Narrative Registry

The canonical brand-narrative truth SSOT — the eighth protocol-layer skill, peer of entity-optimizer (SEO/GEO), creator-registry (influencer), offer-claims-registry (paid), consent-registry (email), launch-registry (launch), channel-registry (social), and the cross-discipline memory-management — and the record the TALE A1 (canon-integrity) veto is judged against. It CURATES the narrative record — registry, not gate: no class: auditor, no cap fields, no veto scoring, no NQS roll-up. It stores dated facts; narrative-quality-auditor judges A1 against those facts, exactly as launch-readiness-auditor judges R1 against stage records.

One durable canon per brand (canon.md) holds: the positioning statement, the main narrative (old world → the shift → the new game → the promised land → proof), the three value pillars each carrying its claim IDs (pointers into the offer-claims ledger — never the adjudicated wording itself), the brand voice rules (register, tone, banned phrases, few-shots from own material), the naming tax (product/feature/tier naming rules, approved and banned terms), and the 25 / 50 / 100-word boilerplate set. Beside it sit two standing files: versions.md (the append-only version history — the narrative-whiplash fact base) and candidates.md (the intake file). This registry writes the human-facing canon that every narrative and channel discipline expresses downstream.

Versioned-canon atomic promotion: a canon re-version supersedes atomically — when canon.md is re-cut, the entire prior version is appended to versions.md (with its version tag and date) before canon.md is rewritten. The prior version is never edited in place; history is append-only. A partial edit that leaves canon.md internally contradictory is exactly the A1 canon-integrity state the gate reads — so a re-version lands as one coherent replacement, not a patchwork.

Scope seams — who keeps what:

• The A1 canon-integrity verdict and the NQS stay with narrative-quality-auditor; this registry supplies the canon, voice, naming, and pillar facts — never a go/no-go or a "coherent narrative" label. No canon record on file = NEEDS_INPUT, not pass-by-default (the same red line as the A1 row in TALE).
• Authoring the durable message house stays with message-system-architect; the narrative arc with strategic-narrative-designer; voice + naming rules with brand-language-codifier; the differentiation truth set with positioning-truth-tracer. This registry records what those skills authored, not what to author.
• Machine-facing entity facts (schema, sameAs) stay with entity-optimizer — narrative-registry owns the human-facing canon (positioning statement, boilerplate, voice); the entity descriptions derive from it, they do not redefine it.
• Per-platform voice adaptation stays with channel-registry's voice-dossier.md — that record points up to this brand-voice canon and never redefines it (channel voice is an adaptation of the canon here).
• Claim adjudication stays with offer-claims-registry — the sole adjudicator. Pillars in canon.md carry claim IDs only; this registry NEVER decides whether a claim is substantiated. Unverified claim wording is routed to memory/claims/candidates.md, never entered as fact.
• Archival stays with memory-management — the sole WARM → COLD executor; the canon is standing state, superseded by a dated re-version, never on a timer.

Scope guard: this skill records narrative facts only. It does NOT compute the NQS, run the T1/A1/L1/E1 vetoes, author the message house, or adjudicate a claim — those belong to narrative-quality-auditor, message-system-architect, and offer-claims-registry respectively. Never fabricate a canon: absence of a canon.md is a fact (NEEDS_INPUT), not an implied "the brand has no narrative problem".