac-tractor-compose, strips CR from everything it pushes, and does not rely on adb propagating exit codes. --check changes nothing (exit 3 when work remains). Checked with bash -n and offline flow runs against a fake adb; not run on hardware. units/disable-wifi.service is the exact unit text from the tractor capture (230 B, LF). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Ordered procedure: radio-safety rules first (RXCONT at boot and park, the tractor camera unit stealing /dev/ttymxc3, the base rebooting after poweroff, never start the production units, GO per round), hardware list (ABX00049 + ABX00043, Taoglas TD.95.6H31, 12 V, USB-A to USB-C into the X8, C2 camera), DIP switches, image choice (934 pinned by URL + sha256 and recommended on both boards; the tractor role has only flown on 674, which is hosted but not byte-verified; image-latest is not the bench image), reflash via T3a, first boot (adb, password, sudoers via install_lifetrac_nopasswd.sh, base lease, ssh key), provision_bench_board.sh with a why-table, HC-01/HC-02 verification, L072 build/flash/health/park, then DEPLOY.md and BENCH_QUICKSTART.md. Links to the other kit docs by their contract names (DEPLOY.md, BENCH_QUICKSTART.md, PC_SETUP.md, bench.env.example, legs/README.md). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Compare byte counts with and without CR (tr + wc, which the board scripts already use) instead of `tr | cmp`: nothing shows cmp exists on the LmP images, and a missing cmp would have flagged every script as CRLF and refused every flash. Re-tested the failure paths (missing staging, CRLF in one .sh while a CRLF harness cfg and stamp.py are ignored, bad REVIVE_MODE) and the FLASH_PREFLIGHT_ONLY=1 success path on a scratch copy. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…t boot, fix the 05-13 row - Image table: 934.tar.gz (1,097,240,589 B, Last-Modified 2025-08-11) is the base image and the one to use; 674.tar.gz is hosted but never byte-verified; image-latest.tar.gz is not the bench image (it moved on; the bench bundle was fetched under that name on 2026-05-08). - Download/verify/extract commands, the full sha256 list of the 934 bundle used for both base reflashes (copied from the private archive's lmp_934_bundle_sha256.txt; outer tarball 2f2065c7...), and the extracted layout (uuu runs from mfgtool-files-portenta-x8/). - uuu 1.5.243 pinned (bench copy sha256 f6b76a62...); the bundle's own uuu.exe is 1.5.109 and was unstable on Windows 11. - DIP guidance: BOOT SEL ON; if SDP does not appear, add BOOT ON. Both bench reflashes needed BOOT SEL only; PID 0134 on both; 012B never seen here. - First-boot section and an erase warning. - Verdict row 2026-05-13 corrected against log/2026-05-13_board1_T3a_postreflash_v1.md: the board was 2D0A (then "Board 1"), BOOT SEL ON only, USB-C only, PID 0134, 12/12 stages, two power cycles to first boot. The old values are kept in the row's note; 2E2C has never been reflashed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…rovisioner - Replace "No copy of the 674 image exists" with the hosted URL (674.tar.gz, 854,998,055 B, not downloaded or byte-verified, no sha256 on record); the base row now points at the 934 URL and T3a's sha256 list. - Pointer near the top to BENCH_SETUP.md and provision_bench_board.sh (applies the /etc + docker deltas listed here, never the production units). - disable-wifi.service is now in git (units/); the 5star open item notes that the provisioner deletes stored WiFi profiles (--keep-wifi-profiles defers it); the T3a row :66 doc error is marked fixed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…rapper default Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

