Moira 6.10.0 release validation

Date: 12 October 2026. Release scope: completed engine work through the selected VED-014 profile; Vedic feature expansion pauses at this boundary. Starting commit: a3a2dabbf2e9f9dbee0cacdd036b6f399188dfe1. Release checkout: C:\dev\moira-release-6-10-0. Runtime: its project .venv, Python 3.14.3, rebuilt native extension and editable distribution reporting 6.10.0. tests/KNOWN_ISSUES.yml remains empty. The local REST/test environment uses FastAPI 0.136.3, Pydantic 2.14.1, Starlette 1.2.1, HTTPX 0.28.1 and pytest 9.1.1.

The user authorized review, repairs, release validation, wiki/engine publication and the 6.10.0 package release. The separate natal-discovery and light-box work was explicitly excluded and preserved in C:\dev\moira. The release checkout contains neither those files nor their shared app/OpenAPI registration hunks. No website application deployment is part of this release.

Code Review

Verdict: APPROVE after the repairs below. Confidence: HIGH for the reviewed Kalachakra change; this is not a claim that every engine family was adversarially reviewed in this pass.

The review followed source row membership and balance into full-parent MD/AD construction, exact/projection boundary ownership, current lookup, immutable result validation, reader/frame receipts, facade admission, REST schemas, serialization and sanitized errors. Frozen independent fixtures and all fourteen source decisions remain authoritative at their stated scopes.

Priority	Finding	Repair and regression
P2	IndexError inherits LookupError; a generated astronomy indexing defect was incorrectly translated to resource-unavailable 503.	Catch indexing defects before resource lookup errors. A real HTTP regression first failed with 503, then passed with sanitized kalachakra_internal_invariant 500.
P2	The frozen Kalachakra fixture hash changed under Git checkout with core.autocrlf=true.	Pin this one fixture to LF. Real temporary-repository clean/checkout tests exercise both Windows and POSIX settings; the numerical data and recorded digest are unchanged.
P1 release gate	Existing main CI compared the September apsidal audit against the newly packaged October manifest.	Bind the historical audit to the byte-exact archived September manifest. Preserve its original hash and identity, all scientific evidence and all tolerances. Separate current-packaged-manifest tests still enforce October bytes.

The first defect changes error classification only. The second repairs test provenance; it changes no catalog, ephemeris or numerical calculation. A stale current API-reference sentence reserving Kalachakra was also reconciled.

The failing baseline was Release Hardening run 38293347900, job 114958556151. The targeted repair run passed 12 tests. Its initial red run had one expected new-regression failure and 11 passing catalog checks.

Integrated validation

Validation is run with MOIRA_NO_DOWNLOAD=1, MOIRA_TEST_MODE=1, MOIRA_STRICT_KNOWN_ISSUES=1 and HYPOTHESIS_PROFILE=moira-ci. Network denial remains active; marked HTTP tests use loopback only. The isolated Windows environment has no OS IANA database. After preflight identified that missing prerequisite, the final runs explicitly set PYTHONTZPATH=C:\dev\moira\.venv\Lib\site-packages\tzdata\zoneinfo to the existing local timezone data (tzdata 2026.3, IANA 2026c). The interrupted unconfigured runs do not count as acceptance; no timezone fallback, numerical tolerance or dependency was changed. Resource discovery resolves C:\Users\nilad\.moira\kernels\de441.bsp; the harness performs its own content/capability admission.

The selected release suites cover all Vedic unit/server/integration filenames matching the delivered families, the Release Hardening and Release Acceptance workflow selections, native/frame checks and the documented local astronomical acceptance cases. The exact file selections and JUnit receipts are retained at C:\dev\outputs\release-6-10-0\test-selection.json, kernel-free.xml and resourced.xml. This is an integrated release selection, not the literal full repository suite or a new all-body/all-epoch external accuracy certification.

The first combined collection exposed six pre-existing duplicate test basenames across unit/server/integration directories. --import-mode=importlib permits their joint collection without renaming tests or changing runtime code. The final kernel-free run uses four isolated pytest workers with --dist=loadfile; no selected file declares a serial marker, and each file's tests remain together. The real-ephemeris selection runs serially.

