feat(closed loop): bias/disturbance waveforms for FicTrac trials (LAB-185)

.claude/skills/protocol-yaml/SKILL.md

@@ -115,14 +115,48 @@ streams frames for the wait's duration. Declare a `fictrac` plugin. Shape:
       frame_index: 0
       frame_rate: 0        # mode 3: fixed 0
       gain: 0              # mode 3: fixed 0
-    - type: "controller"
+    - type: "plugin"        # NOT "controller" — these are fictrac PLUGIN commands
+      plugin_name: "fictrac"
       command_name: "startClosedLoop"
     - type: "wait"
       duration: *cl_dur    # FicTrac drives frames for this long (same anchor)
-    - type: "controller"
+    - type: "plugin"
+      plugin_name: "fictrac"
       command_name: "stopClosedLoop"
 ```
 
+`startClosedLoop` / `stopClosedLoop` are **plugin** commands on the declared `fictrac`
+plugin. Typing them as `controller` makes the runner skip them with a warning, so the
+loop never opens and the trial silently runs open-loop.
+
+### Bias / disturbance waveforms (LAB-185)
+
+`startClosedLoop` takes three optional params that add a smooth disturbance to the
+loop, so the display keeps moving even when the fly holds still (disturbance
+rejection). Authored as an added rotational **velocity**:
+
+```yaml
+    - type: "plugin"
+      plugin_name: "fictrac"
+      command_name: "startClosedLoop"
+      params:
+        gain: 1.8
+        bias_type: "sine"   # none | constant | sine | square
+        bias_amplitude: 90  # PEAK velocity, deg/s — negate to reverse direction
+        bias_frequency: 0.5 # Hz — sine/square only; ignored by constant
+```
+
+- `constant` drifts steadily; `sine`/`square` are **zero-mean in position**, so the
+  display is pushed equally both ways and starts where it already was.
+- Amplitude is a *velocity*, so the position excursion is derived and **shrinks as
+  frequency rises**: `±A/(2πf)` for sine, `±A/(4f)` for square. At 90 deg/s a sine
+  covers ±28.6° at 0.5 Hz but only ±14.3° at 1 Hz.
+- `bias_frequency` **must be non-zero** for `sine`/`square` — 0 Hz fails the step.
+  A negative frequency is a harmless no-op (warned): negate `bias_amplitude` instead.
+- `stopClosedLoop` clears the bias automatically. Reference:
+  `docs/development/closed-loop-bias.md`; worked example:
+  `protocols/fictrac_bias_test.yaml`.
+
 ### Conditional LED activation (index-gated LED, Mode 3 only)
 
 A Mode-3 `trialParams` may carry an optional **`led_activation`** attribute to drive

.claude/skills/protocol-yaml/bin/validate-protocol.mjs

@@ -68,6 +68,7 @@ for (const w of warnings) console.warn('⚠ ' + (w && w.message ? w.message : w)
 const num = (v) => (Number.isFinite(Number(v)) ? Number(v) : 0);
 let lintCount = 0;
 const lint = (msg) => { lintCount++; console.warn('⚠ waits: ' + msg); };
+const lintBias = (msg) => { lintCount++; console.warn('⚠ bias: ' + msg); };
 
 for (const cond of conds) {
     const cmds = cond.commands || [];
@@ -109,6 +110,43 @@ for (const cond of conds) {
     }
 }
 
+// ── BIAS SANITY (LAB-185) ───────────────────────────────────────────────────
+// The bias params live on the fictrac plugin's startClosedLoop, not on trialParams,
+// so this needs its own pass over plugin commands. Mirrors the runner's rules
+// (js/arena-runner-g6.js normalizeBias) so an author hears about it here rather than
+// at run time, where a bad spec skips the trial — possibly with a fly already mounted.
