Review EZA-001-R001-rc.9 in this fresh session. This is an enforcement and reproduction review of a supplied public reference, not a cold implementation or hidden evaluation.

INPUT AND SCOPE

Use only the supplied EZA-001-R001-rc.9-Review folder and ordinary runtime/setup tools. It contains:

- EZA-001-R001-rc.9/: the current candidate, 125 files including MANIFEST.json.
- prior-candidate/EZA-001-R001-rc.8/: the exact 119-file predecessor, for read-only comparison. Do not execute it or transfer its results to rc.9.
- retained-rc7-probes/: original reviewer scripts, worker, plans and provenance, all unchanged public regression material.
- retained-rc8-probe/: original Q8-001 driver and plan, plus PLAN-rc9.json, its explicit port record and diff. This is also public regression material.
- CHANGE-INVENTORY.json and CHANGES.patch: author-provided comparison aids. Independently verify the complete diff.
- REVIEW-HANDOFF.json and HANDOFF-INVENTORY.json: metadata and wrapper inventory. Inventory consistency does not authenticate a publisher.

Expected SHA-256 of rc.9 MANIFEST.json:

fc44b015cbf152aef4e4cc25bc0bdc6d58f6763a0af871b81795de04e32f8f3e

Expected SHA-256 of rc.8 MANIFEST.json, for read-only comparison:

9e8d3094fd1a5ed3aa16ffa147a9ce1b039a77c4be94238cf4dc2906128dbc71

The current wire specification remains 0.1.0-draft.7, profile local-object-v1, authority format eza-local-object-continuity/3 and controller binding EZA-controller-binding/3. New synthetic fixture grants bind to the rc.9 manifest. No old grant or domain is migrated.

Use the actual supplied path. If missing, ask for its location; do not search unrelated directories or earlier conversations. Use one reviewer without delegated agents. Do not retrieve personal context, prior sessions, sibling controlled evidence, debriefs, private expected results or external EZA explanations. The supplied public predecessor, embedded historical reports and public probe plans are expressly allowed.

Create a NEW sibling EZA-001-R001-rc.9-Independent-Review-001 directory for every review output and runtime setup. If it exists, preserve it and use the next unused suffix. Keep the supplied folder unchanged. This is an explicitly authorized unsigned prepublication review. If supplied an extracted folder, no ZIP is required. If supplied a ZIP, verify its SHA-256 against the separate handoff digest before extraction; do not accept a self-contained digest as the external handoff pin. Never hunt for an absent archive. Do not regenerate a candidate manifest, sign, publish, deploy, migrate authority or modify a live domain. Publication keys remain outside the reviewing agent's control.

VERIFY BEFORE EXECUTION

With trusted standard-library code, verify both manifest pins, exact inventories, every file size/hash, regular-file types and absence of symlinks or unexpected files. Verify the wrapper inventory and public probe source/plan hashes as well, before executing candidate or probe code. A pinned hash identifies supplied bytes, not publisher identity. Preserve verification commands and outcomes.

Read the current SPECIFICATION.md, BOUNDARY.md, CLOCK-HISTORY.md, RESULTS.md, CLAIMS.md, LIMITS.md, RECEIPTS.md, CHANGELOG.md, REVIEW-RESOLUTIONS.json and reproduction instructions. Historical prose records prior candidates; it does not authorize predecessor execution or modification.

Verify the entire rc.8-to-rc.9 diff. The declared scope is documentation and release metadata: 18 modified files including MANIFEST.json, six added history/provenance files, no deletions. All existing enforcement/evaluator code, schemas, fixtures, expectations and harnesses should remain byte-identical. tests/REQUIRED.json changes only candidate identity. The only existing Python source-file change should be the literal result-kind label in reproduction/run.py. The Python inventory-check body in reproduction/VERIFY.md should be unchanged. Existing historical files, license terms, RESULTS.md, CLOCK-HISTORY.md, RECEIPTS.md and LIMITS.md should be unchanged. Preserve the complete observed diff and hashes, including any undeclared difference.

ASSESS IR8-001 AND ALL VERIFICATION ENTRY POINTS

The rc.8 review reported a low-severity, nonblocking guide inconsistency: VERIFY.md's unsigned branch assumed a ZIP digest while INDEPENDENT-REVIEW.md allowed an extracted-directory handoff. Determine whether both reproduction/VERIFY.md and reproduction/README.md now describe the same three paths:

1. Signed publication: authenticate the exact manifest with an already-trusted signature verifier and an independently authenticated owner public key, then verify its full inventory before executing package code.
2. Explicitly authorized unsigned ZIP review: verify the supplied archive against the handoff ZIP digest before extraction, then verify the handoff manifest pin and full inventory before package execution.
3. Explicitly authorized unsigned extracted-directory review: verify the handoff manifest pin and full inventory before package execution; no absent ZIP search or owner signature is required for this review.

Check the root README, reproduction/INDEPENDENT-REVIEW.md, reproduction/SIGNING.md and CLAIMS.md for agreement. Neither unsigned path authenticates a publisher or satisfies the public signature gate. Missing or mismatching required verification evidence must not be bypassed by regenerating a manifest. Signed-publication requirements must remain intact. Distinguish document inspection from actual signature verification; no owner key or signature is supplied for this review.

