[shubhamTR]

Summary

Follow-up to #249 (Michael's PR comment / DM): score the v3 sham and mid-probe stages of the SBD place-learning protocols in the data-browser dashboard.

The v3 (p3-sbd-placelearning-coupled-v3-led14) and v4 (…-coupled-v4-led14, …-uncoupled-v4-led14) protocols add three tests around the training: sham_preprobe (before preprobe, closed loop with the LED off), probe_mid (heated test after bout 10) and sham_probe (LED off, after bout 20, before probe). The dashboard's SBD scorer only knew preprobe / training / probe, so those bouts were dropped. Now:

• Stage mapping (analysis-core.js sbdStage): sham_preprobe / probe_mid / sham_probe are stages of their own; training also matches the v4 uncoupled control's train_k<k>_<plus|minus> bout names (the classifier's condition-name fallback uses the same regex, so a renamed v4 copy still routes to sbd-place).
• Protocol-order position per bout, shared across protocol versions: sham before −1, before 0, train_k = k, mid test / sham test = k + 0.5 after bout k, test = N + 1.
• Heat-off shams: the two sham stages carry heatOff = true (no led_activation on the trialParams, ledDrive 0 before), so the LED-on fraction is 0 for them and uniformHeat is false; their preference index uses the same zone geometry as the probes.
• Stage summary + pages: sbdStageSummary carries all six stages (zero bouts when a protocol lacks one), stages (those present, in protocol order), a short label ("sham before", "before", "training", "mid test", "sham test", "test") and the heat-off flag. The occupancy and preference pages show only the stages the loaded runs ran (a mixed v2 + v4 group shows the union), the sequence pages label the tests on the x axis and shade one band per test. CSV rows gain heat_off and stage_label.
• v1/v2 runs score as before: the existing checks pass unchanged, and a new check pins the v1/v2 run's flags, summary and tick labels.
• Semantics mirror the lab's local Python sbd-dashboard (sbd.py: STAGES, SHAM_STAGES, bout positions, stage_summary), not its code.

Files: dashboard/data-browser/analysis-core.js, dashboard/data-browser/plot-specs.js, dashboard/data-browser/README.md, tests/test-dashboard-sbd.js.

Tier

Next candidate.

Release notes

Dashboard

• The SBD place-learning pages now score the v3/v4 sham tests and the mid test. sham_preprobe, probe_mid and sham_probe bouts appear as their own stages ("sham before", "mid test", "sham test") on the time-in-zone, first-entry, LED-on, occupancy and preference pages, scored on the same safe zone as the probe. Sham tests show an LED-on fraction of 0 and are flagged heat_off in the CSV.
• Bout positions line up across protocol versions: sham before −1, before 0, training bouts 1…N, mid test / sham test half a step after the bout they follow, test N + 1. The sequence x axis labels the tests and numbers the training bouts.
• The v4 uncoupled control is recognised (train_k<k>_<plus|minus> bout names). v1/v2 runs are unchanged.

Test plan

• pixi run test -- dashboard-sbd — 26 / 26 checks (17 existing + 9 new: stage mapping, positions, heat-off flags, summary, pages, v1/v2 regression, v4 classification)
• pixi run test — every dashboard test passes; the 7 files that fail on this Windows checkout (test-arena-studio-alt, test-project-skills, test-protocol-roundtrip-v3, test-rig-qc, test-studio-runtime-vars, test-studio-shell, test-three-viewer-partial-columns) do not reference the dashboard and fail identically without this change
• Load a v4 run (p3-sbd-placelearning-coupled-v4-led14) in the data-browser and check the six stages on the preference page

🤖 Generated with Claude Code