[shubhamTR]

Summary

Runs of the rig03 SBD visual place-learning protocols (p3-sbd-placelearning-*, sbd-bias-*; G6 ports of the G4.1 p058 / p066 / p067 protocols) showed up in the dashboard as "generic protocol" — and because the filenames start with p3-, the bare p3 substring would have routed them into the Heisenberg legacy branch. This PR adds the sbd-place family and its analysis pages, and an optomotor family for the optomotor ± opto LED tests.

• analysis-core.js — protocolInfo recognises the family before the p3 checks (filename contains sbd, or conditions = preprobe + train_plus/train_minus / pair_m<k>_p<k>) → {family:'sbd-place', sbdVariant: base|coupled|uncoupled|bias, sbdVersion: v1|v2|v3}. New exported helpers: sbdZone (safe zone from the first training bout's logged led_activation zone; default heat [8,158) → 50 safe frames centred on 183), sbdBouts (one row per Mode-3 interval, not per step — the uncoupled control runs two bouts per pair_m<k>_p<k> step; start frame/side, pattern id → picture rotation k, safe / opposite / LED-on fractions, first-entry latency, PI), sbdStageSummary, sbdOccupancy. Scoring is in frame space, so the uncoupled control's fixed heat map is scored exactly like the coupled protocol.
• analysis-core.js, optomotor family — the optomotor ± opto LED tests (optomotor_led_test, optomotor_test_60deg, optomotor_test_60deg_tf; conditions grating_cw_noled / grating_cw_led) are classified before the p0 rule, which claimed them through grating_cw_*. The filename must start with optomotor_, so protocols with "optomotor" in the name and no opto LED (g6_color_optomotor_test, simple_optomotor_protocol) stay generic. The label names the grating period; the pages are the generic set.
• plot-specs.js — five pages: time in the safe zone per bout (chance line), time to first zone entry, LED-on fraction per bout, heading occupancy around the zone (zone + 15° ramps shaded, 5-bin circular boxcar), stage preference index (t safe − t opposite)/(t safe + t opposite) with "opposite" = the zone rotated 180°. Group mode draws per-fly lines + mean ± SEM; every page exports CSV rows. buildPages dispatches sbd-place before the generic fallback.
• README — "SBD visual place learning" section.
• tests/test-dashboard-sbd.js (new, auto-discovered by pixi run test) — synthetic behavior_v2 log with an uncoupled pair step: classification (incl. the p3- trap, Heisenberg files staying p3, optomotor vs p0, and "optomotor" inside a filename staying generic), zone geometry, bout rows, fractions, PI, stage summary, occupancy, buildPages.

Not in this PR: the v3 sham and mid-probe stages are labeled but not scored (follow-up).

Context: SBD protocols + patterns 46–49 live in reiserlab/cshl-2026-course (protocols/rig03-sr/); the start_frame runner/bridge work shipped in #223.

Tier

next candidate

Release notes

Dashboard

• SBD visual place-learning runs now get their own analysis pages instead of the generic protocol view: time in the safe zone per bout, time to the first zone entry, LED-on fraction, heading occupancy around the zone, and a preprobe / training / probe preference index, with per-fly lines and mean ± SEM in group mode and CSV export on every page.
• The run catalog and the group filter list these runs as sbd-place, so all variants and LED levels form one group. A loaded run shows the full label with the variant and protocol revision, for example "SBD place learning (coupled v3)".
• Optomotor ± opto LED test runs are now their own family, optomotor, instead of being read as p0 optogenetic-intensity runs. A loaded run shows the grating period in its label; the pages are the generic set.

Test plan

• pixi run node tests/test-dashboard-sbd.js — 17/17
• pixi run test on a trial merge with main (2026-11-11) — 74/74 files
• Parsed all 20 SBD run logs in cshl-2026-course/runlogs/rig03-sr through the new family in Node: every file → sbd-place with the right variant, heat [8,158) read from the log, 22 bouts (preprobe + 20 training + probe), alternating +/− start sides, 0 parse errors; group pages build (21 traces each)
• Review (2026-11-10): all 203 run logs in arena-experiments (66b42ce) classified with main and with this branch — 139 labels change, all intended (SBD and optomotor runs); no real p0 or Heisenberg p3 run changes; 0 parse errors
• The 30 real optomotor run logs keep the same catalog and loaded-run labels after the narrower filename rule (3ceb29f)
• Browser check on /next/ once the candidate is published (the Node test exercises the page objects; the UI consumes title/description/figure/csvRows generically)
• No footer / version / release-notes-file edits (release tiers)

🤖 Generated with Claude Code

[mbreiser]

@shubhamTR this is reviewed and good for the next candidate. To save you a round, two small commits are now on your branch and the description is updated. Pull before you push again.

What changed

• 3ceb29f: the optomotor filename rule is narrower. The file name must start with optomotor_. Before, any name with "optomotor" in it matched, so g6_color_optomotor_smoke, g6_color_optomotor_test and simple_optomotor_protocol (no opto LED) would be labeled "optomotor ± opto LED". The condition-name rule is unchanged. Your 30 real optomotor run logs keep the same labels.
• d973b42: README correction. The run catalog and the group filter show the family id (sbd-place, optomotor). The full label ("SBD place learning (coupled v3)") shows on a loaded run.
• PR description: the same catalog correction in the release notes, the optomotor family added to the summary and release notes, and the test plan updated (17/17; 74/74 files on a trial merge with main).

Review result

All 203 run logs in arena-experiments (66b42ce) were classified with main and with this branch. 139 labels change, all intended. No real p0 or Heisenberg p3 run changes.

Still open, as a follow-up

Scoring of the v3 sham and mid-probe stages. Not needed for this PR.

🤖 Generated with Claude Code