GATE A, FUNCTIONAL. Phase 2, 28 July 2026, re run at phase close so the whole
bundle is one measurement of one commit rather than a scrapbook of runs taken at
different times. FIPS on throughout, GOFIPS140=v1.0.0.

VERDICT: PASS. Gate A remains OPEN on exactly one blocker, which is not a
technical result and is recorded at the end of this file.

WHOLE UNIT SUITE
    packages passing: 52   failing: 0

THE EXTERNAL BOUNDARY, internal/platform/scrub, 20 tests, all PASS:
    TestEveryBindingSiteIsBoundByTheMigration
    TestMigrationBindsNothingOutsideTheInventory
    TestNoProvenanceConstantIsDeclaredInApplicationCode
    TestBindingSiteDetectorPositiveControl
    TestCodeAndMigrationVocabulariesAgree
    TestCandidateRowIsRefusedByProvenance
    TestTenantUploadedDocumentTextIsRefused
    TestBackstopCatchesAMisdeclaredCandidateRow
    TestBackstopLimitIsKnownAndDocumented
    TestProvenanceVocabularyIsClosed
    TestGenuinelyPublicNoticeTextIsPermitted
    TestSuitePositiveControl
    TestBoundaryInventory
    TestOneTenantElementRefusesTheWholePayload
    TestEmptyProvenanceSetIsRefused
    TestValidAbstractSignalPasses
    TestClientFactsAreRefused
    TestMarshalRefusesToEmitALeak
    TestEmptyModelVersionIsRefused
    TestUnregisteredModelVersionIsRefused

  Five of those exist to make the suite honest rather than to test a feature.
  TestSuitePositiveControl proves the gate is running at all. TestBindingSite
  DetectorPositiveControl proves the binding site inventory can fail.
  TestBackstopLimitIsKnownAndDocumented asserts the CONTENT scan's honest limit,
  that a misdeclared stripped candidate row is not caught by pattern matching,
  which is the entire argument for provenance being the primary control.
  TestEmptyProvenanceSetIsRefused fixes the case where silence used to pass.
  TestOneTenantElementRefusesTheWholePayload fixes the case where a mostly public
  payload with one tenant paragraph used to pass.

THE RACE, internal/integration, against staging, tag integration:
    TestIngestDuringPromotionElectsOneWinnerAndCorruptsNothing  PASS
        ingest rewrote the notice 29 times, overlapping the race throughout
    TestIngestNeverRewritesAnApprovedRequirement                PASS
        approved requirement survived 20 ingest rounds byte for byte,
        md5 6acbad7d5a24cb161b141e1d5a99e717
    TestPromotionSuitePositiveControl                            PASS

  The overlap count is asserted, not hoped for. An earlier version of this test
  raced two operations that did not actually overlap, and it passed, which
  proved the harness ran and nothing else. The guard now fails the test if the
  ingest loop did not interleave with the promotions.

GATE A IS OPEN, ON EXACTLY ONE BLOCKER, AND IT IS NOT TECHNICAL.
  The remaining item is a REAL SOLICITATION PACKAGE from Joseph. Every functional
  claim above is proven against synthetic fixtures and staging data, which is the
  right way to prove the mechanism and cannot prove the fit. A package that a
  human actually received, with its attachments, its section C, and its
  irregularities, is the only thing that tests whether the scoping and
  normalization path handles the documents this product exists to handle.

  Saying Gate A is closed on synthetic evidence alone would be the kind of claim
  this build has spent the day removing.