docs(evaluator): audit E9 cross-target readiness

[briansrls]

Summary

• adds a docs-only PR-E E9 readiness audit for cross-target harness consumption
• records that PR-D slice 0 + DifferentialEquals slice 1 are planning surfaces, not executable L5/ForAllTargets evidence
• names the implementation gates: LanguageSpec, all Shape A targets, L4/L7 corpus seed, typed structural observation, runner capability dissolution, and approved fixture home
• cross-links the audit from the R3 evaluator dispatch brief and cadence matrix

Verification

• git diff --check
• pre-push hook: cargo fmt --all --check

Scope

Docs-only. No runner implementation, no new TestPredicate variants, no substrate changes, no L5 corpus execution.

[briansrls]

Manager pass: scope is correct for the E9 dispatch. This is docs-only, names the exact implementation gates, and avoids runner implementation, target enumeration, new TestPredicate variants, substrate changes, and L5 corpus execution. The cadence/dispatch cross-links are appropriate.\n\nMarked ready for review so CI and scheduled reviews can run. No manager-blocking changes requested.\n\n— sent from snappy-moth-795

[briansrls]

Review metadata

• Provider / model: codex / codex-default
• Commit: dcb02316 · Trigger: schedule
• Comparison: origin/main @ 4fe07053 ... review/pr-1417-dcb02316 @ dcb02316
• Thinking: 85s wall

---

Verdict: APPROVE

This is a docs-only readiness audit that keeps E9 blocked until the existing PR-D/L5 gates are live. I found no concrete violations of the pinned invariants, modeling discipline, coding, or testing guidance in the diff. No builds or tests were run, per review instructions.

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: dcb02316 · Trigger: schedule
• Thinking: 175s wall

---

Non-blocking — Strengths

• docs/briefs/r3-pr-e9-cross-target-harness-consumption-readiness.md Docs-only audit stays within thesis-aligned verification boundaries: it blocks L5 execution on LanguageSpec, Shape A grounding, L4/L7 corpus, and typed observation gates instead of authorizing raw target output comparison.

ROADMAP — Verified

• PR-E E9 cross-target harness consumption: The new cadence and dispatch rows consistently preserve the readiness-only status and do not introduce new predicates, substrate carriers, runner work, or target enumeration.

✅ No blocking concerns; this is a clean readiness audit that keeps the cross-target lane gated on the right structural prerequisites.

[briansrls]

Review metadata

• Provider / model: codex / codex-default
• Commit: 573d4143 · Trigger: schedule
• Comparison: origin/main @ 0fd90ea2 ... review/pr-1417-573d4143 @ 573d4143
• Thinking: 51s wall

---

Verdict: APPROVE

Docs-only readiness audit is narrow and consistent with the rubric: it keeps E9 blocked until typed observation, LanguageSpec, Shape A grounding, and corpus prerequisites are live, and explicitly forbids runner changes, target enumeration, raw stdout equality, new predicates, and substrate carriers. No concrete invariant, coding, or testing violations observed in the diff. Tests not run per review instructions.

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: 573d4143 · Trigger: schedule
• Thinking: 151s wall

---

✅ Docs-only readiness audit has no blocking concerns; it keeps E9 implementation gated on structural prerequisites rather than authorizing runner, predicate, target-enumeration, or raw-output work.

[briansrls]

Review metadata

• Provider / model: openai-pro / gpt-5-5-pro
• Commit: 573d4143 · Trigger: manual
• Conversation: View conversation

---

1. Story of the diff

This PR adds a docs-only readiness audit for PR-E E9’s proposed cross-target harness consumption path. The new audit document explicitly says E9 is not implementation-ready yet and turns that into a gated plan: PR-D’s slice 0/1 surfaces exist, but strict ForAllTargets execution must wait for LanguageSpec readiness, all Shape A targets, L4/L7 corpus seed availability, typed structural observation, a runner-capability dissolution path, and an approved fixture home (docs/briefs/r3-pr-e9-cross-target-harness-consumption-readiness.md:39-54). The cadence matrix gets a matching PR-E E9 row that keeps this as a readiness-audit lane, with no runner implementation, target enumeration, raw byte equality, new predicate, or substrate carrier authorized (docs/briefs/r2-evaluator-cadence-verification-matrix.md:29). The R3 evaluator dispatch brief then links the readiness receipt back into the parent E9 lane so future work sees the blocker list before attempting implementation (docs/briefs/r3-evaluator-dispatch.md:173-177).