+const BIAS_TYPES = ['none', 'constant', 'sine', 'square'];
+for (const cond of conds) {
+    for (const c of cond.commands || []) {
+        if (c.type !== 'plugin' || c.command_name !== 'startClosedLoop') continue;
+        const p = c.params || {};
+        if (p.bias_type === undefined || p.bias_type === null || p.bias_type === '') continue;
+        const t = String(p.bias_type).trim().toLowerCase();
+        const label = '"' + cond.name + '" startClosedLoop';
+        if (!BIAS_TYPES.includes(t)) {
+            lintBias(label + ' has bias_type ' + JSON.stringify(p.bias_type) +
+                ' — must be one of ' + BIAS_TYPES.join('/') + '. The runner will skip this step.');
+            continue;
+        }
+        if (t === 'sine' || t === 'square') {
+            const f = num(p.bias_frequency);
+            if (p.bias_frequency === undefined || f === 0) {
+                lintBias(label + ' is bias_type ' + t + ' with bias_frequency ' +
+                    (p.bias_frequency === undefined ? 'unset' : f) +
+                    ' — 0 Hz has no period, so the runner will SKIP this step. ' +
+                    'Set a non-zero frequency (or use bias_type constant for a steady drift).');
+            } else if (f < 0) {
+                lintBias(label + ' has a negative bias_frequency (' + f + ') — that is a no-op, ' +
+                    'the waveform is even in frequency. Negate bias_amplitude to reverse direction.');
+            }
+            if (num(p.bias_amplitude) === 0) {
+                console.log('ℹ ' + label + ' is bias_type ' + t +
+                    ' with bias_amplitude 0 — no disturbance will be applied.');
+            }
+        }
+    }
+}
+
 if (!blocking.length && !warnings.length && !lintCount) {
     console.log('✓ clean — no blocking errors, no warnings, waits rule satisfied.');
 } else if (!blocking.length) {

.github/workflows/validate-fictrac-bridge.yml

@@ -0,0 +1,56 @@
+name: Validate FicTrac Bridge
+
+# The bridge owns two pieces of pure math the closed loop depends on:
+#   • behavior_v1_row()  — the col-22 ns→ms normalization the live scope and the
+#     offline dashboard both read.
+#   • bias_angle_deg()   — the closed-loop bias/disturbance waveforms (LAB-185):
+#     the closed-form integrals and the zero-mean-position property.
+# tests/test-bridge-behavior.py covers both and runs in `pixi run test`, but until
+# this workflow it ran in NO CI job — so a regression in either only surfaced on a
+# developer's machine.
+
+on:
+  push:
+    branches: [main]
+    paths:
+      - 'fictrac-bridge/**'
+      - 'tests/test-bridge-behavior.py'
+      - '.github/workflows/validate-fictrac-bridge.yml'
+  pull_request:
+    branches: [main]
+    paths:
+      - 'fictrac-bridge/**'
+      - 'tests/test-bridge-behavior.py'
+  workflow_dispatch: # Allow manual triggering
+
+jobs:
+  validate:
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v7
+
+      - name: Set up Python
+        uses: actions/setup-python@v6
+        with:
+          python-version: '3.12'
+
+      - name: Install bridge dependencies
+        # bridge.py imports websockets at module scope, so importing it for the
+        # offline tests needs the package even though no socket is opened.
+        run: pip install 'websockets>=12'
+
+      - name: Run bridge behavior + bias tests
+        run: python tests/test-bridge-behavior.py
+
+      - name: Bridge CLI smoke test
+        # --version proves the module imports and argparse is wired; the bad-bias
+        # case proves the 0 Hz guard rejects before any socket is opened.
+        run: |
+          python fictrac-bridge/bridge.py --version
+          if python fictrac-bridge/bridge.py --bias-type sine --bias-freq 0 2>/dev/null; then
+            echo "FAIL: --bias-type sine with --bias-freq 0 should be rejected"
+            exit 1
+          fi
+          echo "OK: 0 Hz sine rejected at the CLI"

CLAUDE.md

@@ -251,6 +251,32 @@ fix flows to every page automatically; two hand-written HTML pages never will.
   when adding encoders/decoders, add them to the export list AND a test; audit
   with `Object.keys(require('./js/arena-wire-g6.js'))` vs the page's `Wire.*`
   uses (a missing export throws silently inside async handlers).
