[mbreiser]

2026-10-28 — main (release 2026-10-27, #274 included) merged in; ready for the candidate with the rig-nodes 0.3.0 re-pin PR. rig_nodes.bias_health and the derive hook ship in rig-nodes 0.3.0 (rig-nodes#1 merged there), so the "needs the next package-set release" note below is satisfied by the re-pin PR in the same candidate. Dropped here: the [feature.dev.activation.env] Zenoh URIs for a sibling path checkout (the dev environment uses git source dependencies since 2026-10-27, so the installed package's configs apply). Kept: the protocols, the camera + G6 Foxglove dashboard (stimulus_node, stimulus_core, dashboard_replay, arena_display), the uFMF tools, rig/config/bias/okey-cokey-bfs-u3-51s5m.json, the ww14 run records, and [feature.viz.target.win-64.activation.env] PATH = '%CONDA_PREFIX%\Library\lib;%PATH%' (ww18 2026-10-28: --foxglove fails without it). Still local, for later: tools/ufmf.py → rig-nodes (numpy is now a rig_nodes run-dependency, so the blocker is gone), arena_display.py + tools/patfile.py → pyDisplayTools.

Summary

The okey-cokey rig work (LAB-303) that lived on two branches without a PR, claude/okey-cokey-pairs and claude/okey-cokey-dashboard (pairs is an ancestor of dashboard). Rebased as one branch onto released main, then merged over the extraction branch feat/rig-nodes-extraction (#274) on 2026-10-26 — this PR is stacked on #274 (base retargeted to that branch so the diff shows only the okey-cokey changes; GitHub retargets it to main when #274 lands). A candidate takes #274 and #269 together (order_prs wants the stack explicit).

• Protocols (Grace's shape): protocols/g6_color_optomotor_test.yaml and _smoke.yaml run CW/CCW pairs (15 s each, no gap, 10 s dark between pairs); the smoke variant has the same shape.
• rig/nodes/stimulus_node.py + stimulus_core.py: the expected-vs-observed display + arena-health node behind the camera + G6 Foxglove dashboard (foxglove/arena-okey-cokey.json; pixi run rig -- --stimulus); arena_display.py is a Python port of js/panel-color.js (the unit test runs the JS module under Node to pin the port).
• dashboard_replay.py, tools/ufmf.py (BIAS uFMF reader), the BIAS camera config for the BFS-U3-51S5M, PowerShell launchers (okey-cokey-start.ps1 starts BIAS, the rig and the server through WMI so a restarting desktop app cannot take them down; bias-ufmf.ps1).
• Records: bare-PC setup checklist, run sheet, ww14 setup + results (2026-10-22): the full g6_color_optomotor_test run with flies passed, arena 48/48 trials, BIAS 58/58 replies ok, 70 953 frames at 73.3 fps; open finding R4 (a panel-side glyph seen by eye in the second half; controller telemetry clean).
• rig/pixi.toml: stimulus, test-stimulus, test-stimulus-live tasks; a win-64 viz activation fix (foxglove-bridge DLL path).

What moved with the extraction (2026-10-26)

The shared half of rig/ is now the installed rig_nodes package (reiserlab/rig-nodes, #274). This PR's additions are expressed through it:

Piece	Before (on main)	Now
rig/nodes/bias_health.py (CountRate + bias_drops)	local module	moved to rig-nodes — reiserlab/rig-nodes#1; viz_node.py, dashboard_replay.py import rig_nodes.bias_health; its unit checks moved to rig-nodes tests/test_pure.py
foxglove_json.py --stimulus	the dashboard replay spliced into the local converter	the arena wrapper keeps #274's enrich hook and adds the replay as the generic rewriter's new derive hook (whole record stream → new topics; images by the well-known foxglove.CompressedImage schema name) — the hook is the second half of rig-nodes#1; dashboard_replay.run_streams is unchanged
rig.py --no-fictrac / --stimulus / --stimulus-args	edits to the local launcher	the same options on the rig_nodes.supervise composition (stimulus_node.py is a Child like viz_node.py; the FicTrac adapter and its udp 60000 preflight are skipped together)
_common imports in stimulus_node.py, viz_node.py, test_stimulus_live.py	from _common import …	rig_nodes._common / rig_nodes.supervise

Needs rig-nodes ≥ the next package-set release (the one that ships rig-nodes#1): the pinned 0.2.1 has neither rig_nodes.bias_health nor the derive hook, so on the released channel viz_node.py, stimulus_node.py's replay and test-stimulus fail at import. Until then: cd rig && pixi install -e dev with a sibling ../../rig-nodes checkout on feat/bias-health (or main once merged), and run the tasks with -e dev. When rig-nodes releases, re-pin rig_nodes / rig_interfaces in rig/pixi.toml and re-lock (that is a one-line follow-up on this branch or the next candidate).

Still local, listed for rig-nodes in the record's §3.3: rig/tools/ufmf.py (a BIAS reader, device-generic). Not moved here — a follow-up, with patfile.py → pyDisplayTools.

Tier

next candidate (2026-10-28, with the rig-nodes 0.3.0 re-pin PR; the dashboard has been exercised on ww14 with flies, the protocols ran there). Based on main since 2026-10-28.

Release notes

Protocols

• g6_color_optomotor_test / g6_color_optomotor_smoke now present each color as a CW/CCW pair (15 s each, no gap) with 10 s dark between pairs.

Test plan

• Root pixi run test: 72/72 files (Mac, 2026-10-26, merged head)
• cd rig && pixi run -e dev test-stimulus: 50 checks, 0 failures, 0 skips (incl. the --stimulus replay end to end through the derive hook)
• cd rig && pixi run -e dev test-viz: PASS
• cd rig && pixi run -e dev test (the ROS graph acceptance harness against the sibling rig-nodes build): 62/62 checks (after fixing the dev environment's Zenoh config URIs — editable install, see the second commit)
• cd rig && pixi run -e dev test-stimulus-live: 10/10 (router → bridge → stimulus node on the fixture replay)
• Full run with flies on REISERLAB-WW14, 2026-10-22 (results file in the PR; pre-extraction code)
• Foxglove replay of the full run (not done yet per the results file)
• On a rig PC: pixi install on the released channel once rig-nodes ships the next set, then pixi run rig -- --no-fictrac --bias --foxglove --viz-args "--sources bias" --stimulus-args "--panel-layout four-color"

Linear: LAB-303 (okey-cokey rig), LAB-338 (extraction)

🤖 Generated with Claude Code