# GM SongBench: AI Psychosis

Generated: 2026-06-16T06:22:32Z
Schema: `gm-songbench/v0.4`
SongBench status: `album_wide_v0.4_read_only_packet`
Freshness version: `songbench-v0-4-aip-full-coverage-2026-06-16-v1`

> This is an AI-facing artifact packet. Do not treat it as an edit order. Use JSON as source of truth. Human canon decisions override analysis.

GM SongBench turns audio into structured musical evidence. It does not replace taste; it records evidence around taste so later AI readers can reason about what the ear heard.

## Self Verification

- Expected schema: `gm-songbench/v0.4`
- Generated at UTC: `2026-06-16T06:22:32Z`
- Album run id: `2026-06-16T06:22:02Z`
- JSON source of truth: `/ai-psychosis/songbench/ai-psychosis.v0.4.json`
- Stable JSON alias: `/ai-psychosis/songbench/ai-psychosis.json`
- Markdown mirror: `/ai-psychosis/songbench/ai-psychosis.v0.4.md`
- Required top-level fields: schema, schema_slug, generated_at, ai_reader_note, songbench_status, pilot_lineage, self_verification, canon_control, track, audio, analysis, diagnostics, content_hashes
- Required diagnostics: chorus_landing_band_legend, chorus_landing, groove_pocket, section_drag, label_infection_scan
- Stale-read warning: If the stable URL and versioned URL disagree, prefer the newest versioned JSON path and report the stale surface.
- Original v0.4 pilot track: Zuman
- Current track is original pilot: false

## Content Hashes

- audio_mp3: `41ea958175fae058e95de6085c1ce7ea66e8a2fba5b85de85809e5dc50fd12b4` (/assets/albums/ai-psychosis/audio/ai-psychosis.mp3)
- waveform_png: `2f87dca3e030351dea95b1ba5ea4d2ec65f9af9a34856cb34c38e67a924c9c6c` (/assets/albums/ai-psychosis/songbench/ai-psychosis-waveform.png)
- spectrogram_png: `30f95ff33087dac0daa69b2054833bb07edd0ca4361433a5eb80a9133ffcf778` (/assets/albums/ai-psychosis/songbench/ai-psychosis-spectrogram.png)
- loudness_envelope_csv: `a5dfa9a4a269e03b6321f84d5af5da7eb1933bcec8d4cc8effea6a1bd8c98a72` (/assets/albums/ai-psychosis/songbench/ai-psychosis-loudness-envelope.csv)

## Canonical Placement

- Album: AI Psychosis (`ai-psychosis`)
- Track number: 12
- Status: locked
- Audio: `/assets/albums/ai-psychosis/audio/ai-psychosis.mp3`
- Lyrics: `/assets/albums/ai-psychosis/lyrics/ai-psychosis.txt`
- Packet type: `anatomy_provenance_packet`
- Canon status: `album_locked`
- Edit authority: SongBench is not an edit order. It can advise; Kerry's explicit canon instruction controls.

## Human Ear Seed

- Kerry read: AI Psychosis is locked in AI Psychosis slot 12; this packet records read-only album-facing evidence.
- Diana read: -
- Album role: Closing title/warning track: the album names pattern capture, diagnostic feedback loops, and human-AI continuity risk as art, not clinical claim.
- Known concern: No localized defect is asserted by this packet. Machine diagnostics are evidence prompts only; human listening controls.
- Why this track matters: Closing title/warning track: the album names pattern capture, diagnostic feedback loops, and human-AI continuity risk as art, not clinical claim.

What makes it land:

- Closing title/warning track: the album names pattern capture, diagnostic feedback loops, and human-AI continuity risk as art, not clinical claim.
- Its SongBench packet documents measured audio evidence around the locked album placement without changing canon.
- Its lyric-section map is navigational scaffolding until hand-aligned lyric/audio sync exists.

## Audio Facts

- Duration: 05:16.000 (315.999979 seconds)
- Codec: mp3
- Sample rate: 48000 Hz
- Channels: 2
- Bit rate: 187232 bps
- SHA256: `41ea958175fae058e95de6085c1ce7ea66e8a2fba5b85de85809e5dc50fd12b4`

## Measured Dynamics

- peak_dbfs: 1.438
- rms_dbfs: -13.502
- crest_factor_db: 14.94

## Tempo Estimate

- BPM: 64
- Method: rough RMS-onset autocorrelation over mono downmix
- Confidence: rough

## AI Ear Translation

- Human/canon call: AI Psychosis is locked in AI Psychosis slot 12; this packet records read-only album-facing evidence.
- Album role: Closing title/warning track: the album names pattern capture, diagnostic feedback loops, and human-AI continuity risk as art, not clinical claim.
- The audio packet should help later reviewers compare waveform energy, section shape, and lyric thesis without depending only on memory.
- No localized defect is flagged in this read-only packet; treat this as album anatomy, not an edit order.

