VEDIC_ADVERSARIAL_REPAIR_VALIDATION_2026 10 09 - TheDaniel166/moira GitHub Wiki

Vedic adversarial repair validation — 9 October 2026

Status: LOCAL_COMPLETE. All eleven repairs closed. This receipt governs VA-01 through VA-11 from the review of engine b578ec89b90f709c8e64d5d9405ca06d29fcc69c. The approved plan retains all five packages. Implementation and documentation are local; no commit, push, tag, package publication or deployment is included.

Source packet and admission decisions

Primary witnesses are the user's local editions, inspected as page images where OCR obscured numbers. PDF page numbers below are one-based file pages; printed page numbers are identified separately. The source files remain in C:/dev/ASTROLOGY-BOOKS-DATABASE; no scans are copied into the engine package.

Edition	Local path suffix	SHA256
Raman, Bhava and Graha Balas, 1996	Books by Authors/BV Raman/Bhava and Graha Balas 1996.pdf	`727c8063f64816a0c1c7f6bc577c249629854e980bf860bf4cfd8de16ec9bab2`
BPHS, Santhanam I	Classics books/BPHS-Santhanam-Vol-1.pdf	`13609f1af89541e070c798b8c70c6bfd74a8b9a71a71388913bf898e9f9ddf3f`
BPHS, Santhanam II	Classics books/BPHS-Santhanam-Vol-2.pdf	`ad7172b615568eb812961c71cff352f354c5676b955bc34cfcfdcf431404cd45`

Source/source-derived fixtures are in tests/unit/test_shadbala_source_repairs.py and tests/unit/test_dasha_full_intervals.py. They are distinct from modern DE441 chart regression cases. The original audit's extracted pages, renders, source manifest and failing probes are preserved in C:/dev/outputs/vedic-adversarial-review-2026-10-09. New repair outputs are in C:/dev/outputs/vedic-adversarial-repair-2026-10-09.

Subject	Inspected source and admitted reading	Independent witness
Full-parent Dasha	BPHS II Ch46/51 proportional subperiod rules; full interval precedes birth/horizon intersection	Rational Ketu/Rahu and Yogini Bhramari/Siddha; half-open adjacency through five levels
Temporary relationships	Raman 1996 article 25: inclusive houses2,3,4,10,11,12	All 144 origin/distance combinations; same-sign is enemy
Saptavargaja	Raman PDF 27–28, printed 22–23, article 30; BPHS I PDF 264, printed 265,27.2–4	Raman Sun seven awards7.5,30,7.5,7.5,7.5,7.5,22.5 sum90; alternative scale10,30,10,10,10,10,20
Division assignment	BPHS I Ch6: Hora, Drekkana, Saptamsa, Navamsa, Dvadasamsa and unequal Trimsamsa	Odd/even D7 start, own-sign D12, all selected D30 boundaries; separate from generic harmonics
Ojayugma	BPHS I PDF 264,27.4½: Moon/Venus even, others odd in D1 and D9	Mercury1 degrees → 30;31 degrees → 0; two mixed parity cases→ 15
Nathonnatha	Raman PDF 40–42: linear distance from apparent local noon/midnight	Six exact solar fractions, opposite observer longitudes
Paksha	Raman PDF 43–44: continuous elongation, Moon doubling, nature qualification	179.9 degrees →119.933333333333 Sha; new/full phase neighbors
Tribhaga	Raman PDF 45–46: actual day/night thirds, Jupiter always 60	Six exact boundaries; Mercury/Sun/Saturn day and Moon/Venus/Mars night
Ayana	Raman PDF 64–65, article 75 and Example 33	Signed24-degree formula; Sun×2, Mercury absolute, Moon/Saturn sign reversal
Planetary war	Raman PDF 65–66, printed 60–61, articles 76–77	Strict<1 degree, lesser longitude, fixed-disc quotient, equal debit/credit, unchanged Chesta
War alternative	BPHS I PDF 283, printed 284,27.20	Distinct two-body Shadbala-difference wording; not silently merged into Raman's quotient
Precession fixture	ERFA `p06e` general precession at Einstein TT; retained tropical Moon/true-frame composition	Existing1e-9 balance tolerance unchanged; nutation independently checked at0.001 arcsecond

