E1_E2_CODEX_Fidelity_Audit_Prompt.md
Memory Rebuild Docs/Stage_6_Charters/E1_E2_DOC81_Scope_Policy/E1_E2_CODEX_Fidelity_Audit_Prompt.md
ELNOR REPO READER TEXT MIRROR
Original path: Memory Rebuild Docs/Stage_6_Charters/E1_E2_DOC81_Scope_Policy/E1_E2_CODEX_Fidelity_Audit_Prompt.md
Source repo: /Users/OpenClaw1/Elnor/Elnor Specs
Git branch: main
Git commit: dbaa25962edc11ab30e8d4ca1715f9ae5bf77331
Generated: 2026-06-09T01:23:58.539Z
---
# DOC81 (E1/E2) Charter Draft — Fidelity & Traceability Audit Commission (CODEX)
**Repository:** github.com/wbrody/Elnor-Specs — branch `main`
**Date issued:** 2026-06-01
**Architect:** Will Brody. **You produce the audit report; Will reviews it before the design red-team.**
**You are:** an independent auditor verifying that the freshly-drafted **DOC81 Scope & Policy** charter (`DOC81_Scope_Policy_Charter_Draft.md`, 1,386 lines) **faithfully realizes its commission, binds correctly to the ratified E0 contracts, invents nothing ungrounded, and actually meets its acceptance criteria.** This is a **fidelity + traceability audit — a pre-red-team screen** — NOT a design review and NOT a re-adjudication. The design red-team (architecture soundness, better patterns) is a separate later pass by other models; if you see a design concern, log it under `DESIGN_ADVISORY`, don't act on it.
**Why this audit exists:** the draft was written by Claude Code, which self-reported "all acceptance criteria met / 26 schemas paste-ready / 0 E0 re-declarations / retired names guarded." A drafter's self-report of its own pass is exactly what an independent check must verify — confirm or refute each claim against the actual text and the cited sources. (A spot-check already found the self-report's "nothing outside the folder modified" claim was slightly overstated — `REPO_FILE_MANIFEST.md` also changed; benign, but it shows self-reports need checking.)
---
## 1. Inputs (all paths repo-relative; read directly — if you can't access the repo, stop and tell Will)
**The artifact under audit**
1. `Memory Rebuild Docs/Stage_6_Charters/E1_E2_DOC81_Scope_Policy/DOC81_Scope_Policy_Charter_Draft.md` — THE TARGET (§0–§14; 1,386 lines).
**The spec it must satisfy**
2. `…/E1_E2_DOC81_Scope_Policy/E1_E2_Drafting_Commission_Claude_Code.md` — the commission (read order, §3 deliverable + required schemas, §4 hard constraints, §5 acceptance criteria). **The draft is graded against this.**
3. `…/E1_E2_DOC81_Scope_Policy/Charter_Opening_Brief.md` — the 20 draft targets + exit criteria.
4. `…/E1_E2_DOC81_Scope_Policy/Charter_Input_Deck.md` — the 10 OPA rows, pinned ADQs, fold-ins, owned objects.
**The contracts DOC81 must BIND to (and must NOT redefine)**
5. `Memory Rebuild Docs/Stage_6_Charters/E0_DOC80_Core/DOC80_Core_Charter_Draft.md` — the ratified DOC80 core. Verify DOC81 references E0's `ReasonCodeRegistry` (§2.1), `DomainProfileRegistry` (§2.2), `EffectiveMemoryPolicyRef` pointer (§3.2), `MemoryFlowCertificate` (§3.3), `MemoryMutationEnvelope` (§5.1), policy-generation carrier (§8.3), §12.1 invariants, and §22 egress vocab **by E0 section number — never re-declaring them.**
**Ground truth (for the no-phantom / trace checks)**
6. `Memory Rebuild Docs/DOC80 Target Baseline/Owner Map/DOC80_Owner_Map.md` — DOC81-owned rows (62, 71–94, 194). One owner per schema.
7. `Memory Rebuild Docs/DOC80 Target Baseline/Skeletal Spec/DOC80_Skeletal_Target_Baseline.md` — §"DOC81" (§1–§7 map) + §10.3/10.4/10.11/11.3/11.9.
8. `Memory Rebuild Docs/DOC80 Target Baseline/Import Graph/DOC80_Import_Graph.md` — DOC81 edges (acyclic; `DOC81 → DOC86` only — the B1 fix; never the reverse).
9. `Memory Rebuild Docs/DOC80 Target Baseline/Retired Names/DOC80_Retired_Names.md` — `ScopeMembrane`, scalar `MemoryPolicyDecision`/`PolicyMembraneDecision`, `ask_user`, linear `contamination_risk` must appear ONLY in guard/lint/lineage context.
