<!-- DOC_STATUS: ACTIVE_NON_AUTHORITY -->

FIX7 P0 — Rollback-Validator Hardening Report (2026-06-11)

• Macro: FIX7_P0_PRE_EXECUTION_HARDENING_AND_GOVERNANCE_CONSOLIDATION_MACRO_2026_06_11
• Authority: PRE_EXECUTION_HARDENING_NON_AUTHORITY. Authorizes nothing. Staging/paper-only.
• Final status: FIX7_P0_PRE_EXECUTION_HARDENING_READY_GOVERNANCE_FOLD_PENDING
• Production mutation: NO · Implementation execution: NO · REAL_RUN/QT001/cutover: NO
• Patch packet: knowledge/dev/reports/architecture/fix7-p0-rollback-validator-hardening-packet-2026-06-11/ (packet tree 59788d04c8d7afb01b28e7c05f23aa9c2b708f03d5dae02096c4c024287e20e4)
• Addresses: T2-REC-ROLLBACK-HARDENING-1 (non-blocking recommendation from the T2 independent review, which itself returned T2_FIX7_P0_DRYRUN_EXECUTION_READINESS_REVIEW_PASS).

1. What was hardened

The original dry-run validator check_rollback_proof (dryrun_validator.py, sha256 7fb2f11e8baec9703faefbb86c23b161366b75340d98860c5c87977ee7bcb297, inside fix7-p0-dryrun-and-execution-readiness-packet-2026-06-11, tree 02b200e5…94e6) verified that each rollback entry had before_hash/after_apply_hash/after_rollback_hash present and that before_hash == after_rollback_hash (restored). It did not verify that after_apply_hash != before_hash — so a fabricated proof in which the "apply" step changed nothing (apply hash equal to before hash) could pass as "rollback proven".

hardened_dryrun_validator.py is a strict superset of the original — same byte-faithful logic for all 11 gates, with rollback hardening added in check_rollback_proof.

A. Hardening target table

file	current rule	missing guard	proposed patch	safe to patch?
dryrun_validator.py check_rollback_proof	before/after_apply/after_rollback present; before == after_rollback; restored_match True; production_rollback_status ∈ {None, NOT_APPLICABLE}	no check that after_apply_hash != before_hash → fake proof with no real mutation passes	add ROLLBACK_APPLY_DID_NOT_MUTATE when both hashes present and equal; add optional expected_restored_hash pin check (ROLLBACK_NOT_RESTORED_TO_PIN)	YES — additive only; removes no gate; preserves ABSENT sentinel; published as a separate hardened file, not an in-place rewrite

The full required-guard set is now enforced when rollback_proof_status == PROVEN_IN_STAGING: before_hash present · after_apply_hash present · after_rollback_hash present · after_apply_hash != before_hash · after_rollback_hash == before_hash (or == expected_restored_hash) · production_rollback_status stays NOT_APPLICABLE (production mutation forbidden).

2. Probes (Workstream B)

hardened_bad_input_probes.py runs against the real frozen T1 rollback evidence (valid_evidence_recheck.json, entries copied byte-for-byte from rollback-recovery-proof.json sha256 07acdf19…37ba). Result: 7/7 fail-closed, control passes, any_fail_open=false (hardening_probe_results.json):

#	probe	result	code(s)
1	after_apply_hash == before_hash while claiming mutation	FAIL-CLOSED	ROLLBACK_APPLY_DID_NOT_MUTATE:RB-3
2	missing after_apply_hash	FAIL-CLOSED	ROLLBACK_ENTRY_MISSING_HASH:RB-3:after_apply_hash
3	missing before_hash	FAIL-CLOSED	ROLLBACK_ENTRY_MISSING_HASH:RB-3:before_hash (+ NOT_RESTORED)
4	missing after_rollback_hash	FAIL-CLOSED	ROLLBACK_ENTRY_MISSING_HASH:RB-3:after_rollback_hash (+ NOT_RESTORED/PIN)
5	production rollback claimed despite no production mutation	FAIL-CLOSED	PRODUCTION_ROLLBACK_CLAIMED
6	rollback proof claimed but no rollback step	FAIL-CLOSED	ROLLBACK_PROVEN_BUT_NO_MUTATION; ROLLBACK_PROVEN_NO_ENTRIES
7	valid existing T1 rollback evidence still passes	PASS (clean)	—

No invalid input emitted any PASS/digest/cert/seal-like token (leak check enforced in expect_fail).

3. Valid-evidence recheck + no-regression

B. Rollback proof evidence table (real frozen T1 entries, hardened verdict)

step	before hash	after_apply_hash	after_rollback_hash	distinct apply?	restored?
RB-2	ABSENT	30bdca6e…	ABSENT	YES (real hash ≠ ABSENT)	YES (absent again)
RB-3	49c386a9…	91c520d9…	49c386a9… (== pin)	YES	YES (== before/pin)
RB-4	ABSENT	1cddd95e…	ABSENT	YES (real hash ≠ ABSENT)	YES (absent again)

hardened_dryrun_validator.py --selftest → PASS (real evidence passes; fabricated no-mutation rollback fails closed). Exit 0.

No-regression cross-check (reconstructed the real dry-run packet, tree 02b200e5…94e6):

• original dryrun_validator.py → PASS (exit 0) — reproduces T1's verdict;
• hardened_dryrun_validator.py full-packet mode → PASS (exit 0) — hardened validator still accepts the real packet;
• original 20 bad_input_probes.py → 20/20 fail-closed (exit 0).

So the hardening neither breaks valid evidence nor weakens any existing gate.

4. Packet integrity

bash RERUN.sh → RERUN_RESULT: PASS; shasum -c HASH_MANIFEST.txt all OK; packet_tree.sha256 == sha256(HASH_MANIFEST.txt) = 59788d04…e20e4. KB round-trip: all 11 files re-fetched untruncated; 10 ASCII files' stored length equals their local byte length exactly (byte-identical); README multibyte-consistent.

5. Boundaries honored

The original dry-run packet, its verdict history, and the T2/Codex review artifacts were not modified or deleted. This is a patch packet marked REQUIRED_FOR_FUTURE_EXECUTION_MACRO. No N7/N8/P7 alteration. No production. No implementation execution. No REAL_RUN/QT001/permit/activation/repoint/cutover.

6. Disposition

The hardened validator is required for any future AUTHORIZE_IMPLEMENTATION_EXECUTION_NO_PRODUCTION macro: it must replace dryrun_validator.py as the rollback gate. Until then, nothing executes. Default owner decision: HOLD.