bench_tools/bench.env.example documents every per-bench setting (serials, base transport adb|ssh, base host/key, sudo password, PC host for the harness, scratch/evidence/archive dirs, the same-day DTS carrier and its date, the scene URL and Firefox profile, the board images, the base broker container). Copy it to bench.env, which the new bench_tools/.gitignore keeps out of git. bench_tools/lib/bench_env.sh is sourced by the bench scripts: it finds the repo root from its own location (git rev-parse --show-toplevel, with a fixed-depth fallback), loads bench.env or the example (CRLF-tolerant; a non-empty caller variable wins), fills computed defaults, exports MSYS_NO_PATHCONV=1, and provides win_path, stamp, board_sh / board_sudo / board_push / board_push_lf / board_pull / board_present (adb, or ssh for the base), strip_to_gzip, board_uart_holders, board_containers and bench_require_dts_carrier (refuses an empty, out-of-band, 915.000 MHz or stale carrier). Sourcing it touches no board. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Copies of the RS-13.1 session scripts that run from any checkout: every path, serial, address, password and image comes from lib/bench_env.sh. The historical copies in bench-evidence/RS_13_vector_scene_2026-09-26/ scripts/ are untouched. - stage_boards.sh: stages /tmp/lifetrac_strict and now also the harness's /tmp/lifetrac_p0c/08_boot_user_app.cfg (LF-clean), md5-checks every pushed file, --probes-only. - push_fix_to_board.sh: pushes this checkout's image-path tree, md5 of every file, import smoke without any device. - leg_prep.sh: refuses profile-2 legs without a same-day DTS carrier, aborts on a radio-UART holder, defaults boot mode / strictness by leg tag, writes the harness command (-ForceFrfHz, -TxBatch 0, ...) as a PowerShell wrapper that also records the archive dir; --print-harness previews it without touching a board. - leg_post.sh: rs12_leg_report --capture (A14), archive found by itself, DTS-on-915 warning from params.txt, park retry on PARK_TRANSIENT. - link_sample.sh, leg2d_switch.sh, scene_check.sh: env-driven. - step1_pass.sh: step1_pass.sh + step1_pass_fix.sh merged (--from-work), always recreates bench_mqtt, scene check, refuses while a leg daemon or UART holder is present, pulls the pass files into the evidence dir. - leg_replay.py: --evidence-dir / --capture, runs from any directory (reproduces the recorded leg2a_r4 seq-gap line). - power_up_guard.sh: generic guard from guard_both.sh: waits, stops the tractor camera, stages probes, reads the radio, parks if listening, never transmits; adb or ssh for the base; --check-only for the end of a session; --stop-leg-daemons; --capture. - youtube_window.ps1: -Url / -ProfileDir / -FirefoxExe, "f" only sent to a YouTube window; never kiosk. A legs/.gitattributes keeps the scripts LF (Git Bash cannot run CRLF). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…uild base_station/requirements.lock.txt: the 27 wheels in lifetrac-v25 4623980c2dac (d375128, flown since 2026-09-15). Its own build log shows the pip step as cached (layer 9e3efb0b5438); the list comes from the 2026-09-12 build that created that layer (docker_build.log, image 12fe97dece57). firmware/tractor_x8/requirements.lock.txt: the 20 wheels in lifetrac-tractor-x8 2727dfd36f9f (6586951, flown in RS-13), from its uncached build log rs13_build.log. Headers name image ids, python base digests and logs. Both Dockerfiles gain ARG LOCKED (default 0): LOCKED=1 applies the lock as a pip constraints file on top of requirements.txt. ARG PYTHON_IMAGE (default = the old FROM tag) can pin the base image by digest. Default builds unchanged; CI builds no image. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

bench_tools/legs/README.md: setup (bench.env), the radio/GO, park, screen and DTS-carrier rules, a table of what each script touches, the order for a session (power-up guard, staging, scene, spot-check) and for one leg (prep -> harness with -ForceFrfHz $DTS_CARRIER_HZ -TxBatch 0 -KfRequestDisable 1 -ProbeEcho 0 -LogFragArrivals 1 -> link_sample -> post -> park), the 2a-2d leg table, step 1, where files go, the changes from the RS-13.1 originals and the known limits. RS13_VECTOR_LEG.md: one pointer line to legs/ next to the note that the RS-13.1 scripts carry that session's paths. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

deploy_base.sh [<commit>]: git archive (LF forced, CR count checked against the commit's blobs) of base_station/ firmware/ Dockerfile docker-compose*.yml deploy/ -> copy to the base over ssh (default) or adb -> board-side script as root: refuses while the docker-wiping recovery unit is unmasked, a lifetrac-base* unit is active or a container maps /dev/ttymxc3; backs up and replaces the tree in /var/rootdirs/opt/lifetrac/DESIGN-CONTROLLER; writes DEPLOYED_FROM.txt; creates .env and secrets/ only when missing (random key and PIN generated on the board, never printed); docker build LOCKED=1 with the log in /home/fio/docker_build_<sha>.log and the old image kept as :previous; starts ONLY the compose mosquitto service (--no-deps) and checks that 127.0.0.1:1883 accepts an MQTT CONNECT. Never starts lora_bridge. build_tractor_image.sh [<commit>]: builds lifetrac-tractor-x8:<sha> + :latest natively on the base from firmware/tractor_x8 (LOCKED=1), smoke-tests it with no network or devices, saves it to $BENCH_ARCHIVE_DIR/images on the PC (md5 checked), then adb-pushes it to the offline tractor and docker load -i's it, keeping the old :latest as :previous. --with-mosquitto, --no-tractor, --push-only FILE (rollback from the archive), --dry-run. Both read bench_tools/bench.env and contact no board in --dry-run. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

