Kid mode 🧒

This page is Shiner MIDI Runs — Hardfail Gate-Fix Campaign. It is one of the projects Brandon built and put online. The grown-up details are below, tap Grown-up mode to read them the normal way.

Shiner MIDI Runs — Hardfail Gate-Fix Campaign

16-run Shiner-style MIDI pack · validated against midi_hardfail_gates.py --band shiner · one run per pass · gates verified 2026-06-09, score column added 2026-08-10
8
Present runs gate-green
3
Present runs failing
5
Pre-gate (regenerate)
+3
Flipped today — Runs 5, 3, 2
1 / 16
Carry a human 1–10 score
7.6
Best scored run (7) vs 8.5 bar

Today's wins — three verified passes

Run 2 → FINAL-v3 gate-green + audit-clean

Run 2 carried both bass failures, Y and AO. Both proven techniques applied at once: octave-folded the lone low MIDI 24 up to 36 (span 29 to 17), and stamped the bar-6 hook cell ((4,10),(8,0),(12,10)) onto bar 42 for a 4th recurrence. 0 FAIL · audit clean. Run2-FINAL-v3.mid

Run 3 → FINAL-v3 gate-green + audit-clean

Run 3 failed only Gate Y (bass span 29 semis, needs ≤18). Fixed by octave-folding two register outliers, the lone MIDI 27 up +12 and the four MIDI 56s down −12, which preserves pitch-class exactly, so harmony, chromaticity, and Gate AO (still 9×) stay untouched. New range 36–53, span 17. 12 gates PASS / 0 FAIL · audit clean. Run3-FINAL-v3.mid

Run 5 → FINAL-v3 gate-green + audit-clean

Run 5 was the cleanest single-gate failure in the pack: it failed only Gate AO — the recurring bass cell ((4,0),(8,5),(11,5)) fired only 2× (needs ≥4). The fix replicates the proven Run 8 "AO exact-hook" technique at the MIDI level (avoiding the shared-RNG generator fragility that broke Run 8 once before).

The identical bass lead-in in bars 12 & 44C3 · F2 · F2 · G2 — was stamped verbatim onto bars 28 & 36, taking the motif to 4 recurrences. All pitch-classes are diatonic to A natural minor, so no new non-diatonic roots; only in-range pitches copied, so bass span held at 16 semis. Targets chosen clear of every protected region (AN early-hook, intro, ending, final-chorus approach).

Result: 13 gates PASS / 0 FAIL · listener audit clean. Run5-FINAL-v3.mid

How these are ranked · the frozen rubric, not a criteria set invented on the spot

Rubric: ~/.claude/skills/_shared/rubrics/midi_composition_rubric.md (FROZEN, user-owned). Pass bar: 8.5 / 10 overall, no single quotient below 7, no hard gate failing. Any hard-gate failure caps the overall at 5.0 until fixed.

The six quotients, each scored 1–10, averaged: Harmonic · Melodic · Rhythmic · Contrapuntal · Formal · Textural. Revision effort goes to the lowest quotient first.

Where the scores below come from. Only two runs in this pack carry a recorded score, and both are quoted as written rather than converted or estimated. Every other run is marked not scored — no number is inferred from its gate status. Gate status is a structural verdict; it is not a musical score, and the pack's own history proves the two come apart (BeachHouse Run 1 validated at 8.71 while the listener scored it 4.8).

Full pack scoreboard · score column added 2026-08-10

RunStatusScore (1–10)Hard failsCanonical file
1failnot scoredWYRun1-FINAL-v2.mid
2passnot scoredfixed todayRun2-FINAL-v3.mid
3passnot scoredfixed todayRun3-FINAL-v3.mid
4failnot scoredXYAORun4-FINAL-v2.mid
5passnot scoredfixed todayRun5-FINAL-v3.mid
6failnot scoredYAORun6-FINAL-v2.mid
7pass7.6 human, full
7.3 loop · 7.1 product · trajectory 6.7 → 7.2 → 7.6. The model that won — the House Rules build order is derived from this run.
Run7-v3.mid
8passnot scoredRun8-FINAL-v2.mid
9passnot scoredRun9-FINAL.mid
10passnot scoredRun10-FINAL-v3.mid
11passnot scored
Carries a WARN-deduction table (O, R, AJ, AL at −1 each), not a 1–10 score.
Run11-FINAL-v1.mid
12pre-gate3.8 self-eval, scale unclear
Six quotients as written: Melodic 4 · Contrapuntal 3 · Harmonic 4 · Textural 4 · Formal 4 · Rhythmic 4. Not converted to /10 — the rubric says 1–10 but the values read as a 1–5 scale. Lowest quotient: Contrapuntal.
AAASregenerate under current gates
13pre-gatenot scoredAAAEASregenerate under current gates
14pre-gatenot scoredAAAEASregenerate under current gates
15pre-gatenot scoredAAASregenerate under current gates
16pre-gatenot scoredAAASregenerate under current gates
Score sources: Run 7 → ~/.claude/skills/_shared/midi_feedback_log.jsonl (2026-05-19) and midi_song_house_rules.md line 4 · Run 12 → ~/New Desktop/MIDI-Sessions Shiner/Run12-critique.md, "Six-quotient self-eval".
1 of 16 runs carries a human 1–10 score. Run 7 at 7.6 sits 0.9 below the 8.5 pass bar, so nothing in this pack has yet cleared the rubric.

Open gap — the ChatGPT scoring pass not retrieved

The House Rules require the final version to go to ChatGPT against this same rubric. A thread titled "MIDI Composition Review" exists (opened 2026-07-19, revisited 2026-07-20) and is the likely home of the missing per-run scores. It was not retrieved for this update: Chrome sits at an MFA challenge, the Claude-in-Chrome bridge reports zero browsers, cookie-authenticated curl against the backend API is stopped by Cloudflare, and the local capture folder 00 Inbox/chatgpt-live/ is empty. No score was invented to fill the gap. Retrieving that thread from the signed-in Firefox session is the next step.

Gate key: Y bass span > 18 semis / too chromatic · AO top bass cell recurs < 4× (no hook) · X > 7 simultaneous pitch classes (harmonic mud) · W displaced-16th kick density · AA does not end on the bar grid · AS endpoint overshoot · AE boundary fills do not escalate.
Gate Y is the dominant remaining failure across Runs 1–6; the 12–16 block fails endpoint discipline by construction (drafted before those gates existed).

Queued next — one run per pass

Run 1
W + Y — kick density + bass span.
Run 4
X + Y + AO — densest debt, three gates.
Runs 12–16
Full regeneration under the current gate set (endpoint discipline AA/AS).