VED-011 marriage election implementation evidence

Date: 10 October 2026. Status: selected marriage scope accepted locally. This receipt accompanies the selected source standard and approved implementation plan. Both full HTTP cases and all affected follow-up checks pass. The catalogue reports source_scoped_public. This does not admit a release or deployment. The subsequent performance receipt records an accepted native recurrence optimization and exact full-response comparisons. This document retains the original implementation measurements and fingerprints as historical validation evidence.

1. Scope and evidence ownership

The finite MC Avasthi 2004 ordinary marriage astronomical timing profile has 42 rule-family contracts, nine remedy/composition contracts, M01–M30 and F01–F09 coverage. Personal assessment and the VV/MC Godhuli composition are separately selected. The source worksheet distinguishes textual rules, commentary, and named Moira interpretations. Other VED-011 purposes are open.

The MC and SS scan fingerprints and independent transcriptions are retained in tests/fixtures/marriage_election_sources.json and tests/fixtures/marriage_visibility_cases.json. Private PDFs stay outside the repository. Source arithmetic is tested independently of runtime output. ERFA altitude roots independently check the frozen-horizon time-degree construction at five northern/southern/equatorial cases, at 2e-12 degrees. marriage_visibility_reader_brackets.json adds twelve real-reader input brackets covering all six event roles/branches and both hemispheres. Every endpoint is checked by independent ERFA altitude roots against the separately transcribed source threshold and direction. These coordinates are generated inputs, not a primary planetary-position or event-date oracle. This does not establish a paired external oracle for the entire marriage profile, nor observed physical visibility.

Serving-reader checks use the actual DE441/LE441 kernel, including its lunar tidal identity, native bounded record accessor and loaded clock/frame data. Raw response and diagnostic artifacts are under C:\dev\outputs\ved011-marriage-validation-2026-10-10. Generated worked cases are computation/transport evidence, not independent expected source answers.

2. Runtime and commands

Baseline engine is e7107267942d1926da9bcf8ba3d1299b4205ccf9; generated wiki is 0c79a578942db2d46e11f203bc6a9c57e730ba27. All repository execution uses .venv\Scripts\python.exe, Python 3.14.3, Moira 6.9.9. Validation disables downloads, enables test mode and strict known-issue expiry. Known issues are empty. Version, dependencies and release state are unchanged.

$env:MOIRA_NO_DOWNLOAD='1'
$env:MOIRA_TEST_MODE='1'
$env:MOIRA_STRICT_KNOWN_ISSUES='1'
$env:PYTHONUTF8='1'
$env:MOIRA_MARRIAGE_RECEIPT_DIR='C:\dev\outputs\ved011-marriage-validation-2026-10-10'
.\.venv\Scripts\python.exe -m pytest -o addopts='' -o junit_family=legacy -q tests/integration/test_muhurta_marriage_ephemeris.py

The two expensive HTTP cases can execute in separate processes using -k real_http_windows and -k high_latitude_snapshot; the remaining three deterministic resource-budget cases use -k real_resource_budget. The optional receipt directory records raw requests, responses and timings.

3. Executed verification ledger

These slices overlap and must not be added into a fabricated total.