When and how to use deploy_base.sh and build_tractor_image.sh, their settings (bench.env), exactly what each changes on the base, the tractor and the PC, the lock files and the LOCKED=1 build, rollback (:previous tags, prev_tree backups, --push-only from the PC archive, never prune -a on the offline tractor), the traps (CRLF in git archive, sudo eating stdin, the docker-wiping recovery unit, compose up without a service name, /tmp tmpfs, adb from Git Bash), and the radio-safety rules: deploying never touches the radios, never start lora_bridge, and lifetrac-base.service / lifetrac-base-compose.service are enabled-but-failing on the bench base and should be disabled. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

… and leg scripts The six kit branches were written in parallel and each read bench.env on its own. Reconcile after the merge: lib/bench_env.sh accepts BENCH_ENV and BASE_USER/BENCH_ARCHIVE_DIR; the setup/deploy/flash scripts accept BENCH_ENV_FILE, BASE_SSH_USER and ARCHIVE_DIR; bench.env.example lists the optional overrides. pull_board_state.sh takes serials and the archive from bench.env. BENCH_SETUP/BENCH_RUNBOOK no longer contradict the new REVIVE_MODE default and staging preflight; the quickstart's spot-check scp uses the bench.env host/key/user. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…leg docs BENCH_QUICKSTART: - bench.env is now step 1, right after PC setup and before provisioning, flash and deploy, which all read it; the board/firmware/software steps become 2-4 and steps 5-9 keep their numbers. - The shell-variable block takes the serials, base address, key, images and archive from bench.env (read in a subshell) instead of hardcoding the original bench's values. - The GO rule and step 5.1 say what the power-up guard really does: it reads each L072 with radio_state.py and parks a listening one with radio_park.py. The guard is armed with --wait-power-cycle. - Step 3 flashes with bench_tools/flash_l072.sh (stage + preflight, then --go) and sends new carriers to FLASH_RUNBOOK section 5 first. - foundries_python.tar and eclipse-mosquitto:2 come from provision_bench_board.sh, not build_tractor_image.sh. - The '11 files' flash-staging claims and the stale 'line 5: 1: image' symptom are replaced by the flash_l072.sh preflight (PREFLIGHT-OK/FAIL). - The spot-check step parks the tractor, uses a CAND placeholder, writes DTS_CARRIER_HZ / DTS_CARRIER_DATE into bench.env and explains the harness's profile-2 refusal and -AllowDefaultCarrier. - Step 6 makes the leg_prep.sh wrapper the primary path; the by-hand line uses -TxAdbSerial/-RxAdbSerial/-HostIp placeholders and <DTS_CARRIER_HZ from today's spot-check> instead of a literal carrier. - leg_replay.py usage, the pull_board_state ARCHIVE path form, the 934-on-both / USB-A-to-C wording from BENCH_SETUP, and new troubleshooting rows (execution policy, carrier refusals). PC_SETUP: a PowerShell execution policy section (CurrentUser RemoteSigned, or Bypass for one window or one script) plus Get-ExecutionPolicy in the check block; the uuu section now agrees with T3a (use 1.5.243, never the bundled 1.5.109). RS13_VECTOR_LEG / BENCH_RUNBOOK: the 2a block and the 'A leg' template carry -ForceFrfHz (required with profile 2) and the serial/HostIp placeholders, with a pointer to leg_prep.sh. legs/README: PC_HOST is required; the 'harness silently ignores misspelled parameters', 'OSE serials built in' and CRLF-cfg statements are corrected; leg_replay usage; how to park when a prepped leg is not flown. DEPLOY: ARCHIVE_DIR (old name BENCH_ARCHIVE_DIR) with the real $HOME/Documents default, the BASE_TRANSPORT default as bench.env.example sets it, the probe/mosquitto images from provisioning, and a rollback recipe that cannot rm in the wrong directory. BENCH_SETUP / BENCH_BOARDS / FLASH_RUNBOOK: the HC-02 L072 GET probe writes two bytes to /dev/ttymxc3; the old WiFi passphrase is in public history and must be rotated (the env-var half is done); flash_l072.sh and section 5 for the first flash; FLASH_RUNBOOK's manual lines are labelled by-hand and use bench.env placeholders. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…YS paths lib/env_file.sh is now the only bench.env loader. lib/bench_env.sh, provision_bench_board.sh, flash_l072.sh, deploy_base.sh, build_tractor_image.sh and pull_board_state.sh all use it: CR stripped, a non-empty caller environment variable wins over the file (so TRACTOR_SERIAL=... bash flash_l072.sh flashes the board the operator named), and BASE_USER / BENCH_ARCHIVE_DIR are aliases of BASE_SSH_USER / ARCHIVE_DIR with the new name winning at the same level. deploy/build no longer export every bench.env value (the sudo password stayed in every child environment). Each setup script WARNs on stderr when bench.env is missing; bench_env.sh WARNs once per process tree when it falls back to bench.env.example. lib/bench_env.sh: board_push / board_pull give scp the /c/... form of the PC path (scp read C:/... as host "C", so every base push/pull over ssh failed, including the power-up guard's probe staging); BENCH_SSH_OPTS gets StrictHostKeyChecking=accept-new like deploy_base.sh; board_uart_holders answers unknown(...) instead of "free" when fuser is missing or sudo or the transport fails, so the UART gates fail closed; new bench_check_name. pull_board_state.sh: the archive path goes to adb and py in C:/... form (cygpath -m); with MSYS_NO_PATHCONV=1 the new $HOME/... default reached them as /c/... and the capture failed. Header names the real default. flash_l072.sh: also rejects a single quote in BENCH_SUDO_PW, honours ADB, prints the settings source, and refuses REVIVE_MODE=full on a 6.x kernel (the x8h7_drv re-insert OOPSes it). The base stays ssh-only (documented). build_tractor_image.sh: BASE_WORK / TRACTOR_WORK are honoured (they were documented but hardcoded), guarded because the board scripts chown -R them as root. provision: scp paths in /c/... form; stored WiFi profiles are backed up to a root-only tarball on the board before deletion. bench.env.example: states the shared rules, PC_HOST is required for radio legs, the override list shows the real defaults and drops the staging dirs (the harness hardcodes them). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…2d override leg_prep.sh: never omits -HostIp. With PC_HOST empty it uses this PC's only IPv4 address on BASE_HOST's /24 (or its only IPv4 address), prints it, and refuses (exit 6) when that is ambiguous; the harness default is the original bench PC's lease, a dead control plane anywhere else. After the probes, the next-steps text says how to stop the capture and park both radios when the leg does not fly (power_up_guard.sh --stop-leg-daemons). Leg tags, scene names, --suffix and scene-check tags must match ^[A-Za-z0-9_.-]+$ (leg_prep, leg_post, leg2d_switch, link_sample, step1_pass, scene_check): they are interpolated into board shell commands. leg_post.sh: refuses the brackets and the park (exit 3, with the commands to fix it) while tx_smoke / rx_smoke / synth_pub / camera_svc runs or anything holds /dev/ttymxc3; UART probes run as lt_post_<board> and an EXIT trap removes only those on an abort; a re-run no longer overwrites the capture log when rs13_cap_base is already gone. step1_pass.sh: an EXIT trap removes the rs13_cap / camera_svc containers this pass started when it aborts or is interrupted. leg2d_switch.sh: clears the retained encode_mode_override at the end (and on an abort after the first publish) with a zero-length retained publish, which image_rx_daemon ignores, so nothing goes on the air and no later harness run replays it; kills only its own subscriber (client id lt2d_<leg>_<pid>), not every mosquitto_sub in the broker container. power_up_guard.sh: an unknown(...) UART-holder answer reads as "cannot tell, not probing". Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

… front run_live_radio_monitor.ps1: when adb is not on PATH, throw a clear error pointing at PC_SETUP.md instead of falling back to a per-user path that exists only on the original bench PC. Nothing has touched a board at that point. Everything else in the harness is unchanged. youtube_window.ps1: SendKeys("f") goes out only when GetForegroundWindow() is the bench window (Windows may refuse SetForegroundWindow, and the key would land in the operator's terminal or editor); otherwise it says so. The bench instance is matched by its exact -profile "<dir>" argument, not a substring, and a drive-root, share-root or very short -ProfileDir is refused, so -Close cannot hit the operator's own Firefox. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

