Repository navigation
Release 2026-10-29: Arena Studio v0.101 - #290
Merged
Merged
Conversation
…a rig PC (LAB-338) One command checks the candidate on /next/ (or Production) end to end: the tier's build.json → a scratch worktree at that SHA with its own pixi environment, the candidate's graph with the rig's flags, FicTrac (or the simulator), the Studio emulator pre-check, the tier's Studio in a QC Chrome profile driven over DevTools (build SHA, protocol load, bridge connect, the one Web Serial click, Run experiment), export + runlog-check + closed-loop-report + bench-check, teardown, a pass/fail table and a ready-to-paste validation note. Per-rig rows and the checklist text live in rig/qc/profiles.toml (ww18, ww14, bergamo, mac); the procedure is rig/QC.md; tests/test-rig-qc.py pins the pure half. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
rig/pixi.toml reads the channel at https://raw.githubusercontent.com/reiserlab/rig-nodes/v0.4.0/channel (the tag form; 0.3.0 was pinned by the payload commit's SHA only because its tag had landed one commit early) with rig_nodes / rig_interfaces ==0.4.0, re-locked with pixi 0.83 (`pixi lock` in rig/; the `dev` environment's git sources on rig-nodes main re-pinned to fa83d39). 0.4.0 vs 0.3.0: the ScanImage JSON connector moved to the private 2P rig repo (rig-nodes#4), the host monitor's typed HostStats floats are quantized and its portable process sampling is cheaper (rig-nodes#6); no .msg change. Verified on the Mac with ~/Documents/GitHub/rig-nodes renamed away: `pixi install --locked` (default + typed, 4 s each), `pixi run test` 60/60 (host monitor self_cpu_median 0.006 of a core — the macOS cost gate the 0.3.0 sampler failed here), `pixi run -e typed test` 63/63, `check-typed` lists the 8 types incl. HostStats, a plain `pixi install` leaves the lock as written. Gateway: hello_ack still advertises the scanimage connector (the emulator's --scanimage-cmds node checks need the private node, as on main). Docs: WINDOWS-TEST.md expected harness count 60 / 63, README.md adapters list, the okey-cokey dashboard plan's tiff_run_to_mcap.py line (private since 2026-10-28), the rig-nodes dependency comment. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Rig-specific protocols and per-rig records leave the public repo so a rig's protocol tweaks never need a webDisplayTools release. Rule: a protocol stays here only if a tool, test, QC profile or fixture depends on it. - Removed (copied first to reiserlab/arena-experiments 8db48e6 and reiserlab/scanimage-2026-bergamo 00bccfd, each with PROVENANCE.md): the three bias_camera_closed_loop_poc* protocols (→ protocols/ww18/), fictrac_bias_test, fictrac_direction_test, fictrac_coupling_bias_fly_test, led_activation_quadrant_test (+ patterns), mode3_drill_1trial (→ protocols/shared/); the three 2P protocols + two pattern folders (→ bergamo protocols/); the ww18 and okey-cokey (ww14) run sheets, handover, dashboard plan, BIAS config and .ps1 start scripts (→ rigs/ww18/, rigs/ww14/). - protocols/index.json drops the moved keys; protocols/MOVED.md maps every old path to its new home. - tests/test-protocol-roundtrip-v3.js parses the 2P sweep from a fixture copy (tests/fixtures/v3_g6_2p_color_led_sweep.yaml). - scripts/make-color-optomotor-patterns.js --gs16 and scripts/make-2p-diagnostic-patterns.js take --out <dir> (the 2P targets are no longer in this repo; without --out they explain where they went and exit 2). - rig/README.md, rig/WINDOWS-TEST.md, the protocol-yaml skill, two course docs and the decision record's "stays" row point at the new locations. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…uard miss) GitHub returns one ETag for the raw and the JSON form of a contents URL; after a raw read (?p= open, promote, previews) Chrome revalidated the cached raw body for the later JSON request and served it on the 304, so res.json() threw and every existence check (resolveRepoSaveTarget, the silent-fork guard) answered "no": a changed protocol overwrote the repo copy with no "Overwrite?" prompt. Found on ww18 during the LAB-227 private-repo test, 2026-10-29. GH.run now sends cache: 'no-store'; runBytes keeps caching since nothing JSON is stored. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
… counter The ring block header's `dropped` is the controller's eviction count since its boot, so a Studio connected hours later logged "dropped 1929" with gaps 0 and read as in-run loss (ww18, 2026-10-29). The drainer now keeps the counter at its first block of a ring incarnation (`droppedBase`, reset on a fresh ring) and exposes `droppedSession`; the once-a-minute stats line leads with "evicted this session N (controller lifetime M)". `telemetry_dump` and the soak stats carry the new field; the raw `dropped` stays for the record. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…ostics moved to the private repo, LAB-227) Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…on protocol needs FicTrac) Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…EAK (ww18 rc1 QC findings)
Two findings from the first `pixi run qc` on ww18 (candidate 2026-10-29 rc1, 2026-10-29):
1. bench-check's hold checks assume a still ball. A live fly turned it −1108° in 8 s, the
display followed FicTrac with 0.00 % idx mismatch, and the QC table said FAIL. New
qclib.hold_verdict(): when ONLY hold lines failed and closed-loop-report shows an epoch with
|net turn| >= 45°, the row is SKIP naming the turns; a hold miss on a still ball, or any other
bench-check failure, stays FAIL. ww18's run/reports look text says so. Tests in test-rig-qc.py.
