Re-pin four tests/ui assertions to the values the new defaults produce

A scoped exception to chunk 10a's ownership boundary: main must not be pushed
red, and deferring a behavioural fix into a merge commit is the pattern that
produced both bad merges on 2026-09-07. Only the four failing assertions are
touched — no matcher loosened, no test deleted, no other line in tests/ui/**,
and nothing under src/ui, src/demo, src/styles, index.html or tests/browser.
Every value below was read from a run, not computed by hand.

describe-diff.test.ts "describes what raising the investigation minimum
duration actually did" and parameters-panel.test.ts "the diff badge": both
pinned `strategy serial → random`. Under D58's rules the "after" analysis has
longestAdjacentRun 2 on holes [9,10], which clears serialMinRun 2, so both
sides are now serial and the badge reads `strategy unchanged`. The rest of
each assertion — −7 investigations, the total-errors clause — is untouched and
still agrees with the metrics beside it, which is what the tests are for.

describe-diff.test.ts "reports the quadrant time": 78.21 → 78.11 s and 178.37
→ 178.27 s. "reports the smoothed path and the mean speed": 391.30 → 390.91 cm
and 254.94 → 254.87 cm. Both moved by exactly the two frames D60 no longer
fills; the mean-speed clause in the same test was already right and is
unchanged.

Also: PARAMETER_DECISIONS pointed the five strategy thresholds at O7 and the
two gap-filling ones at O10, both of which now read "Superseded by" in
decisions.md, so the parameters panel and the XLSX sheet were sending a reader
to a superseded entry. They now name D58 and D60.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AiWHKiDDceNzuJdtFpcPkP

