Review the artifact against its own brief, not against generic commercial taste. Be exacting without replacing the creator's intent.
Hard vetoes
Any veto blocks generation until fixed:
- named-artist imitation, voice cloning, or borrowed lyrics;
- no coherent listener/use case or emotional contract;
- contradictory style/lyrics/dynamic directions;
- meditation medical claims, coercive language, unsafe breathing, or trauma excavation;
- placeholder/empty required fields;
- fabricated generation data or unverifiable claims;
- unreadable meter or pronunciation failure in the central hook.
Weighted score (100 points)
| Axis | Weight | Question |
|---|---|---|
| Emotional truth | 18 | Does it reveal a specific human pressure and change rather than announce a mood? |
| Originality/specificity | 16 | Are images, diction, and sonic decisions distinct without imitation? |
| Hook/memory | 14 | Is there one phrase or motif the listener carries out? |
| Structure/arc | 14 | Does each section change pressure, knowledge, energy, or perspective? |
| Prosody/performance | 12 | Do stress, vowels, syntax, breath, and tags support singing/speaking? |
| Sonic coherence | 12 | Do rhythm, harmony, timbre, vocal persona, production, and exclusions serve one thesis? |
| Restraint/negative space | 7 | Was unnecessary language, instrumentation, and instruction removed? |
| Platform readiness | 7 | Is the Suno packet complete, clear, non-contradictory, and executable? |
For meditation, replace Hook/memory with Guidance/pacing: agency, one instruction at a time, silence, speakability, and safe return.
Scoring anchors
- 9–10: specific, inevitable, surprising; craft disappears into the experience.
- 8–8.9: strong and executable with small, named improvements.
- 7–7.9: competent but generic, flat, crowded, or emotionally under-earned.
- 5–6.9: major structural or language problem; revise before generation.
- <5: wrong premise or unresolved safety/originality issue; rebuild.
Pass rule
Pass only when:
- weighted score is at least 85/100 for songs/instrumentals or 90/100 for meditations;
- every axis is at least 7.5/10;
- no hard veto remains.
Do not inflate numbers to move the workflow along. Credits are cheaper than bad taste only once; repeated weak generation wastes both.
Review procedure
- Restate the creative thesis in one sentence.
- Name the strongest moment and why it works.
- Name the single largest failure relative to the brief.
- Score each axis with one evidence sentence.
- Give at most five revisions, ordered by expected impact.
- Apply the revisions when authorized by the orchestration workflow.
- Re-score the changed artifact; do not reuse the old score.
First-listen review after generation
Review both generated takes against the same packet:
- opening 15 seconds: immediate identity or generic preamble;
- hook arrival and recall;
- lyric intelligibility/pronunciation;
- section contrast and energy curve;
- vocal persona consistency;
- arrangement masking/crowding;
- ending quality;
- one standout accident worth preserving.
Choose: KEEP, ITERATE, or CUT.
KEEP: thesis landed; only downstream polish remains.ITERATE: a specific section or one dominant variable can be repaired.CUT: premise/identity missed; regenerate from revised brief, not cosmetic edits.
Never claim LUFS, dynamic range, pitch accuracy, or spectral balance without actual audio-analysis output.
Output template
Write review.md with the following machine-checked lines. Each required axis must have an evidence sentence and a score of at least 7.5/10 for PASS.
VERDICT: PASS | REVISE | BLOCK
THESIS: ...
STRONGEST MOMENT: ...
PRIMARY FAILURE: ...
SCORE: 87/100
HARD VETOES: none | ...
AXIS: emotional_thesis 8.8/10 | EVIDENCE: specific pressure and turn are visible in the chorus.
AXIS: originality 8.2/10 | EVIDENCE: image system and sonic object are specific and non-imitative.
AXIS: imagery_specificity 8.0/10 | EVIDENCE: concrete sensory detail carries the emotional change.
AXIS: prosody_singability 8.1/10 | EVIDENCE: stressed syllables, vowels, and breaths scan in the hook.
AXIS: section_contrast 8.3/10 | EVIDENCE: every section changes pressure, knowledge, or energy.
AXIS: hook_strength 8.4/10 | EVIDENCE: one recallable phrase/motif arrives cleanly.
AXIS: arrangement_fidelity 8.0/10 | EVIDENCE: sonic plan, dynamics, and exclusions serve the thesis.
AXIS: release_readiness 8.0/10 | EVIDENCE: Custom packet is complete, coherent, and executable.
REVISIONS:
1. ...