The completed kernel-free selection passed 8,861 tests in 1,007.69 seconds, with no failures, errors or skips. Its eight warnings are the same two optional test-client deprecations repeated across four workers; no runtime dependency or numerical behavior was changed to suppress them. This includes 250 Kalachakra cases; its two real-resource cases belong to the second selection.

The exact final invocations, after setting the environment above, are:

$selection = Get-Content C:\dev\outputs\release-6-10-0\test-selection.json | ConvertFrom-Json
$paths = $selection.kernel_free
.\.venv\Scripts\python.exe -m pytest @paths --import-mode=importlib -n 4 --dist=loadfile -m 'not requires_ephemeris and not external_network' -o addopts=-ra -v --tb=short --maxfail=10 --junitxml=C:/dev/outputs/release-6-10-0/kernel-free.xml
$paths = $selection.resourced
.\.venv\Scripts\python.exe -m pytest @paths --import-mode=importlib -m 'requires_ephemeris and not external_network' -o addopts=-ra -vv --tb=short --maxfail=10 --junitxml=C:/dev/outputs/release-6-10-0/resourced.xml

The real-resource selection passed 548 tests in 7,067.06 seconds (1:57:47), with 7,210 cases deselected by its complementary marker expression, no failures, errors or skips. The harness admitted and executed 551/551 DE441 resource uses, with one content probe and zero resource failures or skips. The twelve warnings comprise the two test-client deprecations, three existing JUnit record_property compatibility notices and seven existing class-scoped fixture deprecations. None was suppressed or converted into an accepted failure.

The two selections contain 9,409 unique passing tests; a JUnit node-identity check confirms no overlap. This includes 252 Kalachakra tests, including the real-kernel astronomy and chart/direct HTTP cases. Full marriage, VV/KP, housing, business and travel acceptance ran to completion with their existing boundaries, source rules and validation thresholds. Timings describe this integrated run and are not a latency guarantee or a controlled benchmark.

The JUnit SHA-256 fingerprints are:

• kernel-free.xml: bb0ffff96401de0dc3ac2ad5f4a2937386c67f66b50a6c2cdf8cc4e6b307a49e
• resourced.xml: 43119003e5e257ce658cf8953578ca76c2ef6582d3e772aa24bf882845aa4d2a

The release check selection also covers the existing offline eclipse comparison and NASA cases, houses, sidereal and progressions references, and occultation structural checks under their owning tolerances. Secondary-engine corroboration remains distinct from primary-source validation.

The 375-file engine docstring preflight, scoped changed-file Ruff, source correctness lint (F401,F821,F822,F823), release identity, doc consistency, Hellenistic inventory, REST reference, generated wiki and publication-candidate checks passed. This is not a claim of a clean repository-wide Ruff audit. Candidate publication contains 26 documentation entries and 26 release entries.

Hosted candidate checks

The reviewed source candidate fd3496195c4a4f9fb3987f3d4b116153ab98538e was pushed after wiki 171e194. Its Release Hardening and Release Acceptance Matrix both passed. Hardening includes actual Python 3.10 server/OpenAPI compatibility, the Kalachakra feature slice, generated documentation and the repaired historical catalog check. These CI kernel-free gates supplement local real-ephemeris checks.

Publication requires the final receipt commit's hosted gates to pass before tagging. Exact-tag publication and the downloaded PyPI artifacts receive separate verification: PyPI's 20-wheel/one-source-distribution inventory, downloaded Windows wheel and source-archive hashes, actual Windows CPython 3.14 native import, Kalachakra engine/projection/current lookup and real HTTP success/rejection parity. Deferred natal-discovery/light-box additions must be absent. The tagged release record records those final CI and publication outcomes. Candidate artifacts are not relabelled as exact-tag release evidence.

Preserved evidence and next work

The Kalachakra implementation receipt retains the earlier 794-test package checkpoint. Overlapping preliminary and repair runs must not be added to the integrated release test total. Its source fixture SHA-256 remains f646f4ab82907988da8cdff128785a62e0c9502a66019c15c70885505952f823. Historical asteroid audit/support bytes and archived manifest hashes remain unchanged. Publication does not rewrite them to assert new scientific scope.

The remaining-work register retains sixteen selected delivery checkpoints and eight open/bounded packages. Special Lagnas and further research can resume after the break; source conflicts, unsupported Kalachakra traditions, deeper levels and unevaluated eligibility remain explicit. No new Vedic feature scope is opened by release preparation.