Runner: mansart-persistence/mansart-jpa-tck, out of the reactor, using the official 3.2.1 suite and PostgreSQL 17 DDL. By default it runs a derived copy of the spec-tests jar carrying the local fixture patch TCK-BUG-001: the byte-exact backport of upstream commit 1fea05e (jakartaee/persistence#1175), which removes the <delimited-identifiers/> unit default from two orm.xml fixtures. Official artifacts are never modified and TCK_FIXTURES=official runs the untouched jar. The P8 full run passed 2013 tests; compared test by test with the preserved P7 baseline (1063 passes), 950 tests newly pass, with no regression or added/missing tests. This includes one earlier converted-literal query defect exposed and fixed during attribution, not later-milestone implementation. Required SINGLE_TABLE and JOINED, polymorphic loading and JPQL TYPE are delivered. At P7, the focused gates totalled 307 / 312 passed, four XML-mapping-dependent errors and one skipped. The XML-dependent errors were resolved in P10. The initial P10 baseline passed 2121 tests but ignored <delimited-identifiers/>. The pre-P11 untouched run with correct quotation reported 2096 passes, 35 errors, four skips / 2135, with no added/missing tests: 25 regressions are confined to core.annotations.nativequery (1 / 12) and core.entitytest.apitests (7 / 21), whose delimited-default mappings conflict with the unquoted PostgreSQL DDL. core.override 26 / 26, core.callback 66 / 66, core.relationship.descriptors 8 / 8, se.descriptor 1 / 1 and core.StoredProcedureQuery 40 / 40 retain their results. That historical run also contained ten second-level cache errors, subsequently fixed in P11. With TCK-BUG-001 (2026-10-10, provider unchanged) the P10 full run reported 2121 passes, 10 errors, four skips / 2135; the 25 fixture errors passed with zero regressions and the same test identities. P11 subsequently fixed the cache assertions. The latest local patched full run reports 2131 passes, 0 failures/errors, four skips / 2135, unchanged against the previous P11 run. These patched scores are local development results, not official results or formal certification. The P8 gate totals 925 tests: 924 passed, no failures/errors, one upstream skip: metamodel 258 / 259, Criteria 650 / 650, entity graphs 13 / 13 and repeatable named graphs 3 / 3. The P8 runtime/processor/module-path reactor reported 470 passing tests, including populated canonical fields in entity packages neither exported nor opened to the provider. Optional TABLE_PER_CLASS remains explicitly refused; decision D5 stays open. Schema/container contracts, XML, and caching/validation are delivered. Data’s incompatible canonical plural field kinds need a maintainer-approved producer reconciliation; Data is unchanged. No formal certification or Web Profile is sought.

The script exits zero after a reported run even if tests failed, but nonzero on setup or fixture-provenance failures; inspect target/tck-report-persistence.txt and the failsafe XML counters.

Runner: mansart-jakarta-data/mansart-data-tck, out of reactor (standalone POM Model 4.0.0, no <parent>). This decouples the implementation and official TCK release cycles. The historical Maven Model 4.1.0 parsing constraint no longer applies since the move to Maven 3.9.x and Model 4.0.0.

Harness architecture:

Local reproduction:

Major historical bug: BUG-20260505-01 (TCK entities not metamodelled because compiled without the processor in the APT classpath). Resolved by the M7 runtime runner (MansartRuntimeRepoCreator). See BUG.md.

Runner: mansart-transactions/, see TCK.md.

The official TCK distributed by Eclipse is a historical Sun tsharness harness:

No Surefire-scannable jar exists — dependenciesToScan against jtatck.jar only finds support classes, not tests.

Status:

Manual reproduction:

None to date. The Jakarta Data and Jakarta Transactions TCKs run in CI on H2 in-memory, with no external dependency.

Prerequisites: non-public official Jakarta artefacts installed in the local M2 — procedure documented in each runner’s README.