2. Invariant categories

1. LAYER MODEL (substrate vs implementation).

Compliant — this is implementation- and docs-only, not substrate mutation: the audit states “This note does not authorize runner implementation, new TestPredicate variants, substrate changes, target enumeration, or L5 corpus execution” (docs/briefs/r3-pr-e9-cross-target-harness-consumption-readiness.md:3-5). The cadence row repeats the same boundary: “No runner implementation; no target enumeration; no raw stdout/byte equality; no new predicate or substrate carrier” (docs/briefs/r2-evaluator-cadence-verification-matrix.md:29).

1. INVARIANTS.md + modeling-discipline.md.

Compliant — Boundary Discipline / single authority is handled by requiring L5 to consume the existing corpus and approved fixture home rather than inventing a parallel one: “L5 consumes that corpus; it does not invent a parallel one here” (docs/briefs/r3-pr-e9-cross-target-harness-consumption-readiness.md:48) and “uses the existing PR-D/L5 fixture path approved for that claim instead of creating a parallel cross-target suite” (docs/briefs/r3-pr-e9-cross-target-harness-consumption-readiness.md:51). Fail-closed is also explicit: “If any gate is missing, the correct E9 output is a blocked row or docs update, not runner code” (docs/briefs/r3-pr-e9-cross-target-harness-consumption-readiness.md:53-54).

1. CODING.md.

N/A — the diff adds documentation only; it does not add Rust functions, methods, helpers, error/result shapes, global state, or module structure subject to the coding-style rules.

1. TESTING.md.

Compliant — no tests are added, and that is appropriate for a readiness audit that explicitly forbids execution work: “No L5 corpus execution in this audit” (docs/briefs/r3-pr-e9-cross-target-harness-consumption-readiness.md:78) and “No changes to src/v3/compiler/src/test_runner.rs” (docs/briefs/r3-pr-e9-cross-target-harness-consumption-readiness.md:83). The doc also avoids promoting a runnable behavior claim before the required evidence exists.

1. LOCKED DESIGN DECISIONS.

Compliant — the PR references the cross-target design lock as the authority rather than redefining it: “Design lock: docs/design-cross-target-equivalence.md defines semantic observations, corpus policy, oracle validity, float policy, and side-effect normalization for strict L5 evidence” (docs/briefs/r3-pr-e9-cross-target-harness-consumption-readiness.md:10-12). It then requires future execution to compare “typed semantic observations per the design lock” (docs/briefs/r3-pr-e9-cross-target-harness-consumption-readiness.md:68-69).

1. TRACKED vs UNTRACKED DEBT.

Compliant — the existing ForAllTargets raw-command scaffold is identified as non-final, bounded, and tied to a dissolution path. The audit says the raw command/args/expected-exit surface “is not the final typed observation model” (docs/briefs/r3-pr-e9-cross-target-harness-consumption-readiness.md:34-37), requires a “Runner capability dissolution path” that names the typed target capability / observation facts that will retire it (docs/briefs/r3-pr-e9-cross-target-harness-consumption-readiness.md:50), and bars raw stdout/string/byte equality as authority (docs/briefs/r3-pr-e9-cross-target-harness-consumption-readiness.md:49, docs/briefs/r3-pr-e9-cross-target-harness-consumption-readiness.md:81).

3. Verdict

APPROVE. The PR is a clean docs-only blocker/readiness audit: it does not add implementation surface, it prevents premature E9 execution, and it gives the existing scaffold explicit bounds and a dissolution trigger. I did not find a diff-backed invariant violation.