feat(theory): noise-tolerant loop detection rewrite (task L-30)

detectRepeatingProgression redesigned: consecutive-dup collapse,
weak-period candidate rejection (kills self-overlap ghosts
structurally), lengths 2-8, <=1 substitution/insertion per cycle with
a >=2-exact-occurrences evidence gate, recency-weighted linear
coverage scoring. All 17 C-30 fixtures pass plain (smoke 817/817,
expectedFail markers removed). Critic PASS: 15 novel generalization
probes correct or data-faithful, 0.2-0.7 ms/call, diff confined to the
repeating-progression section, fixture contract untouched.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

scripts/loop-fixtures.mjs

@@ -12,11 +12,12 @@
 //   · fixture with expectedFail that PASSES    → smoke FAILS: stale marker, flip it
 // L-30's definition of done = every expectedFail marker removed, all green.
 //
-// ─── FAILURE MAP OF THE CURRENT ALGORITHM ────────────────────────────────────
-// (theory.js `detectRepeatingProgression`, every fixture RUN against it 2026-07-10;
-//  the `today:` comment on each expectedFail fixture is the actual observed output)
+// ─── FAILURE MAP OF THE PRE-L-30 ALGORITHM (historical — all fixed by L-30) ──
+// (theory.js `detectRepeatingProgression` as of C-30, every fixture RUN against
+//  it 2026-07-10; L-30 redesigned the function and removed every expectedFail
+//  marker — all 17 fixtures now pass as plain regression guards)
 //
-// The current matcher: window = last 32 commits; candidate lengths 2–6; EXACT
+// The pre-L-30 matcher: window = last 32 commits; candidate lengths 2–6; EXACT
 // contiguous occurrence counting (scan advances by len on match, by 1 on miss);
 // requires reps ≥ 2; score = reps × len²; returns the canonical (lexicographically
 // smallest) rotation of the best candidate.
@@ -130,87 +131,54 @@ export const LOOP_FIXTURES = [
     expect: null,
   },
 
-  // ── Expected failures of the current algorithm — the L-30 contract ─────────
+  // ── Failure modes of the pre-L-30 algorithm — fixed by L-30 ────────────────
   {
     id: 'vamp-2x4',
     description: 'clean 2-chord vamp played 4× — must still report the pair, not a self-overlap',
     history: reps(['Am', 'G'], 4),
     expect: ['Am', 'G'],
-    // today: returns [Am,Am,G] — the 3-chord self-overlap [Am,G,Am] scores
-    // 2 reps × 3² = 18 and beats the true pair at 4 × 2² = 16 (failure map #1).
-    expectedFail: true,
   },
   {
     id: 'vamp-2-sustained',
     description: 'sustained 2-chord vamp filling the whole window (×16) — the everyday two-chord jam',
     history: reps(['Am', 'G'], 16),
     expect: ['Am', 'G'],
-    // today: returns [Am,Am,G,Am,G] — a bogus 5-chord self-overlap of the pair
-    // wins on len² (failure map #1). A musician vamping Am–G sees a fake
-    // 5-chord progression.
-    expectedFail: true,
   },
   {
     id: 'spurious-substitution',
     description: '3-chord loop ×4 with ONE substituted misdetection (Am read as E7 in the third rep) — must still report the 3-loop',
     history: ['C', 'Am', 'F', 'C', 'Am', 'F', 'C', 'E7', 'F', 'C', 'Am', 'F'],
     expect: ['C', 'Am', 'F'],
-    // today: returns [Am,F,F,C] — the wrong 4-pattern [F,C,Am,F] straddling the
-    // noise scores 2 × 4² = 32 and beats the real loop's 3 × 3² = 27
-    // (failure map #2). This is the user's "doesn't recognize when 3 chords
-    // return": one bad commit and the display shows a 4-chord ghost.
-    expectedFail: true,
   },
   {
     id: 'spurious-2of4',
     description: '3-chord loop ×4 with an inserted misdetection in two different reps (E7, then Dm7) — realistic sustained noise',
     history: ['C', 'Am', 'F', 'C', 'Am', 'E7', 'F', 'C', 'Am', 'F', 'C', 'Dm7', 'Am', 'F'],
     expect: ['C', 'Am', 'F'],
-    // today: returns [Am,F,C,C] — a wrong 4-pattern beats the true 3-loop once
-    // noise appears in more than one rep (failure map #3).
-    expectedFail: true,
   },
   {
     id: 'dup-commit',
     description: 'consecutive duplicate commit of the same chord inside an otherwise clean 3-loop (C Am F C C Am F …) — dups must collapse',
     history: ['C', 'Am', 'F', 'C', 'C', 'Am', 'F', 'C', 'Am', 'F'],
     expect: ['C', 'Am', 'F'],
-    // today: returns [Am,F,C,C] — the duplicate mints a wrong 4-pattern
-    // [C,Am,F,C] at 2 × 16 = 32 vs the real loop's 27 (failure map #4).
-    // App.jsx:322-324 currently suppresses adjacent dup commits, so this exact
-    // history can't arise from today's commit layer — the fixture pins L-30's
-    // "collapse consecutive duplicates before matching" so the detector is safe
-    // standalone (and safe if L-31 changes the commit layer).
-    expectedFail: true,
   },
   {
     id: 'seven-x2',
-    description: '7-chord loop played 2× — beyond the current length-6 cap',
+    description: '7-chord loop played 2× — beyond the pre-L-30 length-6 cap',
     history: reps(['Em', 'G', 'D', 'A', 'Em', 'C', 'B7'], 2),
     expect: ['Em', 'G', 'D', 'A', 'Em', 'C', 'B7'],
-    // today: returns [A,Em,C,Em,G,D] — a truncated 6-chord slice of the loop,
-    // because candidate lengths cap at 6 (failure map #5). The user's
-    // "5 and then 2 others" = a 7-chord form is structurally undetectable.
-    expectedFail: true,
   },
   {
     id: 'eight-x2',
-    description: '8-chord loop (extended andalusian form) played 2× — beyond the current cap',
+    description: '8-chord loop (extended andalusian form) played 2× — beyond the pre-L-30 cap',
     history: reps(['Am', 'G', 'F', 'E7', 'Am', 'C', 'Dm', 'E7'], 2),
     expect: ['Am', 'G', 'F', 'E7', 'Am', 'C', 'Dm', 'E7'],
-    // today: returns [Am,C,Am,G,F,E7] — again a wrong 6-chord truncation
-    // (failure map #5).
-    expectedFail: true,
   },
   {
     id: 'section-change',
     description: 'section change: 4-chord loop A ×3, then 3-chord loop B ×3 — must report B, the loop being played NOW',
     history: [...reps(['C', 'G', 'Am', 'F'], 3), ...reps(['Dm7', 'G7', 'Cmaj7'], 3)],
     expect: ['Dm7', 'G7', 'Cmaj7'],
-    // today: returns [Am,F,C,G] — the STALE loop A: no recency weighting, so the
-    // old section's 3 × 4² = 48 outscores the current section's 3 × 3² = 27
-    // (failure map #6). The display stays stuck on the previous section.
-    expectedFail: true,
   },
 ]