Slice / saved result	Observed result	Evidence scope
All unit/server paths matching muhurta, panchanga, lunar_month, or vedic_second_pass; vedic-regression.xml	1,759 passed; 128 planetary resource uses, all executed	Broad affected Vedic regression before the final Newton repair; affected marriage tests rerun below
Marriage source, enclosure, windows and server suites; root-repair.xml	182 passed; seven real resource uses, no skips	Includes the repaired Newton candidate, strict catalogue/limits, polar seam, source and transport cases
Full targeted marriage + budget slice; targeted.xml	199 passed, one expensive case deselected; ten real resource uses	Earlier focused checkpoint, retained as history rather than substituted for final full acceptance
Public API/doctrine/discovery, native import, SPK reader and server error mappings; public-native-errors.xml	95 passed, two skipped; ten real resource uses, none skipped	Existing optional comparator skips are separate from resource execution and from the new native record tests
Visibility and window contracts after the worked-bracket addition	38 passed	Independent geometry for all six real event branches, complete-empty versus indeterminate-empty output, seed-budget and polar-gap regressions
Synthetic SPK type 2/3 bounded records	Two passed	Exact independently authored coefficients, both native file/reversed storage orders, immutable copies and bounds rejection
Current complete marriage unit/server and three resource-budget cases; marriage-final-focused.xml	209 passed; two full HTTP cases deselected; eleven real resource uses, no skips	Final numerical/transport regression before the separately recorded public export repair
Marriage source/contract and public API drift; public-identity-final.xml	82 passed	Includes identity parity for every owning marriage export across root, vedic and facade
Current marriage, docstring and resource-budget acceptance; marriage-current-acceptance.xml	223 passed; eleven real resource uses, no skips	Precedes the polar parent-domain repair below; its affected cases are rerun separately
Final marriage, docstring and resource-budget acceptance; marriage-final-acceptance.xml	225 passed; eleven real resource uses, no skips	Includes the required-parent versus unused-pad polar witness regressions; only the two full HTTP cases are separately selected
Release-hardening selection, 69 named test files; release-hardening-current.xml	2,676 passed, four public-guard failures, one conditional skip	All 251 planetary and one supplemental small-body resource uses executed; the four failures are repaired and rerun below
Complete public guard, API drift and marriage source suite; public-identity-current.xml	104 passed	All four release public-guard failures resolved; explicit 41-symbol/four-method admission and identity checks include the window types
Earlier six-hour HTTP windows checkpoint; integration-dense-current.xml	One passed; three real resource uses, no skips; 1,815.19-second test	Successful baseline before the polar-domain repair
Final six-hour HTTP windows case, Delhi, Godhuli, 4,096-day history cap; integration-dense-final.xml	One passed; three real resource uses, no skips; 1,778.74-second test	45 certified cells, 88 gap-free supporting certificates, strict serializer, direct interior probes, personal direct HTTP parity and borrowed-reader lifetime; response exactly equals the earlier successful baseline
Real HTTP instant, latitude 70, Europe/Oslo; integration-high-latitude-final.xml	One passed; three real resource uses, no skips; 1,558.76-second test	Completed operation with explicit unavailable Lagna/visibility evidence, exact engine/HTTP parity and retained reader ownership; no eligible result inferred

The broad slice command is:

$testPaths=@(rg --files tests/unit tests/server | Where-Object { $_ -match '(muhurta|panchanga|lunar_month|vedic_second_pass)' })
.\.venv\Scripts\python.exe -m pytest -o addopts='' -q @testPaths

4. Adversarial findings and repairs

• Periodic interval subdivision re-enumerates physical rays when its local angle lift changes. A parent numeric target cannot silently exclude a child representation of the same crossing.
• Native record arguments enclose the serving two-stage floor/carry path; frame, clock, observer and finite-difference motion calculations retain source-derived whole-cell rounding guards and source identities.
• A real node search exposed a false Newton candidate near JD 2461222.3226968, separate from the true descending crossing near 2461222.3225176. The fresh interval value excludes the candidate. Every Newton-contracted leaf is now rechecked before retention. Independent analytic and real-reader regression tests pass; actual tangencies remain unresolved rather than discarded.
• Historical Mercury nature is evaluated at the causal event, with possible intermediate associations retained across uncertain bands. Both 27-star occupancy clearance and unequal 28-star piercing clearance require an actual complete Moon traversal. Missing history is not a clear finding.
• Calendar-owned brackets join the window partition; achieved uncertainty is preserved. A calendar bracket wider than its one-second admission becomes unavailable before constructing the canonical month result.
• Polar Lagna midpoint unavailability cannot certify a whole window. Unsupported Lagna regions retain gaps rather than sampling through None.
• Request caches are owned by the request, including native record caching; finished requests and borrowed readers are collectible. Pools retain a pinned reader order without transferring ownership.
• Preflight history caps cannot be silently exceeded by the fixed calendar parent search. Seed bisections and adaptive apparition calls share the request root budget. Incremental and final output accounting do not silently truncate results.
• Pydantic-wrapped budget exceptions retain the engine's named REST error and structured limit/count/stage details. Catalogue defaults and maxima agree with engine admission and OpenAPI.
• The final public-surface audit found four historical-evidence types absent from moira.facade. Its imports and curated exports now match root and vedic; the regression checks every owning marriage symbol by identity.
• The release docstring gate identified missing actor/vessel documentation in 34 new classes. Their architecture, state and resource ownership are now documented. docstring-only-ast-receipt.json confirms identical computational ASTs before and after these documentation edits; all 13 docstring governance tests pass. The original failure is retained in docstring-current.txt.
• The high-latitude full HTTP attempt reached 2,000,001 reader calls at Jupiter_east_appearance after 2,397.09 test seconds. REST returned the expected named budget error, but the intended unavailable-result acceptance failed. The failed XML and raw budget response remain preserved. Its source history contains actual undefined horizons. Such a witness inside the required adjacent-event parent now records an incomplete domain and returns unavailable availability evidence before futile refinement. It is not a zero-exclusion proof, a latitude blacklist or an increased work limit. A gap outside the required parent does not trigger this rule. Both boundary cases have unit regressions. The real latitude-70 Jupiter/Venus component probe now finishes in 7.43 seconds, at 7,101 evaluations and 171,349 reader calls, retaining both unavailable-domain reasons. Both complete HTTP cases pass after this repair.