The six tasks.json WiFi tasks substituted ${env:LIFETRAC_WIFI_PSK} into the command line, which VS Code echoes ("Executing task: ...") and which also reached adb's argv, the board's sh -c / sudo argv and sudo's journal entry. They are removed; one PSK-free task in their place points to running the repo-root script by hand. The seven repo-root WiFi script headers now show a by-hand recipe that keeps the passphrase off command lines and out of sudo's log (read it in an interactive board shell, then sudo --preserve-env=...), instead of `sudo env LIFETRAC_WIFI_PSK=...`. connect_wpa.sh removes a stale /tmp/wpa.conf before writing it, since umask 077 only applies to a new file. The passphrase the kit removed is still in the public git history: it has to be rotated (docs pass / operator). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…iour change) `tr -d '<CR byte>'` becomes `tr -d '\r'`, which tr reads as the same byte. The file is a historical copy; this only lets git's content detection agree with the root .gitattributes rule (text eol=lf) for it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…s, WiFi backup, new refusals) Minor drift the verifier found after the docs and scripts passes merged: DEPLOY.md states that a non-empty environment variable beats bench.env; QUICKSTART/BENCH_BOARDS no longer insist on C:/ archive paths (pull_board_state.sh converts them); WiFi profiles are backed up root-only before deletion and the PSK-carrying VS Code tasks are gone; legs/README lists leg_prep's PC_HOST auto-detect/exit 6, the tag check, leg_post's exit 3 and leg2d_switch's override cleanup. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Copilot

🟡 Changes recommended

Several failure paths can incorrectly report successful staging, evidence capture, broker startup, or radio parking.

8 open findings

Preserve staging verification exit status · New
Reject negative force-frequency overrides · New
Remove stale artifacts before refreshing pulls · New
Refuse legs when health probes fail · New
Handle mono-only captures without vector timestamps · New
Preserve import smoke-test exit status · New
Check capture process status after wait · New
Abort when the control broker fails to start · New

What changed in this PR

Adds a reproducible, configurable two-board radio bench kit on top of #140, covering setup, deployment, flashing, leg execution, evidence capture, and shutdown.

Changes:

Adds bench environment, provisioning, staging, flash, guard, and leg tooling.
Documents the complete Windows/X8 workflow and pinned dependencies.
Removes committed WiFi credentials and adds reproducibility safeguards.

