Elnor Repo Reader

ELNOR REPO READER TEXT MIRROR
Original path: Memory Rebuild Docs/Stage_6_Charters/E1_E2_DOC81_Scope_Policy/E1_E2_R3_Application_Commission_Claude_Code.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 Commission (Claude Code)

**Repository:** `wbrody/Elnor-Specs`, branch `main` (local working tree) · **Date issued:** 2026-06-07 · **Architect:** Will Brody
**Task:** apply the R2 design-review adjudication to `Memory Rebuild Docs/Stage_6_Charters/E1_E2_DOC81_Scope_Policy/DOC81_Scope_Policy_Charter_Draft.md` (currently R2 + 2 fidelity fixes, 2,310 lines), producing **R3 in place** (same filename; git is the version record). Then write the application report. **No git commands — the architect commits.**

## 1. Authority hierarchy (higher wins on any conflict)

1. **`E1_E2_R2_Adjudication_Card.md`** (this folder) — THE instruction layer: §0 headline (unified-state pipeline adopted NOW), §1 clusters V1–V24 + minors + confirmed/retracted, §2 seed adjudications, §3 declines D-R1–D-R6, §4 the edit-cluster plan. **The three architect confirms in card §4 are APPROVED — land them.**
2. The **R1 adjudication card** (`E1_E2_Adjudication_Card.md`) — its §1-bis rulings R-1…R-5, forks, and declines REMAIN binding; R3 exists partly to make R-1/R-2 actually executable (V4/V5).
3. The two review bodies (paste-ready code): `Reviews/E1_E2_R2_Design_Review_Claude_Opus_4.8.md` (CL — all three rounds incl. addenda) and `Reviews/RT-0001_CHATGPT_CHATGPT_CONSOLIDATED_REVIEW.md` (GPT — all three passes incl. the QF pack).
4. The current R3-target draft text.

**The standing rule, now with an R2 lesson attached: never paste a review patch wholesale — and treat MERGE SEAMS as the highest-risk lines you will write.** Both defects CODEX found in R2 were born where two sources were combined without re-deriving the combined semantics (the obligation-fix ordering; the self-reported brand sweep). Wherever the card merges CL + GPT bodies (V1, V3, V11, V15, V16, V17), re-derive the merged behavior end-to-end yourself and state in the report HOW the seam was reconciled.

## 2. Read order

1. The R2 adjudication card — in full, twice.
2. The target draft — in full (you must know what V1 replaces before replacing it).
3. CL review (the walked scenarios in Part A are your test cases; §2.1 LVL_* maps; the A-B1/A-B3/A-B4 paste-ready fixes; the fix-composition map in Round 3).
4. GPT review (the `PolicyEffectiveState` §15 shape; Q1–Q5; the QF1–QF17 quantitative pack; the №-numbered second-pass fixes).
5. Ground truth as cited: ratified E0 (§2.2 `DomainProfileRestrictivenessVector` for V2; §7.1; §22), R1 card, Owner Map, Round D R0.2.