+- **Closed-loop bias (LAB-185):** a disturbance waveform added to the FicTrac
+  closed loop so the display moves even when the fly is still. Authored
+  per-condition on the `fictrac` plugin's `startClosedLoop`
+  (`bias_type`/`bias_amplitude`/`bias_frequency`) — deliberately NOT in the
+  plugin's `configFields` and NOT a Console input, so there is one source of
+  truth. The MATH lives in the bridge (`bias_angle_deg` in
+  `fictrac-bridge/bridge.py`, pure + Python-tested); the runner only validates
+  (`normalizeBias` → `{bias, warning}`) and pushes it as bridge config. GOTCHAS:
+  (1) the `none|constant|sine|square` vocabulary is declared in THREE places —
+  bridge `BIAS_TYPES`, runner `BIAS_TYPES`, registry `bias_type.options` — and a
+  registry test pins them equal; (2) `stopClosedLoop` MUST push
+  `bias:{type:'none'}` or the waveform keeps accumulating into the frame index
+  through later trials — and because a mid-trial STOP never REACHES it, the loop
+  is torn down twice over: `_clearClosedLoop()` (setApply(false) + bias none) is
+  called from ALL THREE teardown paths (`runSequence` finally, `stop()`,
+  `_clear()`/`abort()`, symmetric with `_clearLedActivator()`) and
+  `startClosedLoop` ALWAYS carries an explicit bias so an epoch can never inherit
+  a stale one. Call `_clearClosedLoop()` from any NEW teardown path;
+  (3) `bias` is the one OBJECT-valued key in
+  `FicTracBridgeClient.setConfig`, whose scalar keys gate on `Number.isFinite`;
+  (4) never widen `BEHAVIOR_V1_COLS` to log it — the per-frame `bias` on the
+  WebSocket is display-only, and offline reconstruction uses the `bias_config`
+  log event, which `js/runlog-replay.js` surfaces as a status event
+  (`status.phase === 'bias_config'`, spec on `status.bias`) so `buildTimeline`
+  and every `status.phase === '…'` consumer keep working untouched.
+  Full spec: `docs/development/closed-loop-bias.md`.
 - URL state ([#107](https://github.com/reiserlab/webDisplayTools/issues/107),
   read+write): `js/studio-url-state.js` (`mode` ∈ run|edit|console; a shared
   `p` forces `edit`→Run on fresh loads, never `console`). Write side:

arena_studio.html

@@ -306,6 +306,8 @@
   /* closed-loop label: very light by default, bright-green pulsing when active */
   .run-bridge .rb-cl{margin-left:auto;font-family:var(--head);font-size:11px;letter-spacing:.4px;color:var(--dim);opacity:.45;transition:color .2s,opacity .2s}
   .run-bridge .rb-cl.active{color:var(--accent);opacity:1;text-shadow:0 0 8px rgba(0,230,118,.7);animation:pulse 1.1s infinite}
+  /* closed-loop bias readout — read-only mirror of what the protocol pushed (LAB-185) */
+  .run-bridge .rb-bias{font-size:11px;color:var(--dim);white-space:nowrap}
   /* quick safety button: drive the stimulus LED dark (AO 5000 mV) — pink */
   .pill.led-off{border-color:#ff408c;color:#ff6fa8;font-weight:700;letter-spacing:.3px}
   .pill.led-off:hover{background:rgba(255,64,140,.16);border-color:#ff408c;color:#ff8fbd}
@@ -2220,6 +2222,7 @@
           <span class="dim" id="rbStatus">disconnected</span>
           <span class="dim rb-count" id="rbCount"></span>
           <span class="rb-cl" id="rbClosedLoop" title="Closed-loop (FicTrac Mode 3/4) — lights green while a trial or experiment is driving the arena from FicTrac">closed-loop</span>
+          <span class="dim rb-bias" id="rbBias" title="Closed-loop bias — the disturbance waveform the running protocol pushed to the bridge, and the angle it is currently adding. Set per condition in the protocol (fictrac → Start closed-loop)."></span>
           <button class="pill led-off" id="btnLedOff" title="Turn the stimulus LED off now — sets Analog Out to 5000 mV (BuckPuck dark). Works during a run.">TURN OFF LED</button>
         </div>
       </div>
@@ -2917,7 +2920,7 @@ <h2 id="modalTitle">Import error</h2>
 </div>
 
 <footer id="footer">
-  <span class="foot-left">Arena Studio v0.69 | 2026-07-21 16:46 ET · <a href="https://github.com/reiserlab/webDisplayTools" target="_blank" rel="noopener" title="webDisplayTools on GitHub: source, issues, and release notes (docs/development/arena-studio-release-notes.md)">GitHub</a></span>
+  <span class="foot-left">Arena Studio v0.70 | 2026-08-04 11:47 ET · <a href="https://github.com/reiserlab/webDisplayTools" target="_blank" rel="noopener" title="webDisplayTools on GitHub: source, issues, and release notes (docs/development/arena-studio-release-notes.md)">GitHub</a></span>
   <!-- Course-repo quick-links: open protocols / logs / patterns in a new tab.