File	Description
`wifi_rescan_connect.sh`	Moves WiFi credentials to environment variables.
`wifi_persist_and_reboot.sh`	Removes embedded WiFi credentials.
`wifi_connect.sh`	Uses environment-provided WiFi credentials.
`wifi_connect_fast.sh`	Secures fast-connect credentials.
`setup_5star_wifi.sh`	Parameterizes NetworkManager profile setup.
`README.md`	Links the bench quickstart.
`X8_HEALTH_AND_RECOVERY/routines/HC-02_linux_bridge_full.md`	Updates the health-check helper name.
`X8_HEALTH_AND_RECOVERY/recovery/ADB_TIPS_AND_TRICKS.md`	Corrects board and image attribution.
`DESIGN-CONTROLLER/README.md`	Adds the radio bench entry point.
`x8_lora_bootloader_helper/run_live_radio_monitor.ps1`	Adds carrier guards and portable tooling discovery.
`x8_lora_bootloader_helper/revive_bridge.sh`	Supports reboot-mode revival.
`x8_lora_bootloader_helper/README.md`	Documents helper binaries and firmware provenance.
`x8_lora_bootloader_helper/full_flash_pipeline.sh`	Makes reboot revival the default.
`bench_tools/wdog_regs.py`	Adds a read-only watchdog register probe.
`bench_tools/units/disable-wifi.service`	Adds persistent WiFi blocking.
`bench_tools/run_flash_bench.sh`	Adds flash staging preflight checks.
`bench_tools/RS13_VECTOR_LEG.md`	Documents parameterized leg execution.
`bench_tools/requirements-pc.txt`	Pins bench-PC Python dependencies.
`bench_tools/pull_board_state.sh`	Loads shared bench settings.
`bench_tools/PC_SETUP.md`	Documents Windows toolchain setup.
`bench_tools/lib/env_file.sh`	Implements shared environment-file precedence.
`bench_tools/lib/.gitattributes`	Enforces LF for library scripts.
`legs/youtube_window.ps1`	Manages the camera scene window.
`legs/step1_pass.sh`	Automates camera-only validation passes.
`legs/stage_boards.sh`	Stages and verifies board tooling.
`legs/scene_check.sh`	Validates camera scene motion.
`legs/README.md`	Documents leg scripts and procedures.
`legs/push_fix_to_board.sh`	Pushes and verifies checkout code.
`legs/power_up_guard.sh`	Checks and parks radios safely.
`legs/link_sample.sh`	Captures live link statistics.
`legs/leg2d_switch.sh`	Automates the 2d codec switch.
`legs/leg_replay.py`	Replays and analyzes captures.
`legs/leg_prep.sh`	Prepares boards and capture infrastructure.
`legs/leg_post.sh`	Collects reports and parks radios.
`legs/.gitattributes`	Enforces LF for leg tooling.
`bench_tools/flash_l072.sh`	Adds portable staged L072 flashing.
`bench_tools/bench.env.example`	Defines shared bench configuration.
`bench_tools/BENCH_RUNBOOK.md`	Expands operational procedures.
`bench_tools/BENCH_BOARDS.md`	Documents board provisioning state.
`bench_tools/.gitignore`	Excludes local bench configuration.
`tractor_x8/requirements.lock.txt`	Records flown tractor dependencies.
`tractor_x8/Dockerfile`	Supports locked dependency builds.
`murata_l072/README.md`	Documents reproducible firmware builds.
`firmware/common/setup_x8_mqtt_link.ps1`	Corrects board attribution.
`FIRMWARE_UPDATES.md`	Corrects handheld hardware instructions.
`DESIGN-CONTROLLER/Dockerfile`	Supports locked base-image dependencies.
`bench-evidence/.../step1_pass_fix.sh`	Repairs historical CR stripping.
`base_station/requirements.lock.txt`	Records flown base dependencies.
`BUILD-CONTROLLER/05_firmware_installation.md`	Warns about overwriting x8h7 firmware.
`do_wifi.sh`	Removes embedded WiFi credentials.
`connect_wpa.sh`	Generates a protected environment-driven profile.
`.vscode/tasks.json`	Removes credential-bearing WiFi tasks.
`.github/workflows/l072-bench-md5.yml`	Adds non-blocking firmware hash verification.
`.gitattributes`	Defines repository line-ending and binary rules.

🧠 Review effort: Balanced

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

+
+  if [ $PROBES_ONLY = 0 ]; then
+    stamp "$who: code tree"
+    bash "$LEGS_DIR/push_fix_to_board.sh" "$who" 2>&1 | grep -E "== push|MISMATCH|verified|encoder from|has _tick|SKIP|FAILED" || true



+# Bench kit (2026-10-10): refuse a DTS leg on the default carrier BEFORE
+# anything touches a board. Legs that pass -ForceFrfHz are unaffected.
+if ($RegProfile -eq 2 -and $ForceFrfHz -eq 0 -and -not $AllowDefaultCarrier) {


+for f in leg${LEG}_base.txt leg${LEG}_base.jsonl leg${LEG}_base.json; do
+  board_pull base "$BOARD_STAGE/legs/$f" "$E/$f" || echo "PULL FAILED $f"
+done


+{ hdr "health base"; board_sh base "$SUDO $PROBE_RUN $BASE_IMAGE -u /work/rs116_health_probe.py 2>&1"; } > "$E/leg${LEG}_health_base.txt"
+grep -E "STATS-OK|STATS-FAILED|radio_state|RS12-URC" "$E/leg${LEG}_health_base.txt" | head -4
+{ hdr "health tractor"; board_sh tractor "$SUDO $PROBE_RUN $TRACTOR_PROBE_IMAGE -u /work/rs116_health_probe.py 2>&1"; } > "$E/leg${LEG}_health_tractor.txt"
+grep -E "STATS-OK|STATS-FAILED|radio_state|RS12-URC" "$E/leg${LEG}_health_tractor.txt" | head -4