## 3. The work — four edit clusters, in this order (card §4)

**Cluster 1 — Pipeline (V1, V2, V5, V6, V8, V23, V4, V22 + the V12 flags-carrier piece).** Rebuild §3.0/§3.2 around `PolicyEffectiveState { point, disclosure_vector }`:
- `PolicyPoint` becomes the **4 capability axes** (CL B.3): `disclosure_class` leaves `PolicyPoint`/`BOTTOM`/`TOP`/`meetPoints`/`leqPoint`; disclosure travels ONLY in the vector half; the scalar is **derived exactly once** at `finalizeDisclosureAndCoherence` (then `makeCoherent` reads it). `MemoryPolicyDecision` KEEPS its declared 5-axis shape (Round D-sourced) — its scalar is consistency-guarded against its vector (V10), never independently consumed by the meet.
- Pipeline steps (normative order): 0 generation/freshness → 0b egress structural (destination present+typed) → **0c crosswalk consumption** (V6: row resolved by `relation_to_destination`; dispositions enforced; `combined_floor = meetFloors(scope, crosswalk)`; `egress_attestation_required` carried) → 1 applicable filter (+ malformed-axis F1 + scalar/vector-mismatch guard V10) → **1b egress applicable check** (V8: ≥1 *applicable* destination-matching decision) → **1c empty-applicable internal baseline** (V4a: same-principal internal action + no R-1 basis ⇒ domain-profile-only baseline; hard block retained for egress/write/learn/delegate/carryover/cross-principal) → 2 meet incl. always-present domain contribution **as a state** (point + vector) → 3 floor application **action-aware** (V4b: rule-3 floors gate EGRESS actions; internal actions floor only on the R-1 four, decided from `conservatism_floor_basis { r1_qualifying, flags }` — add that field to `ScopeResolutionResult` NOW, sourced from V12's preserved flags) → 4 sticky-restrictive **(mode='ordinary' only)** as a state-meet → 5 obligations (resolve + conflicts + vector ceilings) → 6 finalize (derive scalar once + cohere LAST) → **7 action closure + predicate gate** (V23, `ActionPermissionResult` carried).
- **V5:** `mode='restamp_reeval'` — no sticky; clamp to `MIN(root ceiling, current scope floor)` per axis + ceiling vector; §3.4 names the mode; QF10's resources-only guard on batch approvals.
- **V2:** `domainContribution(request)` becomes buildable — CL's LVL_* level maps + per-action axis selection (E0 9-axis → DOC81 4+vector; `source_authority` stays DOC82), emitting point + vector (`disclosureVectorCeilingFor(LVL_DISCLOSE[…])`). Constants marked `(seed — architect confirm, §13.6)`.
- **V22:** the normative action/exposure/relation paragraph — internal injection/attach = internal retrieve path, never `carryover`; egress actions outbound-only; internal relations never on egress decisions; + its 2 lints/2 fixtures.

**Cluster 2 — Disclosure (V3, V9, V11, V21).** The merged `deriveDisclosureClass` (card V3's exact decision tree — never falls through to `full`); GPT's parameter-map typed obligation union covering EVERY `PolicyObligationKind` (+ `require_safe_label` vector ceiling, CL A-M4); `meetDisclosureVectors` preserving rendering refs (conflict ⇒ drop-to-none); per-action `disclosure_class`+`vector` on `PolicyUIExport.scope_items`; `PolicyStampInvalidation.notice_disposition` (summary ref forbidden on `no_notice`); per-axis `binding_constraint` attribution on `EffectiveMemoryPolicy` (optional field); the three wall one-liners (V21 — invariants + lints only, explicitly non-gating, dormant-for-solo noted).

**Cluster 3 — Quant + lifecycle (V7, V10, V13–V20).** Sticky-prior in cache key + result (V7); lattice-error wrap + domain/floor/prior `isMember` guards + the scalar/vector mismatch runtime guard (V10); `persistence_kind` markers on the six derived objects — choose per object, derived vs audit-durable, and say which in the report (V13; NO new E0-named base — D-R2); capability-obligation **emission lint** + consumer-precedence line (V14); cluster `lifecycle_state` + `equivalence_generation_id` + `maximum_bottleneck_spanning_tree` + contradiction disposition (V15); cascade freeze-completion gate + `cascadeComplete()` formula + nested published-views + the lawful-edge predicate with the three-state boolean (V16); the QF helper pack — confidence-component selector, coverage math, threshold polarity model, contamination confidence minimum, `UnitInterval`, zero-bucket⇒coverage-proof, topic-ambiguity formula, canonical hash spec, quota math (V17); collection `evaluateCollectionDisposition` with `governance_resolution_state` (V18); restamp `issued_memory_flow_certificate_ref?` + invariant (NOT a second object — D-R1) + the disclosure-vector ceiling comparison (V19); the schema-integrity batch incl. `PolicyStampScope` collapsed into `PolicyStamp.scope_items`, `effect_id`, `DomainPolicyThresholdsRef`, legal-hold discriminated union + branded `hold_ref`, destructive vs non-destructive visibility effects split, `contextCompatible` contract, `principal_scope` sentinel, conditional vocab-version (V20).