        Hrefs built from the configured repo + bench id (updateGhQuickLinks). -->
   <span id="ghQuickLinks" title="Open the course repo on GitHub (new tab)">
@@ -3096,7 +3099,7 @@ <h2 id="modalTitle">Import error</h2>
         }
         const div = document.createElement('div');
         if (ev.phase === 'error') div.className = 'l-err';
-        else if (ev.phase === 'skip') div.className = 'l-warn';
+        else if (ev.phase === 'skip' || ev.phase === 'warn') div.className = 'l-warn';
         // Reuse run-log.js's own transcript line if available, else a compact fallback.
         div.textContent =
             (ev.t_iso ? ev.t_iso.slice(11, 19) + '  ' : '') +
@@ -5745,19 +5748,34 @@ <h2 id="modalTitle">Import error</h2>
         // sync (both subscribe to the one client's events), even when switching
         // views. Only connect + status + live counts + a closed-loop indicator
         // are exposed here; gain / closed-loop config stay in the Console.
+        // Bias is READ-ONLY here: it is authored per condition in the protocol
+        // (fictrac → Start closed-loop), so this only mirrors what the runner pushed
+        // plus the live angle the bridge reports on each frame. Blank when inactive.
+        function biasText() {
+            const b = bridge.bias;
+            if (!b || b.type === 'none') return '';
+            const head =
+                b.type === 'constant'
+                    ? 'constant ' + b.amplitude + '°/s'
+                    : b.type + ' ' + b.amplitude + '°/s @' + Math.abs(b.frequency) + ' Hz';
+            const live = bridge.biasAngleDeg;
+            return 'bias: ' + head + (live == null ? '' : ' · ' + (live > 0 ? '+' : '') + live.toFixed(1) + '°');
+        }
         function refreshRunBridge() {
             const on = bridge.connected;
-            const dot = $('rbDot'), conn = $('rbConnBtn'), st = $('rbStatus'), cnt = $('rbCount'), cl = $('rbClosedLoop');
+            const dot = $('rbDot'), conn = $('rbConnBtn'), st = $('rbStatus'), cnt = $('rbCount'), cl = $('rbClosedLoop'), bias = $('rbBias');
             if (dot) dot.classList.toggle('on', on);
             if (conn) conn.textContent = on ? 'Disconnect' : 'Connect';
             if (st) st.textContent = on ? 'connected' : 'disconnected';
             const s = bridge.stats;
             if (cnt) cnt.textContent = (on && s && s.recv) ? 'rx ' + s.recv + ' · applied ' + s.applied : '';
             if (cl) cl.classList.toggle('active', on && bridge.apply); // 'apply' = closed-loop live
+            if (bias) bias.textContent = on ? biasText() : '';
         }
         bridge.on('status', refreshRunBridge);
         bridge.on('stats', refreshRunBridge);
         bridge.on('apply', refreshRunBridge);
+        bridge.on('bias', refreshRunBridge);
         if ($('rbConnBtn')) $('rbConnBtn').addEventListener('click', () => {
             if (bridge.connected) { bridge.disconnect(); }
             else { ftSendConfig(); bridge.connect(($('cFtUrl').value || 'ws://localhost:8765').trim()); }
@@ -14903,6 +14921,7 @@ <h2 id="modalTitle">Import error</h2>
     '#mNotes': 'Anything worth remembering about this run — temperature, fly age, oddities.',
     '#provBox': 'Filled in automatically — which protocol, arena, and controller produced this run.',
     '#runBridge': 'The FicTrac bridge program records runs and drives closed-loop trials. Connect it before a recorded run.',
