CC 6.1.2.1.2: ratify the shipped n_eff dominance gate (spec catches up to the wire — #6)

[emooreatx]

What

The Constitution's normative text lagged the shipped wire on a MUST-bearing surface: shipped persist rejects version-1 AggregationMetaV1 tiers at aggregated-tier admission (fail-closed dominance gate, CIRISVerify#167 / CIRISPersist#357), while CC 6.1.2.1 still pinned version = 1 with no n_eff. An implementation conforming to the text alone would be rejected by the substrate. This PR ratifies the shipped v2 surface and pins the constants the wire already runs on.

Changes

• CC 6.1.2.1 — preimage ratified at version 2: trailing u32_be(n_eff), append-only (v1 preimages are byte-identical to the original layout, so all pre-existing signatures/vectors verify unchanged — new vector, not a re-cut). New normative versioning paragraph: v1 tiers MUST be rejected at admission.
• New CC 6.1.2.1.2 dominance-gate (spine key 19.7.1.2, matching the impl's own section number): n_eff = round((Σ m_i)² / Σ m_i²) (Kish / inverse-Simpson over per-member content masses), gate n_eff ≥ min_ratio · source_count with CEG-pinned min_ratio = 0.5 — a normative constant, not an implementation knob (closing the same two-conformant-implementations-disagree defect Noise floor lacks an adversary/prior model — "MUST be below the floor" is unverifiable (aggregation ≠ anonymization) #6 raised about ε; persist's CHANGELOG had flagged that CC didn't pin it). Gate runs after signature verification, before persistence; stable rejection token aggregation_meta_dominated.
• Honest-limits paragraph (normative honesty) — the two accepted residuals stated in place: (1) masses are aggregator-attested, not on the wire — a lying n_eff is admitted (accountable-signature trust, slashable post-hoc; the conformance suite pins this deliberately); (2) mass-dominance ≠ content-similarity — 900 near-duplicates under distinct ids at equal mass honestly pass. Post-1.0 fallback: content-similarity N_eff (lead CIRISVerify#191; fixture CIRISConformance#71 — Evidence sweep as of ciris-server 0.5.152: nine claims are established on a closed issue, and ten shipped controls have verified symbols #55 closed as harness-complete).
• CC 6.1.2 shortcut sentence now points at the normative gate; the conformance note records the append-only canonical_bytes_v2 vector.
• R9 rewritten: mass-dominance moved from known exposure to mitigated since RC2; residual narrowed to the three items above.
• claims.tsv: CLM-noise-dominance open → established (test_542_dominance_gate + CIRISServer#155, both backed by the pinned manifests at decimal 6.1.2); new CLM-neff-dominance-gate (staged — flips established when the CIRISServer/CIRISConformance manifests refresh with the 6.1.2.1.2 decimal) and CLM-noise-content-multiplicity (open, the R9 residual, tracked by CIRISVerify#191).
• Dual-ID spine: toc.tsv + all five codebook maps (n_concepts 429 → 430); Σ added to the PDF Unicode map; PDF rebuilt.

Verification

• tools/check_claims.py: structurally valid, coverage 133/133 (100%), established 118 / staged 29 / open 1; the only new warnings are the expected pending-manifest ones for decimal 6.1.2.1.2.
• build_pdf.py: builds clean.
• Wire facts verified against shipped code: preimage + gate semantics from CIRISVerify/src/ciris-verify-core/src/holonomic/aggregation.rs (signing_preimage, effective_source_count, passes_dominance_gate), admission enforcement + MIN_DOMINANCE_RATIO = 0.5 from CIRISPersist/src/fountain/aggregation.rs, v2 vector CIRISPersist/tests/vectors/holonomic_v19_7/aggregation_meta/canonical_bytes_v2.json.

Closes the mass-dominance half of #6; the disposition comment on #6 covers the rest.

🤖 Generated with Claude Code

Residual cleanup (second commit)

CIRISConformance#55 and CIRISEdge#266 turned out to be closed, so the residual had no live tracker. Opened and wired in: CIRISVerify#191 (lead — v3 wire surface: content-similarity N_eff + auditable masses), CIRISEdge#323 (fold-time masses/multiplicity in aggregate_symbols), CIRISPersist#430 (admission gate extension), CIRISConformance#71 (the 900-near-duplicates-under-distinct-ids fixture — absent from every suite today), CIRISServer#239 (un-ignore the faithful aggregation-erasure measurement now that edge#266 shipped; fix the stale test-(d) tripwire).

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[emooreatx]

Superseded by the rc3 release branch (cut from claude/rc3-cut-summary-lo4me4 + commit 6175ffb). RC3 ratifies the wire at v3 directly (flag-day, version < 3 inadmissible), which strictly contains this PR's v2 ratification; 6175ffb carries this PR's surviving content forward — the CC 6.1.2.1.2 spine section (n_eff definition + pinned min_ratio = 0.5), the R9 rewrite, and the claims rows — and fixes two byte-order errors in the cut's v3 preimage listing against the shipped wire + golden vector. Nothing from this branch is lost.