Elnor Repo Reader

ELNOR REPO READER TEXT MIRROR
Original path: Memory Rebuild Docs/Stage_6_Charters/E1_E2_DOC81_Scope_Policy/E1_E2_R3_CODEX_Application_Audit_Prompt.md
Source repo: /Users/OpenClaw1/Elnor/Elnor Specs
Git branch: main
Git commit: dbaa25962edc11ab30e8d4ca1715f9ae5bf77331
Generated: 2026-06-09T01:23:58.539Z

---

# E1/E2 (DOC81) — R3 Application-Fidelity Audit Commission (CODEX)

**Repo:** github.com/wbrody/Elnor-Specs — branch `main` · **Date:** 2026-06-07 · **Architect:** Will Brody.
**You are:** an independent auditor verifying the **R3 application** of the R2 design-review adjudication landed **exactly** as the card directs — nothing skipped, improvised, or merged-wrong, no architect ruling diluted, no decline reintroduced. **Fidelity, not design** (design is settled; log any design concern as `DESIGN_ADVISORY`, don't act).

**Why this audit:** the applier (Claude Code) self-reports R3 complete + self-audited clean. Self-reports get checked here — the **R2** report falsely claimed regression check-6 passed when bare-string IDs remained; an independent CODEX pass caught it. Stale/over-claimed checks and **merge-seam defects** (where two review bodies were combined without re-deriving the result — the source of both prior R2 fidelity defects) are your priority class.

## 1. Inputs (paths under `Memory Rebuild Docs/Stage_6_Charters/E1_E2_DOC81_Scope_Policy/`; read directly — if no repo access, stop and say so)
1. `DOC81_Scope_Policy_Charter_Draft.md` — the R3 draft (~2,682 lines). THE TARGET.
2. `E1_E2_R2_Adjudication_Card.md` — THE AUTHORITY: §1 clusters V1–V24, §1.3 minors, §2 seeds, §3 declines D-R1–D-R6, §4 plan. **The card's §4 architect confirms are approved — verify they landed.**
3. `E1_E2_R3_Application_Commission_Claude_Code.md` — the commission (cluster plan, hard constraints, regression + meta-checklist).
4. `E1_E2_R3_Application_Report.md` — the applier's self-report. **Verify; do not trust.**
5. Source bodies: `Reviews/E1_E2_R2_Design_Review_Claude_Opus_4.8.md` (CL), `Reviews/RT-0001_CHATGPT_CHATGPT_CONSOLIDATED_REVIEW.md` (GPT).
6. Ground truth: `E1_E2_Adjudication_Card.md` (R1 — §1-bis R-1…R-5 still binding); ratified E0 (`../E0_DOC80_Core/DOC80_Core_Charter_Draft.md` §2.2/§7.1/§22); Owner Map; Round D.

## 2. What to check
**2.A Mapping table.** Re-derive every **V1–V24, each §1.3 minor, each §2 seed row, D-R1–D-R6** against the actual R3 text: landing exists at the claimed location, matches the card's merge decision (not just the raw review body), PASS/FAIL per row with `§+line`. Any claimed-but-absent or materially-divergent landing is a finding.

**2.B The merge seams (highest value — re-derive end-to-end, don't pattern-match).** For V1, V3, V11, V15, V16, V17 the card MERGES CL+GPT bodies. Walk each merged result yourself:
- **V1 unified pipeline:** `PolicyPoint` = 4 capability axes only (no `disclosure_class`); disclosure travels ONLY in the vector; scalar derived EXACTLY ONCE at finalize; `makeCoherent` last. Walk the meet steps 0→7 in order: generation/freshness → 0b egress structural → 0c crosswalk consumed → 1 applicable (+F1 +scalar/vector mismatch guard) → 1b egress-applicable → 1c empty-applicable internal baseline → 2 domain-as-state → 3 action-aware floor → 4 sticky (ordinary only) → 5 obligations → 6 finalize → 7 action predicate. Confirm no step meets disclosure as a bare scalar anywhere.
- **V3 `deriveDisclosureClass`:** never falls through to `full`; `source_title ∨ count=exact ∨ summary=full_reason → full`; `generic_reason_only`/`redacted_reason → redacted_summary`. Enumerate all four `SummaryFidelity` values and confirm each maps conservatively.
- **V11/V15/V16/V17:** the disclosure-rendering merge, equivalence cluster lifecycle+spanning, cascade freeze-gate+formula+lawful-edge predicate, and the QF helper pack each present and internally consistent.