2. Teardown on Windows left the graph's nodes and the router as orphans ("graph stopped", then
"ports free again" FAIL on 7447/8765/60000). qc.py stops rig.py with
rig_nodes.supervise.signal_group(SIGINT), which on Windows is CTRL_BREAK_EVENT on the process
group; Python delivers that as SIGBREAK, which rig.py did not handle, so it died before its
`finally` stopped the children. rig.py now registers SIGBREAK where it exists (bridge_node and
host_monitor already did). QC.md documents both.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
mbreiser
force-pushed
the
release/2026-10-29
branch
from
October 29, 2026 20:41
f49f34f to
303853e
Compare
…a whole-epoch hold Michael on the rc1 QC: with FicTrac live the ball can always move, so a check that the display HOLDS frame 25 / 75 for a whole epoch is the wrong test on a rig. What the two epochs exist to verify is the v0.91 start frame: the first post-tare index equals the trialParams frame_index (closed-loop-start-frame.md). closed-loop-report now emits `first_idx_after_tare` + `opens_on_start_frame` per epoch (and marks "200 @25 ✓" in the table); qc.py records "epochs open on their start frame" from it; the rest of the epoch is the existing idx-mismatch check. The ww18 profile drops its `holds` (bench-check still judges completion, arena acks, channel, build; `holds` stays available for a bare or blocked ball, with hold_verdict's SKIP when the ball turned). Tests in test-closed-loop-report.py cover the open on 25 / 75 and a ≤ 3.3 bridge opening on 0. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…f a tier on a rig PC (LAB-222, LAB-338)
# Conflicts: # rig/docs/okey-cokey-foxglove-dashboard-plan.md
…sion; telemetry line reports in-session evictions (LAB-227)
mbreiser
force-pushed
the
release/2026-10-29
branch
from
October 29, 2026 20:45
303853e to
41c1f51
Compare
Contributor
Author
|
✅ Validated
Build label: candidate 2026-10-29-rc3, sha 41c1f51, PRs 286, 285, 288, 289; page meta 41c1f51, channel next, footer Arena Studio v0.101 | 2026-10-29 16:45 ET.
Remarks: panel inventory "unsupported" on this firmware (expected). Three cosmetic zenoh "Unable to publish link event: session closed" ERROR lines at bridge shutdown (rig-nodes follow-up). rc1 of this candidate had failed on the two QC-tool findings fixed in #285 (hold checks vs. a live fly; Windows teardown); rc2 was superseded unvalidated. Artifacts on ww18: C:\qc\2026-10-29\summary.md, validation-note.md, reports\ (runlog-check, closed-loop-report, bench-check). |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Release candidate
2026-10-29rc3 · candidate commit41c1f51d4d· built on main59a377b05eLive on Next: https://reiserlab.github.io/webDisplayTools/next/
9018462064pixi run qc— the five-minute release validation of a tier on a rig PC (LAB-222, LAB-338)9614a91a9091bf95849e4090716d45Promotion checklist
pixi run releasechecks)Validated 41c1f51 …(pixi run tiers -- validate --rig … --by …). Any change to a constituent PR = rebuild (rc+1) = validate again.releaselists recent rig runs — advisory)fictrac-bridge/changed, rig PCsgit pull+ restart the bridgePromote with
pixi run release(merges with a merge commit — never squash — and tagsrelease-2026-10-29). Manifest:releases/2026-10-29.json.🤖 Generated with Claude Code release tooling (
scripts/tiers/candidate.py)