[Bug] Require physical broker tests in the native execution context #521

Describe the Bug

Seven physical Mac broker tests live in an ordinary portable test module. The Linux candidate review correctly reports their nine default parameter instances as skipped setup outcomes. No checked-in native job supplies the two required build roots or runs these exact functions. Existing native jobs prove related contracts with different assertions.

To Reproduce

Inspect the exact-head candidate3.12 job for PR #519 at ccbff95. It emits nine TEST_OUTCOME_NOT_PASS errors from tests/unit/specfact_code_review/run/test_native_execution.py, all skipped, phase setup.

The seven functions require SPECFACT_NATIVE_EXECUTION_BUILD or SPECFACT_NATIVE_STALLED_STARTUP_BUILD. No checked-in workflow provides those inputs. Default repetitions produce nine cases. The existing parametrization produces 7R + 2 cases, or 702 at R=100.

Expected Behavior

• Portable request, session, path and source assertions execute in ordinary Linux discovery.
• Every existing physical function, assertion, guard and parameter identity remains in an explicit required native context with valid normal and separately signed stalled inputs.
• Missing inputs, empty collection, skipped, failed, errored or duplicate outcomes fail native acceptance.
• All current nine OS/ABI cells, 100-repetition checks, delivered component bytes, signatures, cleanup deadlines, customer gates and job budgets remain intact.

Actual Behavior

Linux cannot execute the physical corpus and fails its nonpassing-test gate. The current nine-cell candidate acceptance, customer matrix and three-OS boundary lane execute different contracts. Historical local signed-stall evidence is recorded in the native TDD document, but is not a current checked-in producer or CI invocation.

Environment

• Linux candidate workflow on Ubuntu24.04, CPython3.12.
• Physical target is the existing Darwin ARM64 support matrix.
• PR fix(ci): preserve fixture identities and require native broker proof #519, branch codex/ci-fixture-identities.
• Repository CI/test ownership maintenance for specfact-code-review.

Validation Output

The linked service log reports authoritative FAIL with nine test errors. Their seven function identities are the self-test plan, held-worker cancellation/timeout/EOF, postlaunch file storm, unsafe launch tree, postlaunch symlink, stalled signed bootstrap and controller EOF during startup.

Additional Context

Parent Feature #163. Proposed OpenSpec change is ci-05-native-test-ownership. This is a separate maintenance correction from #518's fixture identities and #520's PID publication reader. It does not reopen rejected #460 or authorize a runtime release.

Use the established explicit proof ownership and existing native acceptance lane. A genuine stalled fixture must reach marker-stage error106 and independent survivor checks. Its producer must operate only on isolated test inputs with actual signing and verifier admission. Production C, signed module assets and delivered archive bytes remain unchanged. Preserve the frozen PID59 test file. A separate old single-owner Hatch expectation must receive actual baseline RED before its minimal ownership assertion is corrected.

Specification, tests and genuine failing evidence precede implementation. Local source/fixture proof, full physical execution, normal gates and independent review remain separate requirements. Current runtime and monetary cost are unmeasured. If the preserved corpus cannot fit existing budgets, retain the failure and reconsider the design without reducing acceptance or raising limits.

The owner has authorized useful guideline-compliant overnight corrections in this chat. No merge, rebase, force push, retargeting, new warning waiver or protected machine authority is granted. The separate independent Z3 mismatch remains visible and is not fixed by this ownership correction.