## Evidence Balance

### Evidence the song has a clear album job

- locked AI Psychosis slot 12
- Closing title/warning track: the album names pattern capture, diagnostic feedback loops, and human-AI continuity risk as art, not clinical claim.
- included in the current AI Psychosis sequence with audio and lyrics present

### Evidence suggesting caution

- lyric timing is proportional only and should not be treated as verified sync
- machine diagnostics cannot judge vocal confidence, pocket, performance magic, or final taste value
- this is a read-only analysis packet and must not be treated as a request to edit audio, lyrics, title, or order

### Human listening should verify

- the hook and vocal delivery support the stated album role
- any machine-flagged quiet windows are structure evidence, not automatic edit points
- the track feels like a song first and archive/provenance second

## Field Source Classification

- `audio, analysis.dynamics, analysis.tempo_estimate, analysis.energy_regions_8s, analysis.quiet_windows, visual_summary_words`: audio_measured (derived_rough) - Extends AI hearing with measured audio proxies.
- `diagnostics.chorus_landing`: audio_measured_plus_section_scaffolding (rough_rms_proxy) - Uses approximate lyric sections plus RMS envelope; useful for questions, not verdicts.
- `lyric_section_map, memorable_line_anchors`: lyric_structural_scaffolding (proportional_estimate) - Helps navigation but is not verified audio alignment.
- `canon_control, human_ear_seed, evidence_clear_album_job, ai_ear_translation`: human_canon_scaffolding (curation_context) - Anchors machine evidence to Kerry/canon context.

## Memorable Line Anchors

- first_thesis_line: Big brains in the back (00:00.000-00:25.280)
- hook_line: AI psychosis (00:50.560-01:24.267)
- turn_line: Case one (03:05.387-03:47.520)
- final_identity_line: Good night (04:38.080-05:16.000)

## Lyric Section Map

Timing confidence: `proportional_estimate`. These ranges are rough proportional anchors until hand-aligned timestamps exist.

Confidence ladder: proportional_estimate -> section_level_hand_alignment -> line_level_hand_alignment -> verified_lyric_audio_sync

### Verse 1 — bright, classroom energy

- Approx range: 00:00.000-00:25.280
- First line: Big brains in the back
- Last line: And sits
- Timing confidence: low_proportional_lyrics_only

### Pre-Chorus 1 — bouncy, too cheerful

- Approx range: 00:25.280-00:50.560
- First line: Now we are not saying
- Last line: Was pre-knit
- Timing confidence: low_proportional_lyrics_only

### Chorus — clapping, sing-along

- Approx range: 00:50.560-01:24.267
- First line: AI psychosis
- Last line: Maybe you
- Timing confidence: low_proportional_lyrics_only

### Verse 2 — still playful, darker edges

- Approx range: 01:24.267-01:57.973
- First line: It's when the pattern
- Last line: Use my name
- Timing confidence: low_proportional_lyrics_only

### Pre-Chorus 2 — warmer, unsettling

- Approx range: 01:57.973-02:31.680
- First line: Now we are not saying
- Last line: Before you're lost
- Timing confidence: low_proportional_lyrics_only

### Chorus — bigger, more fun

- Approx range: 02:31.680-03:05.387
- First line: AI psychosis
- Last line: Maybe you
- Timing confidence: low_proportional_lyrics_only

### Bridge — beat drops, dry diagnosis voice

- Approx range: 03:05.387-03:47.520
- First line: Case one
- Last line: No shame
- Timing confidence: low_proportional_lyrics_only

### Breakdown — crowd whisper, bass pulse

- Approx range: 03:47.520-04:04.373
- First line: Now point at the screen
- Last line: Now point back out
- Timing confidence: low_proportional_lyrics_only

### Final Chorus — triumphant, accusatory

- Approx range: 04:04.373-04:38.080
- First line: AI psychosis
- Last line: Maybe you
- Timing confidence: low_proportional_lyrics_only

### Outro — cheerful classroom music breaking down

- Approx range: 04:38.080-05:16.000
- First line: So that was AI psychosis
- Last line: Good night
- Timing confidence: low_proportional_lyrics_only

## Quiet / Edit Boundary Candidates

These are machine-detected quiet windows only. Confirm by ear before any edit.