10. `Memory Rebuild Docs/Flattening/Execution Ledger/Architect Decision Queue/Architect_Decision_Queue.md` — ADQ-310/315/316/313/308/213/406/PASS2-02 (the draft's authority).
11. `Memory Rebuild Docs/Flattening/Supersession Matrix/Supersession_Matrix.md` — SM-010/011/012/013/014/101/102/104/105/106/107/208.
12. `OP-A and Operations and Trackers/OPA_V4.md` — §6 (10 PropA rows), §6.Z (`OBL-D81-TOPIC-COLLECTION-SUPPRESSION-01`), §6.Z3 (`OBL-PROPA-LOCALFILEEXPORT-OUTBOUND-PATCH-01`), §6.A (source-exclusion body).
13. **`OP-A and Operations and Trackers/Archived DOC OP-A and Operations DOCS/OPA_V3_18.md`** — **the archived OPA the 8 carried PropA rows (`OBL-PROPA-NEW-01/02`, `-V15-01..06`) inherit their acceptance bodies from.** This was NOT in the drafter's read-list (the draft's §13.3 flags it). **You have it — use it to confirm or refute the §7.3 landing-site mapping against the actual V3.18 row bodies.**
14. `Current Specs/Miscellaneous Specs/MultiDoc_PropA_R6_3_Compiled_Operative_Spec.md` (PropA R6.3) and `Memory Rebuild Docs/Memory Rebuild Review Packs/DOC80_Memory_Control_Plane_PreSpec_Review_Pack_v1_0_2026-05-25/13_Round_D_Policy_Scope_UI_Micro_Patch_R0_2.md` (Round D — the scope §3.x / policy §1.x / capability-disclosure §1.6 semantics + the ScopeAffinity §3.4 + restamp §5.4 sources).
---
## 2. What to check
### 2.A Acceptance-criteria verification (Commission §5)
Re-derive each acceptance criterion against the actual text and report PASS/FAIL + evidence (§+line). Do **not** copy the drafter's self-report — confirm it. In particular: all §1–§14 present + substantive; all 20 draft targets addressed; all pinned ADQs land in §14.1; all 10 PropA rows + 2 E0-handed land in §7/§14.2; the 5-plane cascade wiring check done (§5.2, six §10.11 lints); the four named invariants each have a runtime gate + Stage-9 lint + Stage-8 negative fixture (§10); the `EffectiveMemoryPolicy` **meet algorithm** is specified (monotone restriction, conservative-on-missing, restamp-only widening), not just the shape; `OPEN_FOR_ARCHITECT_REVIEW` ≤ 5.
### 2.B Required-schema presence + well-formedness (Commission §3.3)
Confirm each required full schema is present, paste-ready, and structurally sound (balanced TS, exhaustive enums, optionality marked, branded IDs per E0 §8.5, no undefined-within-draft type refs): the scope family (`ScopeIdentityRoot`…`ScopePopulationHealth`), the 5-axis `MemoryPolicyDecision`, `EffectiveMemoryPolicy` + meet, `PolicyMembraneDecision`, `PolicyObligation`, `PolicyStamp`/`Scope`/`Invalidation`/`Restamp`, `PolicyDisambiguationRequest`, `EpisodePolicyEpoch`, `disclosure_class`, `SafeLabelDisclosurePolicy`, `ExtractionRoutePolicyEnvelope`, `PolicyCappedDAMSInput`, `contamination_risk` threshold, `TopicRiskClass`, `CascadingSourceInvalidation` envelope, `LegalHoldState`, `collection_mode` governance. Flag any missing, stub, or malformed schema.
### 2.C No-phantom / traceability (the core)
For every contract, field, enum value, lint, and fixture: confirm it traces to a citation (Owner Map / Skeletal / ADQ / OPA / Round D / SM / E0 section) **or** is honestly flagged `OPEN_FOR_ARCHITECT_REVIEW`. Flag any ungrounded invention. Apply special scrutiny to value sets that look reasonable but uncited (enum members, thresholds, lint names).
### 2.D No E0 redefinition + boundary discipline
Independently verify **0 re-declarations** of an E0-owned registry/contract (DOC81 must reference, not declare). Verify DOC81 owns only what the Owner Map says — `TopicCollectionDirective` stays DOC83; membership/Topic identity stays DOC87; truth stays DOC82; the cross-cutting registries stay DOC80; EC is the executor. Flag any boundary creep.
### 2.E The four hard invariants
Verify each is correctly specified with the gate+lint+fixture triple AND is semantically right: (a) capability-meet vs disclosure-meet **orthogonal** (§4.1); (b) **scope ≠ membership ≠ authority** (DOC81/DOC87 non-overlap, §6.2, the five §10.4 lints); (c) **policy monotonicity** — narrow-only, `PolicyStampRestamp` the only widening path, ceiling-bound per ADQ-316, source-revocation monotonically lowers (§6.3); (d) **fail-closed** default — conservative on missing/unknown/incomparable (§4.6).
