[mbreiser]

Summary

One command on a rig PC validates a webDisplayTools tier end to end in about five minutes of wall time, with the person at the rig only confirming what they see. It replaces the hour of hand-driven QC of 2026-10-27/28 (docs/development/handover-qc-windows-2026-10-26.md; results on #274).

cd rig
pixi run qc -- --tier next            # the candidate on /next/, rig detected from the hostname
pixi run qc -- --tier next --rig ww14 --auto   # no Enter prompts: a Claude session drives it

What it does (rig/qc/qc.py): preflight (pixi ≥ 0.80, ports 7447/8765/60000 free, the rig's services up) → the tier's build.json → a scratch git worktree at exactly that SHA with its own pixi environment (pixi install --locked, plus typed/viz per rig; the rig's checkout and .pixi are never touched) → the candidate's pixi run rig with the rig's flags (flags that build does not know are dropped with a warning) → FicTrac started or found running, frames counted at the gateway, simulator fallback → tests/studio_emulator.py pre-check (a broken candidate fails here, before any browser) → Chrome (a QC profile with no course token) on the tier's arena_studio.html?mode=run&rig=…, driven over DevTools: build SHA + footer, protocol load, bridge connect, the one human click (Web Serial port picker), firmware + panels, Run experiment → pixi run export, runlog-check, closed-loop-report, bench-check → teardown, pass/fail table, validation-note.md and the exact pixi run tiers -- validate … line.

• rig/qc/profiles.toml — the per-rig rows (ww18, ww14, bergamo, mac: flags, protocol, FicTrac command, Foxglove layout, required ports, report holds) and the checklist text Michael reads at each step. A new rig is a row.
• rig/qc/qclib.py — the pure half, pinned by tests/test-rig-qc.py (77 checks, runs in the root pixi run test).
• rig/QC.md — the procedure page: what to type on ww18 / ww14 / Bergamo, the two clicks, where artifacts land (C:\qc\<date>\), safety on a rig PC, adding a rig.
• rig/pixi.toml — the qc task. The branch includes rig: pin rig-nodes 0.3.0 + re-lock with pixi 0.83 (#280, LAB-338) #281 (rig-nodes 0.3.0 re-lock) so the rig workspace installs on pixi 0.83.

Tested on this Mac against candidate 2026-10-28-rc2 (06bad57) on /next/: PASS end to end in 26–35 s wall with a warm cache — graph (router, fictrac, host, bridge, stimulus, viz, foxglove), FicTrac-MacOS on the looped sample video (50 fps), emulator pre-check, Chrome shows build 06bad57 / v0.100 with the protocol loaded and 51 frames/s at the bridge, Foxglove opened on ws://localhost:8766, export + bag closed + closed-loop report 0 % mismatch. --tier prod on main 8d5b952 fails cleanly at the install step (main's rig lock predates pixi 0.80, #280 — fixed by today's release). The controller path (click → firmware/panels → Studio.runOnce(true) → bench-check) could not run here (no controller); the first exercise is the ww18 dry run.

Finding on the way: a never-used Chrome profile holds ws://localhost:8765 from the Pages origin on the Local Network Access permission prompt (Chrome 155: no error, "connecting…" forever). A DevTools Browser.grantPermissions localNetworkAccess does not unblock it; --disable-features=LocalNetworkAccessChecks does (the flag the Foxglove desktop app itself runs with), so the QC Chrome uses it; --lna-prompt shows the prompt instead.

Linear: LAB-222 (two-tier releases — the validation step), LAB-338 §7 item 8 (the cross-repo smoke on a rig PC). No HTML, no footer, no release-notes file touched; nothing Prettier-formatted.

Tier

next candidate (rig-side tooling + scripts only).

Release notes

none

Test plan

• pixi run test (root): 73/73 files, incl. the new tests/test-rig-qc.py
• cd rig && pixi run qc -- --tier next --rig mac --auto end to end on this Mac (rc2 06bad57)
• --no-browser, interactive mode (piped Enter + a remark), --tier prod failure path
• ww18 dry run (pixi run qc -- --tier next --rig ww18) — controller path, real FicTrac via fictrac.bat
• ww14 dry run (pixi run qc -- --tier next --rig ww14) — BIAS connector pre-check, four-color smoke

First run on ww18 (candidate 2026-10-29 rc1, 2026-10-29)

pixi run qc -- --tier next --rig ww18 --auto ran end to end in 104 s: candidate checked out and installed, graph + real FicTrac (96 fps), emulator pre-check PASS, Studio on /next/ with build f49f34f, controller connected on Michael's click, run completed, export, bag, runlog-check and closed-loop-report (0.00 % idx mismatch) all PASS. Two findings, fixed on this branch:

• The start-frame check was the wrong test on a live rig. A live fly turned the ball −1108° in 8 s; the display tracked it (0.00 % idx mismatch), so bench-check's "holds frame 25/75 for the whole epoch" lines failed. Michael: with FicTrac live the ball can always move. The check is now "epochs open on their start frame": closed-loop-report emits first_idx_after_tare / opens_on_start_frame per epoch (the first post-tare index must be the trialParams frame_index, 25 / 75), qc.py records it, and the rest of the epoch is the idx-mismatch check. The ww18 profile drops holds; holds stays for a bare or blocked ball, where qclib.hold_verdict() turns a hold miss into SKIP when the ball turned.
• Windows teardown orphans. signal_group(SIGINT) is CTRL_BREAK_EVENT on Windows, delivered to Python as SIGBREAK, which rig.py did not handle: it died before its finally, leaving the nodes and the router running ("ports free again" FAIL). rig.py registers SIGBREAK now.

Also: the bergamo profile runs the public g6_2x10_smoke protocol while the 2P diagnostics protocol lives in the private repo (LAB-227).

🤖 Generated with Claude Code