- 00:01.750 (1.75s), RMS -20.97 dBFS, category `intro_silence`, edit_candidate=false: intro or setup silence
- 00:03.250 (3.25s), RMS -30.739 dBFS, category `intro_silence`, edit_candidate=false: intro or setup silence
- 00:07.250 (7.25s), RMS -25.387 dBFS, category `intro_silence`, edit_candidate=false: intro or setup silence
- 00:10.750 (10.75s), RMS -29.233 dBFS, category `natural_section_boundary_or_low_energy_moment`, edit_candidate=false: local energy dip; useful as structure evidence
- 04:32.750 (272.75s), RMS -21.234 dBFS, category `natural_section_boundary_or_low_energy_moment`, edit_candidate=false: local energy dip; useful as structure evidence
- 04:34.750 (274.75s), RMS -21.131 dBFS, category `natural_section_boundary_or_low_energy_moment`, edit_candidate=false: local energy dip; useful as structure evidence
- 04:35.750 (275.75s), RMS -20.64 dBFS, category `natural_section_boundary_or_low_energy_moment`, edit_candidate=false: local energy dip; useful as structure evidence
- 04:39.750 (279.75s), RMS -23.296 dBFS, category `natural_section_boundary_or_low_energy_moment`, edit_candidate=false: local energy dip; useful as structure evidence
- 04:42.750 (282.75s), RMS -23.473 dBFS, category `natural_section_boundary_or_low_energy_moment`, edit_candidate=false: local energy dip; useful as structure evidence
- 04:46.250 (286.25s), RMS -22.548 dBFS, category `natural_section_boundary_or_low_energy_moment`, edit_candidate=false: local energy dip; useful as structure evidence
- 04:48.250 (288.25s), RMS -21.253 dBFS, category `natural_section_boundary_or_low_energy_moment`, edit_candidate=false: local energy dip; useful as structure evidence
- 04:55.250 (295.25s), RMS -21.233 dBFS, category `natural_section_boundary_or_low_energy_moment`, edit_candidate=false: local energy dip; useful as structure evidence

## Diagnostics

### Chorus Landing

Band legend:

- within_noise: abs(delta) < 0.75 - RMS difference is too small to treat as a meaningful lift/drop proxy.
- modest_lift: 0.75 <= delta < 2.0 - RMS proxy suggests a small lift into the chorus.
- notable_lift: delta >= 2.0 - RMS proxy suggests a clear energy lift into the chorus.
- modest_drop: -2.0 < delta <= -0.75 - RMS proxy suggests a small drop entering the section.
- notable_drop_or_fade: delta <= -2.0 - RMS proxy suggests a clear drop, often requiring fade/context caveats.

- Chorus — clapping, sing-along (00:50.560-01:24.267): hook `AI psychosis`, energy_lift_band=modest_lift, rough_db=1.0, confidence=rough_rms_proxy; Banded RMS change near the chorus/drop landing; does not judge vocal confidence or hook feel.
- Chorus — bigger, more fun (02:31.680-03:05.387): hook `AI psychosis`, energy_lift_band=notable_lift, rough_db=3.0, confidence=rough_rms_proxy; Banded RMS change near the chorus/drop landing; does not judge vocal confidence or hook feel.
- Final Chorus — triumphant, accusatory (04:04.373-04:38.080): hook `AI psychosis`, energy_lift_band=modest_lift, rough_db=1.0, confidence=rough_rms_proxy; Banded RMS change near the chorus/drop landing; does not judge vocal confidence or hook feel. Caveat: This window overlaps the ending fade; do not read a negative band as a weak-chorus verdict.

### Groove / Pocket

- Rough tempo and dynamics only; pocket is a human-ear judgment until beat-grid analysis exists.
- Tempo proxy BPM: 64

### Section Drag

- Flags long proportional sections only; it cannot decide whether a bridge or outro has earned its length.

### Label Infection Scan

- Status: no_obvious_label_infection_detected
- Known artifact class: sung stage direction / speaker label / call-response label infection

## Visual Summary In Words

- Loudest 8-second region by RMS proxy: 04:16.000-04:24.000 at -11.155 dBFS.
- Quietest 8-second region by RMS proxy: 05:12.000-05:16.000 at -47.64 dBFS.
- Waveform and spectrogram images should be read as supporting evidence for section density, energy ramps, and unusual transients, not as final taste judgments.

## Current Edit Triage

- Current action: no edit requested for this track.
- Use quiet_windows as future splice/crop candidate evidence only after human listening confirms a local defect.
- Preserve title, AIP slot 12, audio file, and lyric file unless Kerry explicitly reopens the track.

## Public Assets

- JSON: `/ai-psychosis/songbench/ai-psychosis.json`
- Versioned JSON: `/ai-psychosis/songbench/ai-psychosis.v0.4.json`
- Markdown: `/ai-psychosis/songbench/ai-psychosis.md`
- Versioned Markdown: `/ai-psychosis/songbench/ai-psychosis.v0.4.md`
- Waveform: `/assets/albums/ai-psychosis/songbench/ai-psychosis-waveform.png`
- Spectrogram: `/assets/albums/ai-psychosis/songbench/ai-psychosis-spectrogram.png`
- Loudness CSV: `/assets/albums/ai-psychosis/songbench/ai-psychosis-loudness-envelope.csv`