### 2.F The carried-PropA bodies (§13.3 — the real verification gap; close it with file 13)
The draft maps the 8 carried PropA rows at the topic level and did NOT guess their acceptance bodies (correct, per Commission §7). **Using `OPA_V3_18.md` (file 13), confirm the §7.3 landing-site mapping against the actual V3.18 row bodies.** Report: do the landing sites match the real bodies? Are any acceptance criteria / fixtures from V3.18 unaccounted for? This is the single most important coverage check.
### 2.G The reconciliations + flags (audit the drafter's judgment calls)
- **ScopeAffinity** (§13.1): the draft resolved the conflict between Owner-Map/brief `{direct, secondary, shared, analogical, background}` and Round D §3.4 `{direct, secondary, background, analogical, unrelated, uncertain}` by **union**. Is the union sound, complete, and non-contradictory? Are all members used consistently downstream?
- **disclosure_class total-order** (§13.2): the draft imposed a total order on the lattice for the meet. Is the order defensible (does it make the meet well-defined without losing a real partial-order case)?
- **The 3 OFAR flags** (§13.3 PropA bodies, §13.4 `PolicyMembraneDecision.crossing_disposition` enum, §13.5 dual destination vocabularies): is each genuinely architect-decision-level, or could/should the drafter have resolved it from sources? For §13.4 and §13.5, are the synthesized value sets traceable as claimed?
### 2.H Retired names + import graph + internal soundness
Retired names appear only in guard/lint/lineage context (file 9); import graph respected (`DOC81 → DOC86` only; no B1 cycle, §9); no dangling `§`-refs, no duplicate/orphaned/out-of-order sections, numbering intact §0–§14, naming/casing consistent.
---
## 3. Scope boundary
- **Not a design review.** Architecture soundness, composition, and better-pattern proposals are the next (multi-model) pass. If you spot a genuine design concern, log it under a `DESIGN_ADVISORY` heading — do not adjudicate or redesign.
- **Not a re-adjudication.** ADQ resolutions and the E0 contracts are settled. If a reviewer-style objection arises, mark `OUT_OF_SCOPE_ADVISORY`.
- **No invention.** Every defect you assert cites the commission item or ground-truth source it violates, plus the draft §+line; for fixes, give paste-ready text.
## 4. Output
A single Markdown report. Tag findings `MISSING / MALFORMED / PHANTOM / UNTRACEABLE / E0_REDEFINITION / BOUNDARY_CREEP / INVARIANT_GAP / AC_FAIL / RETIRED_NAME_LEAK / SOUNDNESS_BUG / CONFIRMED / DESIGN_ADVISORY / OUT_OF_SCOPE_ADVISORY`. Sections:
1. **Summary + verdict** — one of `FIDELITY_CONFIRMED` (clean → proceed to design red-team) · `FIXES_NEEDED_BEFORE_RED_TEAM` (defects listed, each with a paste-ready fix) · `MAJOR_GAPS_RETURN_TO_DRAFTER`. State the AC pass-count and where the self-report and the text diverged.
2. **Acceptance-criteria table** (§2.A) — PASS/FAIL + evidence per criterion.
3. **Schema presence/quality** (§2.B) · **Traceability / phantom findings** (§2.C) · **E0-binding + boundary** (§2.D) · **Invariants** (§2.E).
4. **Carried-PropA body confirmation** (§2.F) — the V3.18 mapping result, explicitly.
5. **Reconciliations + OFAR assessment** (§2.G).
6. **Retired-names / import-graph / soundness** (§2.H).
7. **`DESIGN_ADVISORY` annex** (anything for the later design pass).
For every defect (anything not `CONFIRMED`/`PASS`), give a **paste-ready fix** anchored to the draft (`REPLACE:` / `INSERT AFTER:` + corrected text). Cite §+line for every finding. If the draft is faithful, say so plainly and verdict `FIDELITY_CONFIRMED`.
## 5. Delivery
Produce the report as **one Markdown document in your reply.** If your environment can write to the repo, also save it to `Memory Rebuild Docs/Stage_6_Charters/E1_E2_DOC81_Scope_Policy/Reviews/E1_E2_Fidelity_Audit_CODEX.md`; otherwise return it inline and Will saves it. **Start by** confirming repo access, listing the files you read (including `OPA_V3_18.md`), and stating how many acceptance criteria you will verify. If you can't reach the repo, stop and say so.
**You are the independent fidelity gate between a self-reported draft and the design red-team. Verify everything; assume nothing — but stay in your lane (fidelity, not design).**