**2.C The walked cases (the R-1/R-2 executability proof).** Walk against the FINAL text and report outcomes: (a) CL scenario (d) — same-principal cross-matter INTERNAL retrieve returns full content (not floored to reference_only); (b) GPT empty-applicable internal retrieve — domain-profile baseline, NOT `policy.no_applicable_decision` block; (c) CL scenario (e) — an autonomous restamp after a lifted transient floor RESTORES up to ceiling via `restamp_reeval` mode (not stuck at the prior lower value); (d) the V1 leak case — a scope FLOOR's disclosure ceiling survives to the final scalar (the A-B1 class). Each must resolve per the card; a failure is Blocking.

**2.D Declines + rulings (binding non-changes).** D-R1 (zero `RestampIssuance`/second-object — the MFC field is an optional post-issuance field on the existing restamp); D-R2 (no `E0DerivedProjection` E0-named base — `persistence_kind` marker only); D-R3 (`MAX_SCOPE_TRAVERSAL_DEPTH = 16`); D-R4 (capability-obligation = emission lint only, no runtime step-5b); D-R5 (V21 wall items non-gating, one-line invariants); D-R6 (V1 is present as the headline). R1 declines D1–D12 still hold (spot-check: no `risk_level: unknown` swap, no `visible_action_disposition` field, crosswalk is a constraint table). R-1 four-basis exhaustiveness, R-2 firewalled-only authority, R-5 principal keys.

**2.E Regression checklist (re-run independently, PASS/FAIL + evidence).** The commission §5 eleven points — esp. (6) **zero bare-string primary IDs on owned schemas** (the R2 falsification — grep `_id: string;` and confirm every hit is a generation counter / `ExternalVocabularyValueRef` coordinate, non-primary) + every new R3 schema branded; (7) gate+lint+fixture per new invariant; (11) **the meta-checklist** — every table has a pipeline consumer, every metric a formula, every hash a canonicalization, every stateful object transition rules, every optional field a presence condition. Name any residual violator.

**2.F No-phantom + structure.** Sample ≥20 NEW R3 contracts/helpers/lints: each cites source + `(Adj V-n)` or a `(seed — architect confirm)` marker. Confirm: balanced fences (~82); no duplicate type definitions (the `PolicyPoint` axis change must not have left a stale 5-axis copy — grep for any `PolicyPoint`-shaped block still carrying `disclosure_class`); §-numbering monotonic; OFAR count = 0; the V1 spine functions (`meetPolicyState`, `meetAllStates`, `finalizeDisclosureAndCoherence`, `disclosureVectorCeilingFor`, `evaluateActionPredicate`, `domainContribution`) all defined and shape-consistent with what the meet constructs.

## 3. Output
Write `Reviews/E1_E2_R3_Application_Fidelity_Audit_CODEX.md` (workspace-write). Verdict `READY_FOR_DELTA_REVIEW` / `FIXES_NEEDED_BEFORE_DELTA_REVIEW` (any 2.B seam defect, 2.C walked-case failure, missing ruling, or mis-landed Blocking-cluster ⇒ the latter). Findings by severity (Blocking / Substantive / Minor / DESIGN_ADVISORY) with `§+line` + the card clause violated + a paste-ready fix; then the 2.A mapping results, 2.C walked-case outcomes, 2.E checklist results, one-paragraph bottom line. Cite line evidence for PASSes on 2.B/2.C/2.D too. **Start** with repo-access confirmation + your one-line bottom line. Be strict; do not rubber-stamp.