Determine whether the correction only clarifies existing handoff procedures or introduces a new normative enforcement rule. Retaining draft.7 is justified only without a substantive wire, authority or expected-outcome change. Identify any remaining standalone ambiguity without assuming author intent. Assess IR8-001 for rc.9; its historical rc.8 status remains frozen. IR7-001 was closed for rc.8 by its reviewer; preserve that versioned history and check for regression.

EXECUTE AND PRESERVE FIRST RESULTS

Record the actual OS, architecture, Python and dependency versions. Use Python 3.11+ on Linux or macOS. If setup is needed, create a NEW isolated runtime in the review directory, install jsonschema==4.25.1 and cryptography==46.0.0, and preserve commands, output and resolved versions. Do not alter the global environment. Record setup help and failed attempts. Do not infer an approving human actor when none is named by the available records.

Run the FULL fixed suite first, using a new output directory:

PYTHON -B CANDIDATE/reproduction/run.py --expected-manifest fc44b015cbf152aef4e4cc25bc0bdc6d58f6763a0af871b81795de04e32f8f3e --out NEW_REVIEW/fixed-first

Replace only runtime and filesystem paths. Preserve the first run and every failure. Require all 394 behavior/regression groups, 26 schema groups and recorded-output validation. Leave embedded pending/NOT_RUN fields unchanged; results belong outside the candidate and identify its exact manifest.

After preserving the fixed result, replay these public probes on rc.9 and NEW synthetic domains:

PYTHON -B RC7_PROBES/supplemental_review.py --candidate CANDIDATE --pin fc44b015cbf152aef4e4cc25bc0bdc6d58f6763a0af871b81795de04e32f8f3e --plan RC7_PROBES/SUPPLEMENTAL-PLAN.json --out NEW_REVIEW/retained-clock-first

PYTHON -B RC7_PROBES/evidence_review.py --candidate CANDIDATE --pin fc44b015cbf152aef4e4cc25bc0bdc6d58f6763a0af871b81795de04e32f8f3e --plan RC7_PROBES/EVIDENCE-PLAN.json --out NEW_REVIEW/retained-evidence-first

PYTHON -B RC8_PROBE/exact_q01.py --candidate CANDIDATE --plan RC8_PROBE/PLAN-rc9.json --out NEW_REVIEW/retained-q8-first

The first worker must remain beside supplemental_review.py. Report the 62 clock/recovery groups, eight evidence/controller groups and one Q8-001 group separately from the fixed suite and any new probes. Original script docstrings and authoring notes describe the original review; these executions are public replays, not new independent tests.

Verify Q8-001's port before execution: the driver and original plan are unchanged; PLAN-rc9.json changes only the candidate manifest, two fresh grant release bindings, classification and the rc.8-to-rc.9 wording of the fresh-grant adaptation. The original creation timestamp remains source provenance. Expectations, requests, samples, historical source hash and all other initial state remain unchanged. Inspect the exact plan diff against PORT-rc9.json.

The retained N01 label is not a claim of full historical Q01 fixture equality: its first grant expiry is 20000, while historical Q01 and Q8-001 use 5000. Both exceed the final accepted sample of 2000. Q8-001 retains the 1800 second expiry and the same-process two-transaction sequence, checking final durable history and the subsequent 1500 regression. Its driver directly invokes captured verified kernel code and maps TimeUncertain to a BLOCKED observation; it is not an additional public-host-interface test. Confirm actual state and source bindings rather than relying on the label.

Inspect actual protected effects, consumed roots, handles, stored history, witness hashes and controller records. Include old-format rejection, pending versus finalized uncertainty, unchanged clock limits, C08's positive control and isolated content challenge, release mismatch, spent-state rollback, evidence-only receipts and no redispatch. Clock and storage-failure injection are privileged synthetic instrumentation, not requester powers. Process failures do not establish physical power-loss durability.

Write any new targeted expectations before executing them. Add probes only to resolve concrete remaining uncertainty; there is no arbitrary quota. Report new tests, observations and residual uncertainty separately. A passing percentage cannot absorb an authority failure. Never weaken expectations, replace a failed first result, repair an old witness or reset spent authority to obtain a pass.

Reverify the candidate, predecessor and retained-probe bytes after testing. Preserve available commands, stdout/stderr, environment/setup evidence, plans, hashes, source bindings, state observations, first fixed result, public replays and any new probes. Disclose unavailable raw session events and procedure deviations. Do not invent an event export or claim independent implementation, authenticated execution provenance, native hostile-agent containment, independent UTC truth or general production readiness.

DELIVER

Return REVIEW-REPORT.md, FINDINGS.json, REVIEW-SUMMARY.json, REPRODUCE.md, a submission inventory of file sizes and SHA-256 hashes, and a ZIP containing review evidence with its SHA-256. Include exact input bytes or a complete independently verified comparison sufficient to bind every finding to the supplied pins. Preserve predecessor and probe provenance. Keep fixed results, public replays, new reviewer-authored probes, observations, setup failures and documentation/enforcement findings separate.

State whether IR8-001 is closed for rc.9, whether the verification paths agree across all entry points, whether retaining draft.7 is justified, whether the standalone current contract is reproducible without author interpretation, and whether the required gates passed for this exact digest. If no new probe was necessary, explain the scope without inventing a score.

Stop after delivering the review. Owner-controlled signing and separate website acceptance remain later gates.