The earlier full HTTP run correctly returned an uncertified result and its acceptance assertion failed on the node candidate. It is preserved in integration.xml (one failure, three passes), not reported as successful acceptance. Superseded interrupted attempts are not passing validation. The two optional comparison skips in the public/native slice are test_native_daf_catalog_matches_jplephem_on_moira_written_kernel and test_native_chebyshev_payload_matches_live_jplephem_segment_data: that development-only comparator is unavailable. No package was installed for them. The new bounded-record tests and required serving DE441 cases execute without it. The release-hardening skip is the existing conditional test_no_aspect_voc_last_aspect_is_none: its 30-day scan found no matching no-aspect window. It is unrelated to marriage and is not a resource skip.

5. Operational measurements and closure gate

An earlier full point diagnostic completed in 1,706.73 seconds with 186,742 evaluations, 1,253,571 reader calls, 49,460 then-counted root iterations, 640 transitions and 335 historical cells. Its one numerical gap was the subsequently repaired node candidate; it is not final admission evidence. Its raw ordinary assessment had complete source-rule coverage. The same raw evidence round-tripped for all five regional selections, with remedies off and all seven enabled (ten cases). This is shared-evidence parity, not ten independent ephemeris runs.

Two southern-hemisphere apparition-only runs retained Jupiter and both Venus branch witnesses with no numerical gaps. Their elapsed times were 145.07 and 166.38 seconds; respectively 20,160/21,158 evaluations and 278,495/289,996 reader calls. These are component measurements, not full marriage-request timings.

The final six-hour Delhi HTTP window calculation took 1,701.96 seconds (the complete test, including extra interior/transport checks, took 1,778.74 seconds). It measured 188,201 evaluations, 1,286,318 reader calls, 65,877 root iterations, 705 transitions, 45 output cells and 334 historical cells. All 88 supporting certificates are gap-free under the declared arithmetic model; every ordinary source-rule assessment has complete coverage. All 45 cells are restricted, so the eligible list is empty and the indeterminate-cell list is also empty. This is a completed negative search for this selected profile, not missing search evidence.

The result contains 46 boundary bands, 62 shared evidence parts and 613 distinct findings. Its compact response is 1,662,692 UTF-8 bytes; conservative output accounting charges 1,885,295 bytes. The widest achieved band is 3.745665 seconds, preserved instead of being relabelled with the requested 0.1-second leaf tolerance. Supporting node/apparition domains reach 512 days on either side of the anchor, far beyond the six-hour requested interval. Raw request, positions, findings, bands and certificates are preserved in real-http-windows.json; real-http-personal-direct.json records the same evidence with an explicitly incomplete personal context.

The repaired high-latitude HTTP instant returned HTTP 200 in 1,510.44 seconds (1,558.76 seconds for the full test). It used 167,429 evaluations, 1,113,819 reader calls, 59,056 root iterations, 584 transitions, 334 historical cells and 1,383,574 conservatively accounted output bytes. The compact response is 1,229,746 bytes. Its known restrictions are retained with incomplete coverage. Lagna and both planetary availability contexts are explicitly unavailable. The Jupiter and Venus required-history records retain 248 and 231 undefined-horizon witnesses respectively; the numerical status is correctly not_certified, not an eligible or complete-coverage claim. real-http-high-latitude-snapshot.json retains this full result.

Neither successful measured request required a raised evaluation, reader, root or output limit. The supporting-history override in the dense case was 4,096 days, but its actual largest discovery radius was 512 days; its search did not approach the default 2,048-day bound. The default 300,000 evaluations, 2,000,000 reader calls and 100,000 root iterations provide roughly 50 percent headroom above the larger observed run. Other caps are finite operational ceilings with independently tested exhaustion, not promises that every admitted-duration request completes within all default budgets. The complete table is in the source standard and the engine/REST catalogue.