Author: alexbastuba

README.md

@@ -81,16 +81,11 @@ npm run typecheck
 npm run build   # static site in dist/, deployable as-is
 ```
 
-`npm test` runs 1,090 unit tests. Most cover the pure layers — metrics, event detection,
-cleaning, strategy, maze geometry, the export writers, the session file and the MP4 parser; a
-smaller set covers the DOM-free parts of the interface. Sixteen skip
-without `BARNESTRACK_SAMPLE_DIR` (below); fifteen of those are the sample-video tests and one skips
-for its own reason, so pointing that variable at the clips leaves a single skip. At the time of
-writing four fail: three in
-`tests/ui/components/describe-diff.test.ts` and one in `parameters-panel.test.ts`, all pinning
-numbers and a strategy class that the search-strategy and gap-filling defaults changed on
-2026-09-07. They are the interface chunk's to re-pin and are listed here rather than quietly
-excluded.
+`npm test` runs 1,090 unit tests, all passing. Most cover the pure layers — metrics, event
+detection, cleaning, strategy, maze geometry, the export writers, the session file and the MP4
+parser; a smaller set covers the DOM-free parts of the interface. Sixteen skip without
+`BARNESTRACK_SAMPLE_DIR` (below); fifteen of those are the sample-video tests and one skips for its
+own reason, so pointing that variable at the clips leaves a single skip and 1,089 passing.
 
 Two optional extras:
 

docs/known-limitations.md

@@ -20,6 +20,22 @@ session it is found (D39).
   is honest; the exported number is not obviously wrong to a reader who does not open the card.
   Smallest fix: flag when the marked frame's last positioned event point is within the entry radius
   of a hole that is not the target, naming that hole — a second branch beside the one D57 added.
+- **The two halves of the definitions panel read differently.** Since chunk 10a the 24 parameter
+  definitions end in their unit alone, while every metric definition still ends in a decision id —
+  "…never from frame 0 (seconds; O5)." — because `tests/ui/review-format.test.ts` pins
+  `METRIC_DEFINITIONS[key]` to `/\(.+; [OD]\d+\)\.$/` and that file belongs to the interface
+  chunk. A user reading the panel sees this project's internal numbering on one half and not the
+  other. Smallest fix: strip the suffixes from `METRIC_DEFINITIONS`, give `METRIC_DECISIONS` an
+  explicit table as `PARAMETER_DECISIONS` now has, and re-point that assertion.
+- **`gapFilling.enabled` is not a `trials.csv` column, though `gap_fill_max_duration_s` is.** D60
+  made the on/off switch the load-bearing one — filling is off by default — and the column set
+  carries only the ceiling it would apply if it were on. Two cohorts differing solely in that
+  setting are distinguishable by `parameters_hash` and by `parameters.json`, both of which travel
+  in every export, so nothing is unreconcilable; but a reader comparing threshold columns alone
+  sees identical rows. Not fixed here because a new column means regenerating
+  `skills/barnestrack-cohort/SKILL.md` and its binding test for a value the hash already covers.
+  Smallest fix: a `gap_fill_enabled` bool column in `thresholdColumns()`
+  (`src/export/rows.ts`) and `TRIAL_COLUMN_META`, plus that regeneration.
 - **Every maze nudge re-derives the whole cohort, from a step the user is not looking at.** The app
   shell refreshes every step on every store notification, and the Review step re-derives whatever
   the store invalidated so its figures and its export line are not left claiming that analysed

src/analysis/parameters.ts

@@ -205,19 +205,20 @@ const ANALYSIS_PARAMETER_DECISIONS: Record<AnalysisParameterPath, string> = {
   trialCutoff_s: 'O5',
   'targetQuadrant.holeSpan': 'O6',
   kinematicsSmoothingWindowFrames: 'O9',
-  'gapFilling.enabled': 'O10',
-  'gapFilling.maxDuration_s': 'O10',
+  // D60 superseded O10, D58 superseded O7: the panel must not name a superseded entry
+  'gapFilling.enabled': 'D60',
+  'gapFilling.maxDuration_s': 'D60',
   'kinematics.speedWindowFrames': 'O11',
   'kinematics.duplicateTimestampFactor': 'O11',
   'kinematics.dropGapFactor': 'O11',
   noseConfidenceCutoff: 'O16',
   outlierVelocityThreshold_cmPerS: 'O17',
   'trialCensoring.censorToCutoff': 'O5',
-  'strategy.spatialMaxErrors': 'O7',
-  'strategy.spatialMaxHoleDistance': 'O7',
-  'strategy.spatialMaxCentreCrossings': 'O7',
-  'strategy.serialMinRun': 'O7',
-  'strategy.centreZoneRadiusFraction': 'O7',
+  'strategy.spatialMaxErrors': 'D58',
+  'strategy.spatialMaxHoleDistance': 'D58',
+  'strategy.spatialMaxCentreCrossings': 'D58',
+  'strategy.serialMinRun': 'D58',
+  'strategy.centreZoneRadiusFraction': 'D58',
   'quality.goodMinPositionedFraction': 'D30',
   'quality.poorMaxPositionedFraction': 'D30',
 };

tests/analysis/parameters.test.ts

@@ -109,7 +109,10 @@ describe('definitions', () => {
     }
     // the strategy rules cite Gawel et al. 2019, Table 1 rather than D58
     expect(PARAMETER_DEFINITIONS['strategy.serialMinRun']).toContain('Gawel et al. 2019, Table 1');
-    expect(PARAMETER_DECISIONS['strategy.serialMinRun']).toBe('O7');
+    // the decision each default *now* comes from: O7 and O10 were superseded by D58 and D60, and
+    // the panel must not send a reader to a superseded entry
+    expect(PARAMETER_DECISIONS['strategy.serialMinRun']).toBe('D58');
+    expect(PARAMETER_DECISIONS['gapFilling.enabled']).toBe('D60');
     expect(PARAMETER_DECISIONS['quality.goodMinPositionedFraction']).toBe('D30');
     expect(PARAMETER_DECISIONS['trialCensoring.censorToCutoff']).toBe('O5');
   });

tests/ui/components/describe-diff.test.ts

@@ -352,7 +352,7 @@ describe('describeDiff over a real re-derivation', () => {
     expect(sentence).toContain(
       `total errors ${before.metrics.totalErrors} → ${after.metrics.totalErrors}`,
     );
-    expect(sentence).toContain('strategy serial → random');
+    expect(sentence).toContain('strategy unchanged');
   });
 
   it('reports the quadrant time, which only the quadrant span moves', () => {
@@ -364,7 +364,7 @@ describe('describeDiff over a real re-derivation', () => {
       targetQuadrant: { holeSpan: 9.5 },
     }));
     expect(after.metrics.targetQuadrantTime_s).toBeGreaterThan(before.metrics.targetQuadrantTime_s);
-    expect(describeDiff(before, after)).toContain('target quadrant time 78.21 s → 178.37 s');
+    expect(describeDiff(before, after)).toContain('target quadrant time 78.11 s → 178.27 s');
   });
 
   it('reports the smoothed path and the mean speed, which only the filter width moves', () => {
@@ -373,7 +373,7 @@ describe('describeDiff over a real re-derivation', () => {
       kinematicsSmoothingWindowFrames: 31,
     }));
     const sentence = describeDiff(before, after);
-    expect(sentence).toContain('smoothed path length 391.30 cm → 254.94 cm');
+    expect(sentence).toContain('smoothed path length 390.91 cm → 254.87 cm');
     expect(sentence).toContain('mean speed 2.19 cm/s → 1.43 cm/s');
   });
 

tests/ui/components/parameters-panel.test.ts

@@ -433,7 +433,7 @@ describe('the diff badge', () => {
     );
     const badge = container.querySelector('.diff-badge')!;
     expect(badge.textContent).toContain('−7 investigations');
-    expect(badge.textContent).toContain('strategy serial → random');
+    expect(badge.textContent).toContain('strategy unchanged');
     panel.destroy();
   });