**Cluster 4 — Bookkeeping.** §13.6 re-seeded per card §2 (THRESHOLD split — 1.0/0.0 fallback-only + shipped operational seeds ≈0.85/0.7 marked confirm; `local_file_export` floor → `normal_policy_check` with attestation retained; depth stays 16 — D-R3); the §1.3 minors (race attribution → EC per-object CAS; fixture-name align; bracket consistency; attestation compose rule = fail-safe OR; `user_instruction_policy_bound` invariant; unknown exposure-context fails closed); V24 forward notes (default.ordinary profile obligation; delegation envelope → DOC11; as-of read-contract stub + the §3.2 valid-time claim scoped honestly per CL A2-M6); §10/§11 roll-ups extended (all new lints/fixtures incl. the NI-1 **disclosure-axis extension** — `fixture.property.monotone_down_under_add_and_remove` must now exercise disclosure); §0 brand block extended for new refs; counts/header → "R3 — R2 design-review adjudication applied; awaiting fidelity audit + delta re-check"; footer.

## 4. Hard constraints

1. Modify ONLY the draft; create ONLY `E1_E2_R3_Application_Report.md` (this folder). No git. No other files.
2. No-phantom: every new contract/field/lint/fixture cites source + `(Adj V-n)` provenance; numeric seeds marked `(seed — architect confirm, §13.6)`.
3. No E0 re-declaration (the V2 maps REFERENCE E0 §2.2's vector; V13 uses a local marker, not an E0 base); no retired names; §9 stays downward-only; one-owner held (`ContaminationRiskMeasurement` stays DOC84-declared).
4. Declines are binding non-changes: D-R1 (no RestampIssuance object), D-R2 (no E0DerivedProjection base), D-R3 (depth 16), D-R4 (no runtime step-5b obligation reconciliation), D-R5 (wall items non-gating), plus ALL R1-card declines (D1–D12) still hold.
5. Keep the charter's voice: paste-ready TS, branded IDs, per-contract Lifecycle/Unhappy-paths/Lints/Fixtures/Cross-charter blocks. Keep the file structurally valid after every cluster (balanced fences, monotonic §s, no duplicate type declarations — `PolicyPoint`'s axis change ripples; sweep for stale 5-axis references).

## 5. Regression + acceptance (run after Cluster 4; report results)

(1) no E0 contract re-declared; (2) no retired name; (3) §9 downward-only; (4) §7.3↔§14.2 consistent; (5) one-owner incl. F7/F9; (6) zero `| string` escapes AND zero bare-string primary IDs on owned schemas (**S-1 persistence** — the prior sweep must survive your edits; new schemas branded); (7) gate+lint+fixture triple per new invariant (incl. R-1 floor-degradation lint, restamp-mode lint, crosswalk-consumption lint, sticky-cache lint, mismatch guard); (8) golden scenario + NI-1 disclosure-axis extension landed; (9) §0/§13/§14 counts + header updated; (10) §1-bis boundary audit — R-1 four-basis exhaustiveness (walk CL's scenario (d) and GPT's empty-applicable case against YOUR final text and state the outcomes), R-2 silent restamps (walk CL's scenario (e)), R-3 surfaces, R-5 principal keys; (11) **the meta-checklist (both reviews' rule, now the acceptance criterion): every table has a consumer in the pipeline; every metric has a formula; every hash has a canonicalization; every stateful object has transition rules; every optional field has a presence condition.** Sweep the final text against all five clauses and list any residual violator (or state none).

## 6. Report

`E1_E2_R3_Application_Report.md`: per cluster — what changed, where (§ + lines), seam-reconciliation notes (per §1's rule); the walked-case outcomes from check 10; then the full mapping table — **V1–V24, every §1.3 minor, every §2 seed row, D-R1–D-R6 (as verified non-changes)** — each to its landing lines or an explicit blocked-report (never silently skip). An independent automated CODEX audit checks this report against the text next; self-report accuracy is itself audited (the R2 report's check-6 claim was falsified — don't repeat that).