6. Lineage and publication boundary

The governing objects are source finding, raw witness, finite named exception, separate evidence coverage and a half-open certified partition. Discrete tables follow their collated source; historical occurrence and clearance have explicit state/parent ownership. Numerical helper boundaries follow the serving SPK record, clock, frame, horizon and interval-isolation objects. No external engine slot order, oracle-tuned threshold, quadrant repair or copied marriage-calendar staging is used. Secondary calendar agreement does not carry the proof burden.

The source-level numerical review closed its identified false-exclusion defects under binary64_RN_gradual_underflow_libm_4ulp.v1. The conditional arithmetic assumptions are explicit; this is not platform-independent formal verification. Both final full composition runs pass at this declared scope.

The final current-source marriage/governance slice comprises 227 passing tests: 225 in marriage-final-acceptance.xml, plus one in each final HTTP receipt. No required resource test is skipped. The wider latest-per-node reconciliation contains 3,901 passes and three separately disclosed skips; overlapping runs count once. The original failed receipts remain preserved, with the affected tests' successful follow-ups identified above.

At the implementation checkpoint this package was local and uncommitted. The user subsequently authorized wiki-first source publication; the register's publication checkpoint records that authorization and the remaining scope. Release and website/Urania adoption are separate. Other VED-011 purpose profiles remain open and are not included in this marriage admission.

7. Final source identity and documentation checks

source-state-final.json records 41 runtime, test, source-fixture and workflow files against the baseline above, hashing UTF-8 text with LF newlines. Every fingerprint matches the accepted implementation files before the publication-only whitespace normalization recorded in section 8. All 35 Python files in that manifest also parse with Python 3.10 grammar. The manifest excludes documentation, generated publication metadata and the compiled native binary; the serving binary has its own identity below.

Evidence object	SHA-256
source-state-final.json	2c5c8d8bce8d44897fec85419ede827d4c2a5ce68b6bea406f1867643da1e9a2
latest-results-final.json	96667414108de08883599c53b04c077cae1b6b68c188193f216c4ddbb018c20d
Serving DE441/LE441 kernel	9a824e6a8ce3d39f8a513b17278ff7b7ad226f964c76d2019b0a51d8808d6ff5
Numerical source tables	b3eb6825d4bf354bbbd69832c13c63ee094cbacba0bb4844442fc97b72f7329a
Active native backend	2a39bf99dced0d2f5bf6b538f6aef2e0e6d932a005ac6dbfa1fccdff77960ff1

After both full HTTP runs passed, the final catalogue admission flag and typed REST schema were checked again: admission-final.xml has two passes and no skips or failures. These repeat existing cases and do not increase the distinct-test totals above.

Final checks pass for documentation consistency, release identity, Hellenistic generated inventories, REST route inventory, website-document publication metadata and the generated Git wiki mirror. CI's source-lint selection (F401,F821,F822,F823) and both repositories' git diff --check also pass. The broader scoped Ruff comparison retains 68 unchanged legacy diagnostics (60 root-module E402, five facade F811 and three router E402), with no new diagnostics; ruff-baseline-comparison.json preserves that comparison. No unrelated lint cleanup is claimed.

8. Source-publication verification

The authorized publication starts from engine e4e9a72 and generated wiki 9edb51f. The intervening small-body and visibility-documentation commits are preserved. All 41 source fingerprints, seven aggregate XML fingerprints and the active native-backend fingerprint above still match.

On that combined checkout, the four test_muhurta_marriage* unit files, marriage server tests, docstring governance and marriage integration tests were rerun with -k 'not real_http_windows and not high_latitude_snapshot': 225 passed, two deliberately deselected, zero skipped, in 46.74 seconds. All eleven selected DE441 resource uses executed. The exact invocation used the same project runtime and strict no-download environment as section 2, plus -o addopts='' -o junit_family=legacy -q; its XML is publication-acceptance.xml in the receipt directory. The two expensive full HTTP receipts remain the unchanged successful runs recorded above, rather than being reported as repeated publication tests.

The staged whitespace gate subsequently identified one terminal blank line in tests/unit/test_muhurta_marriage.py, which had been untracked during the earlier unstaged whitespace check. Publication removes that blank line. publication-whitespace-receipt.json preserves both normalized-text hashes and confirms an identical full Python AST, including source locations. Original validation manifests remain immutable; publication-manifest.json records the exact normalized files being committed.