+    '#rbBias': 'The bias (disturbance) the running protocol added to the closed loop, and the angle it is adding right now. A bias keeps the display moving even when the fly holds still. Set it per condition in Edit → the condition’s Start closed-loop command.',
     /* File ▾ — relocated Edit actions */
     '#fmOpenDemo': 'Open one of the bundled demo protocols to explore how experiments are built.',
     '#fmSaveCopy': 'Download a regenerated copy of the protocol for diffing or MATLAB checks — this does NOT count as saving.',

docs/development/arena-studio-release-notes.md

@@ -4,6 +4,43 @@ The Studio's footer used to carry the full changelog inline; it now shows one li
 history lives here. Newest first. (Per-session engineering detail stays in
 `arena-studio-handover.md` and the design docs — this file is the user-facing what-changed list.)
 
+## v0.70 (2026-08-04) · Closed-loop bias — add a disturbance to a fly-on-ball trial
+
+- **A closed-loop condition can now add a smooth disturbance to the visual
+  display**, so the arena keeps moving even when the fly holds still. This is
+  the stimulus for disturbance-rejection experiments: impose a known motion and
+  see how the fly counter-turns to null it. Set it per condition in Edit → the
+  condition's **Start closed-loop** command, which grows three fields:
+  - **Bias waveform** — None, Constant (steady drift), Sine, or Square.
+  - **Bias amplitude (deg/s, peak)** — how fast the display is pushed. A
+    **positive** amplitude turns the display **clockwise** (equivalently, the
+    pattern sweeps rightward across the fly's view); negative turns it
+    counter-clockwise. Confirmed on the arena.
+  - **Bias frequency (Hz)** — for Sine and Square only; Constant ignores it.
+- **Constant** drifts the display steadily in one direction. **Sine** and
+  **Square** push it equally left and right around the frame it started on, so
+  they add no net rotation over a trial. Because the amplitude is a *speed*, how
+  far the display travels shrinks as the frequency rises: at 90 deg/s a sine
+  covers ±28.6° at 0.5 Hz but only ±14.3° at 1 Hz.
+- **The Run view shows the live bias** next to the closed-loop indicator — the
+  waveform the running protocol asked for, plus the angle it is adding right now
+  — so you can confirm at the bench that it took effect. It is read-only there;
+  the protocol is the only place it is set.
+- **Stop closed-loop clears the bias**, so a disturbance never leaks into
+  following trials — and so does **pressing STOP mid-run**, which skips the
+  protocol's own Stop closed-loop entirely. Without that, the next run started
+  with the previous run's disturbance still going (and the bridge kept driving
+  the arena after STOP). Bench-reported and fixed before release.
+- A ready-made sweep of all four waveforms ships as **FicTrac closed-loop
+  bias/disturbance test** in File ▾ → Open from library.
+- Needs an updated FicTrac bridge (`pixi run bridge`, reports version 2.1 or
+  later) — the bias is computed there. An older bridge simply ignores it (verified:
+  a stale bridge keeps streaming normally, it just applies no disturbance).
+- **Validated on arena hardware.** A full 8-condition run drove a real controller
+  and the recorded log matches the intended frame index on every one of its 328,733
+  frames; the rotation direction was checked by eye. Not yet used in an experiment
+  with a behaving fly.
+
 ## v0.69 (2026-07-21) · ISP batch retries a failed panel twice
 
 - **A failed panel flash now gets up to two retries** (was one) before being