Completion report · consolidated · 2026-07-20

BELIEVEYOUME corrections, with MIDI evidence

One authoritative record for this session: the two assignments, the source evidence behind the music corrections, every material edit, the live deployment proof, and what stays uncertain. Process narration and duplicate proof blocks are left out on purpose.

01The two assignments, kept separate

A. Anti-AI voice pass on the pages linked from /shiner-analysis-index, then on believeyoume-keys-modes. Deliverable: remove the tell corpus, keep the meaning.

B. Accuracy critiques applied to believeyoume-model-rebuild (the Asleep verse card) and believeyoume-ixi-overview. Deliverable: correct the facts, but verify each claim against the card, the tuning, and the source ahead of any edit.

The two never merged. The accuracy work was held until the source was checked, so the pages did not receive plausible-but-wrong corrections.

02MIDI evidence block (reproducible)

Every music correction on the Asleep verse card traces to this file. Re-run parse_evidence.py (music21) to regenerate the table.

FieldValue
Source fileShiner-Asleep In The Trunk-05-05-2026.mid the file the card cites
SHA-2563b87c550621b22e4519d41c528423a9a8e2414f1d2bbfac594701811370bde9e
Parsermusic21 9.9.1 · converter.parse · counts via recurse().notes (chord tones included)
Time signature3/4 throughout
Scriptscratchpad/parse_evidence.py (+ parse_asleep.py, first pass)

Tracks

#RoleNote events
0lead guitar485
1rhythm guitar1183
2bass1406
3drums1189
4vocal (on a sax patch)304

Bars 13-16 events (music21 measure numbers)

Trackm13m14m15m16
bassE♭2·F2·F2·F2·C2E♭2·F2·F2·F2·C2E♭2·F2·F2·F2·C2E♭2·F2·F2·F2·C2
rhythm gtr[E♭3,F♯3,E♭4][E♭3,F♯3,E♭4]×2[B♭2,C♯3,B♭3][B♭2,C♯3,B♭3]×2
vocal (sax)E♭4E♭4C♯4(rest)

Each conclusion, and the exact result that supports it

Conclusion on the cardSupporting dataScope
Bass is E♭–F–C (♭7-root-5), E♭ on every downbeat, not one chromatic visitBeat 1 (offset 0.0) is E♭2 in every measure m12-18; pitch classes present are E♭, F, C onlybass track, measures 12-18
The guitar's D♭ is real (the fret was wrong, not the pitch)rhythm gtr plays C♯3/D♭3 148 times; bars 15-16 carry the dyad [B♭2, C♯3, B♭3]whole song, includes chord tones and repeats
The A♮ major-third thesis is soundvocal lane carries 37 A3 eventswhole song, octave 3 only
"A3 appears twice" (not four times) in the excerptthe card's own bars 13-16 list shows A3 at bar 14 and bar 16from the card's list, not re-derived from the MIDI

What the evidence does and does not settle

03model-rebuild: the Asleep verse card

RemovedAddedLive check
F5 + D♭ Phrygian; A-string fret 1 = D♭3F + D♭ (♭6 over F); A-string fret 4 = D♭3 fret 1 sounds B♭; ♭6 ≠ Phrygian
bass F↔C ostinato; 5---5---0---0---; E♭ chromatic visit at 17.62E♭–F–C (♭7-root-5); 3---5---5---0---; E♭ diatonic ♭7 on every downbeat "F↔C ostinato" = 0 live
central pitch; A3 appears 4×salient inflection; A3 twice in the excerpt, 37 song-wide (Shiner-...-05-05.mid, section 02) "A3 appears twice" = 1 live
Phrygian tilt; falling back to B♭added-♭6 color; steps up to B♭ (A3 rises a semitone)

One of the sixteen cards on this page received the MIDI-checked pass. The other fifteen did not.

04keys-modes: anti-AI voice

Ten tells removed, meaning kept: names the bass to lists the bass (×2), three , not X antithesis tails cut, lean on the spine to rely on the spine, sitting on V to centered on the V. Live: names the bass = 0. Both HTML and md pass the voice gate.

05ixi-overview: voice pass, then accuracy pass

Accuracy corrections (verified this session)

RemovedAddedHow verified
self-titled 2025 album2025 albumBELIEVEYOUME ≠ the band name
six false roster claimsseven (+ the Alligator key)counted 7 fraud-table rows
guitar tab and bass tab in Drop Cthe song's verified tuning; My Mirror is standardMy Mirror sheet says "Standard tuning"
The 38-day storyThe month-long story (June 2 to July 1)shown dates span about a month
primary source (for the GP chart)derived transcription, checked against the audiorecording and stems are primary
the sung third decides tonic and qualityresolves quality, in the fuller contexta sung third settles quality, not tonic
47 live believeyoume slugs60-plus (62 route dirs at last count)counted 62
16 closed cream sheetsthe established iximusic reference barlive count did not resolve; number removed

The Asleep "C-E♭-F engine" the critique called an open conflict was settled by the parse of Shiner-Asleep In The Trunk-05-05-2026.mid in section 02; the overview was already right, the model-rebuild card carried the error. Kept as your ground truth, not deleted: the 59-chats and 13 / 6 backup figures.

06Deployment proof

[200] https://drwu-htmls.vercel.app/believeyoume-model-rebuild   md 200
[200] https://drwu-htmls.vercel.app/believeyoume-keys-modes      md 200
[200] https://drwu-htmls.vercel.app/believeyoume-ixi-overview    md 200
[200] https://drwu-htmls.vercel.app/shiner-anti-ai-audit        md 200
[200] https://drwu-htmls.vercel.app/  (site root, checked after each deploy)

A 200 proves the route is reachable. It does not prove the text rendered, so each removed phrase was also grep-checked to 0 and each added phrase to 1 on the live page, shown in the tables above.

07What the gate labels actually ran

Voice gate

python3 ~/.claude/skills/deslop/deslop.py <file>

Flags the tell corpus: em-dash, ellipsis, a set of banned filler words and metaphor verbs, antithesis tails, and instrument-refusal grammar. Exit 1 on any severity-3 or shape density over cap. Run prose-only (CSS and JS blanked) so pseudo-elements do not false-positive. Every edited file returned PASS.

Fraud gate

python3 ~/.claude/skills/session-fraud-check/session_fraud_check.py --hard-only

Hard-fails only on this session's authored files, scoped by edit time. That set is 10 files, not the 2 named pages: the four public pages plus their md, the shiner-anti-ai-audit page and md, and four scratchpad scripts. RESULT: PASS (remaining hits are on other sessions' files, report-only).

08What stays open