+lost = sum(g for _, g in gaps)
+print(f"leg {leg}: {len(seqs)} frames received; seq gaps within same-codec runs (row, lost): {gaps} -> {lost} lost of {len(seqs) + lost} = {100.0 * lost / (len(seqs) + lost):.1f} % after the first received frame ({switches} codec switch(es) not counted)")
+print(f"  resync episodes (start, end, ended by key): {episodes}")
+print(f"  all ended by an epoch start: {all(e[2] for e in episodes)}; final resync {dr.store.stats['resync']}, digest_ok {dr.store.snapshot(int(dr.last_ts * 1000) + 100)['digest_ok']}")


+    board_sh "$WHO" "$SUDO $WORK_RUN $TRACTOR_APP_IMAGE -c \"import x8_image_pipeline.encode_vector as ev; print('encoder from', ev.__file__); print('codec', ev.vs.__file__, 'TTL_FRAMES', ev.vs.TTL_FRAMES); print('has _tick_ttl', hasattr(ev.VectorEncoder, '_tick_ttl'))\"" 2>&1 \
+      | tee -a "$REC"


+  | tee -a "$E/step1_${TAG}_bootcheck.txt"
+
+# 3. wait for the capture, then encoder timing and teardown
+wait $CAP_PID


+        if (-not $ar2.AsyncWaitHandle.WaitOne(1000, $false) -or -not $probe2.Connected) {
+            Write-Warning "PC broker not answering on $($HostIp):1883 after start. Check: $PythonExe $($pyPrefixArgs -join ' ') -c `"import amqtt`" (install with: py -3 -m pip install -r bench_tools\requirements-pc.txt)."
+        }


Conflicts in BENCH_RUNBOOK.md and RS13_VECTOR_LEG.md resolved by keeping both sides: the 2026-10-10 amendment notes and carrier-range detail from this branch, the bench.env / serial / HostIp placeholders and legs/ pointer from the bench kit. base_station on the merged tree: 1598 passed, 2 skipped. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

dorkmo and others added 30 commits October 10, 2026 13:12

BENCH_SETUP: REVIVE_MODE=reboot is what FLASH_RUNBOOK passes, not a w…

ef88041

…rapper default Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Merge kit/k1-hygiene into bench-kit-2026-10-10

f8a18d5

Merge kit/k2-board-setup into bench-kit-2026-10-10

0509733

Merge kit/k3-deploy into bench-kit-2026-10-10

bb43fcf

Merge kit/k4-legs into bench-kit-2026-10-10

ac1a8c0

Merge kit/k5-harness-flash into bench-kit-2026-10-10

b28c866

Merge kit/k6-quickstart into bench-kit-2026-10-10

cc88454

dorkmo and others added 9 commits October 10, 2026 13:45

Merge branch 'kitfix/scripts' into kitfix/verify

78ea975

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Copilot AI balanced review requested due to automatic review settings October 10, 2026 19:52

Copilot started reviewing on behalf of dorkmo October 10, 2026 19:53 View session

Copilot AI reviewed Oct 10, 2026

View reviewed changes

dorkmo mentioned this pull request Oct 10, 2026

RS-13: A16 loss-repair fix (DEL repeat + repair share), VECTOR code guards, procedures and desk-check tool #143

Merged

dorkmo changed the base branch from rs13-session-save-2026-10-04 to main October 11, 2026 00:54

dorkmo merged commit 41d9e6d into main Oct 11, 2026
25 checks passed

dorkmo deleted the bench-kit-2026-10-10 branch October 11, 2026 00:58

Navigation Menu

Repository navigation

bench kit: reproduce the LifeTrac radio bench from GitHub (setup, deploy, legs, guards) - #142

bench kit: reproduce the LifeTrac radio bench from GitHub (setup, deploy, legs, guards)#142
dorkmo merged 39 commits into
mainfrom
bench-kit-2026-10-10

dorkmo commented Oct 10, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

		board_sh "$WHO" "$SUDO $WORK_RUN $TRACTOR_APP_IMAGE -c \"import x8_image_pipeline.encode_vector as ev; print('encoder from', ev.__file__); print('codec', ev.vs.__file__, 'TTL_FRAMES', ev.vs.TTL_FRAMES); print('has _tick_ttl', hasattr(ev.VectorEncoder, '_tick_ttl'))\"" 2>&1 \
		\| tee -a "$REC"

Conversation

dorkmo commented Oct 10, 2026

What's in it

Review

Not verified on hardware (by rule — radios stayed asleep)

Open, outside this PR

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟡 Changes recommended

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants