What changed in every version of the standard.

Score delta: <3 points on the golden-eval set.

• M8 — Voice & POV Integrity

  Changed

  Refactored from "one coherent narrator" to "INTENTIONAL POV." Deliberate narrator switches that align with structural beats (K-pop multi-voice, hip-hop featured-verse contrast, gospel call-and-response) no longer score as drift failures. Random narrator switches without semantic function still do.

• M11 — Memorability

  Changed

  Refactored from a single 60-minute recall test to a 4-signal cumulative read: hook integration + phonemic distinctiveness + chorus repetition strategy + one-listen recall. Cumulative / oral-tradition / ritual-repetition forms are no longer false-positive low-scored. Mantric chants that earn their repetition through structural intention now register at the canonical band.

• Refactor descriptive, not prescriptive

  Note

  Both M8 and M11 changes describe WHEN the metric applies; they don’t change WHAT the metric values. Score deltas on the golden-eval set average under 3 points — within the MINOR threshold per RFC-0001.

Score delta: <2 points on the golden-eval set.

• Anti-Platitude rule

  Added

  Anti-inflation rules expanded from 4 to 5. New rule: lines that resolve with generic emotional summaries ("all I need is love," "this is my truth") hit the rubric’s lowest Specificity + Voice band regardless of surface polish.

• Antagonist Ceiling rule

  Changed

  Clarified to require evidence. The dedicated critical voice that challenges every score now must produce specific lines / specific failure modes — vague disagreement does not lower the score.

• Historical Context Anchor rule

  Changed

  Anchored explicitly to the published corpus, especially the Hank Williams S-band entry at composite 95. "Professional craft" now points at a citeable artifact, not an abstract claim.

• Migration

  Note

  Existing scored content auto-rescores on next eval. The reproducibility seal’s rubricVersion field now reads "1.1.0."

Ratifying RFC: RFC-0002-anti-platitude-rule

• Reproducibility seal in API

  Added

  The seal field landed in /api/v1/score (B1199). Every score response now carries the ed25519-signed envelope binding rubricVersion + model + temperature + buildSha + composite + grade.

• Public model card

  Added

  Published at /scoring/standard/model-card (B1197). Pins the runtime contract: which Anthropic model, what temperature, where the prompt lives, what limitations apply.

• Versioning policy formalized

  Added

  RFC-0001 documented the version-bump rules: MAJOR for >5-point golden-eval delta, MINOR for clarifications, PATCH for docs/typos. Every later bump follows this contract.

• No rubric content changes

  Note

  PATCH bumps don’t change the rubric itself — same metrics, same weights, same anti-inflation rules. Score deltas: zero.

Ratifying RFC: RFC-0001-rubric-versioning-policy

• 12 metrics across 3 tiers

  Added

  Initial public release. Craft (25%): M1 Prosody, M2 Structure, M3 Rhyme, M4 Economy. Expression (40%): M5 Specificity, M6 Imagery, M7 Emotional Truth, M8 Voice. Impact (35%): M9 Transcendence, M10 Arc, M11 Memorability, M12 Genre.

• 4 anti-inflation rules

  Added

  Gravity (default 50), Burden of Proof (claims earned by lines that show), Antagonist Ceiling, Historical Context. Anti-Platitude rule (5th) was added in the 1.1.0 bump.

• Grade scale

  Added

  Locked: S+ (96+), S (91-95), A+ (86-90), A (80-85), B+ (73-79), B (65-72), C+ (55-64), C (45-54), D+ (35-44), D (25-34), F (<25). Same scale used by every later version.