[jonahgabriel]

Cites OMN-19711 (plan task B2 of knowledge-base-internal beta/plans/2026-09-25-golden-chain-event-tests-for-pr-validation-plan.md, section 3.3, parent OMN-19680). Part B was approved by the operator ruling of 2026-09-26 (ledger RULING, lane validation-plans-83-ruling, ticket OMN-19680).

What

scripts/ci/chain_replacement_parity.py turns the coverage-parity rule for deleting test cases T in favour of chain cases C, which cover handler modules H, into a receipt. Each of P1 to P5 is PASS, FAIL or MISSING:

check	rule
P1 mutation	every mutant of H killed before the deletion is killed after it, compared by mutant id with zero escapes. Zero mutants counts as MISSING
P2 coverage	per file of H, covered lines and branch arcs after the deletion are supersets of those before it (set comparison, zero tolerance)
P3 walker	every walker ERROR path T asserted is asserted by C
P4 determinism	20 runs, 20 distinct seeds, all pass, identical event lists per case
P5 cost	C's summed case time is at most T's

Any FAIL or MISSING, and any unreadable or invalid input, exits 1.

• evaluate is pure and unit-tested. measure is lab-only: it runs coverage and mutmut 3 over H in scratch copies and never edits the worktree's config. It runs C under 20 seeds, each setting PYTHONHASHSEED and a seeded case order; pytest-randomly is not a dependency here, so this stands in for it. It also records junit times. The default comparison suite is T plus C, which is stricter than a full-suite comparison. --baseline adds paths to both sides.
• mutmut 3 goes in a new parity dependency group, so CI's dev sync is unchanged. The lock adds only mutmut, libcst, pyyaml-ft and setproctitle.

Acceptance evidence

• AC1 and AC2: uv run pytest tests/unit/scripts/ci/test_chain_replacement_parity.py -q gives 25 passed. A planted lost branch fails P2. A planted surviving mutant fails P1. A no-op deletion passes. test_missing_input_fails_closed covers each measurement set to None, plus an absent inputs file and a malformed one, and all exit 1. ruff and mypy are clean.

Lab step (h201, this branch merged locally with OMN-19710's head d6572dad)

measure over the four pilot handlers (handler_delegation_workflow.py, handler_delegation_routing.py, handler_compliance_loop.py, handler_routing_intent.py). T was the 21 cases of test_omn17802_v2_terminal_route_boundary.py and C was OMN-19710's 2 real delegation chains. The run took 433 s of wall time and exited 1, as it should on this pair:

• P1 FAIL: 4,685 mutants (the plan's 2,000 to 4,000 was an unmeasured estimate). Of the 1,182 killed by T plus C, only 868 are still killed by C alone, so 314 would escape.
• P2 FAIL: 2 lines and 2 branch arcs lost.
• P3 PASS, P4 PASS (20 of 20 seeds), P5 PASS (0.371 s against 1.605 s).

The first lab run found two defects, which are fixed in the second commit. (1) mutmut copied only the named handler files, so the tests imported the original code and every mutant read "no tests". (2) junit ids dropped the test class, so the time of every class-based case read MISSING.

Not runtime-affecting: a lab-only CI tool plus a dependency group that CI does not sync.

Out of scope

The deletion gate (OMN-19712, stacked on this branch) and the pilot (OMN-19713). Nothing is armed or merged by this lane. It is handed to the landing lane by a ledger MSG.

Evidence-Ticket: OMN-19711
Evidence-Source: OCC#11459

[github-actions]

✅ Architectural Review — PASSED

Errors: 0
Warnings: 0

What this checks

Rule	Description
ARCH-TOPIC-001	No hardcoded Kafka topic strings in handler code
ARCH-DI-001	No event_bus=None bypass
ARCH-DI-002	No direct Handler instantiation outside workflow_runner/adapters
ARCH-DI-003	No reinvented DI containers
ARCH-TOPIC-002	contract.yaml topics follow `onex.{cmd

---

No architectural violations found.

Static architectural lint — no model inference (OMN-14176).

[github-actions]

✅ Hostile Reviewer — PASSED

Blocking findings (critical/error): 0
Total findings: 5
Models succeeded: qwen3-review,gpt-oss-review

---

Gate semantics

Verdict	Meaning	Blocks merge?
passed	>=2 models succeeded, no critical/error findings	No
blocked	CRITICAL or ERROR findings found	Yes
degraded	Fewer than 2 models succeeded (infra unavailable/timeout)	Yes (OMN-15110)

Powered by omniintelligence.review_pairing.cli_review — node-based adversarial review via HandlerLlmCliSubprocess (OMN-8524)

[jonahgabriel]

Closing as superseded. omnimarket#2950 was stacked on this branch and squash-merged to dev as befa9cc, carrying every commit of this PR (head a350f3d).

Verification from the canonical clone: git diff origin/dev a350f3d4 -- scripts/ci/chain_replacement_parity.py tests/unit/scripts/ci/test_chain_replacement_parity.py is empty, and the only remaining differences in pyproject.toml and uv.lock are newer dev content (the 0.4.234 version bump and entries from other merged PRs); the parity dependency group this PR added is on dev. Nothing is left to land.