The Raman standard horoscope positions used for the seven-award Sun witness are from PDF 7: Sun 180°53′55″, Moon 311°17′19″, Mars 229°30′34″, Mercury 181°31′34″, Jupiter 84°00′49″, Venus 171°09′56″, Saturn 124°22′41″. Inputs are transcribed sexagesimal values, not fitted from present ephemerides. The component arithmetic tolerance is1e-12 Sha; whole-result sum validation uses the existing1e-6 Sha representation tolerance.

The bphs_santhanam_27 selector names an award scale, not a complete BPHS Shadbala implementation. It shares the explicitly selected Raman D1-only Moolatrikona prerequisite. Both use actual D1 degrees and D1 compound relations. No one-degree probe determines dignity. Complete worked-chart agreement is not claimed: retained osculating Chesta, sign-based Drig, ingress-based Abda/Masa and required Rupas are separately identified conventions. Sun's minimum remains6.5 Rupas; no threshold was adjusted to fit repaired scores.

Raman Example 33 contains visible arithmetic/rounding discrepancies. At printed Mars declination−22.45 degrees, the stated formula gives1.9375 Sha, while the printed result is1.84. Venus−4.96 gives23.8, while the print says23.75. Sun's 38.125 is printed 38.12. Tests follow the stated algebra at the printed input precision and disclose these differences; they do not widen a tolerance to turn the printed chart into an oracle. Actual true declinations can exceed 24 degrees; signed algebra is preserved and requires its context receipt.

Moon nature uses a named half-open interpretation of the eighth-bright to eighth-dark tithi interval, [84,264) degrees. Mercury association uses the explicit same-D1-sign malefic convention, with optional supplied benefic or malefic classification. Neither boundary/association rule is presented as the only classical reading.

The targeted war search inspected Raman articles 76–77, BPHS27.20 and their surrounding strength passages. These witnesses prescribe two-body behavior; no simultaneous multi-party allocation was established there. The admitted moira_simultaneous_raman_raw_pairs_v1 therefore computes each pair once against immutable pre-war bases, sums credits/debits, preserves signed nets and leaves Chesta unchanged. Exact longitude ties use a stable pair label and zero amount. This is an explicit Moira composition, not a claim that no other source could discuss multi-way wars. Fixed disc values9.4,6.6,190.4,16.6,158.0 arcseconds are the inspected table, not modern astronomical disc diameters.

Online primary-text corroboration included the available BPHS transcription and Raman 1996 book transcription. The hashed local editions and inspected page images govern the admitted rules.

Compatibility and operational contracts

Dasha period responses add full interval endpoints; alternate periods add explicit year metadata. The first visible child is the active child on the original full axis, not a newly restarted child of the birth balance.
Generic Varga partition uses exact binary-input integer ratios. The 65,880 D1–D60 boundary corpus has zero wrong signs and zero label/coordinate disagreements; 4,320 explicit D60 profile checks retain zero disagreements. Tiny negative cyclic inputs remain immediately below360 and no longer crash.
Full Shadbala and standalone Kala require an explicit context; standalone Sthana requires complete D1 positions. Dated callers derive context once. Synthetic inputs never silently borrow unrelated ephemeris positions.
REST Shadbala positions are apparent geocentric; observer coordinates govern houses and solar geometry. This repairs the previously mixed topocentric chart/geocentric context. Source/geometry/policy receipts travel through chart/full/network/Bhava and Lagna. Direct Lagna input uses the dataclass shape and preserves complete strength evidence.
Canonical war amounts are Kala adjustments. REST retains its existing shashtiamsas_transferred name with explicit adjustment_component. Deprecated chesta_transferred is unpopulated by new engine calculation.
Output validation checks component sums, thresholds, frame/epoch, source entries and ledger derivation. Corrupted external input rejects; internal invariant failure is 500. Polar full strength is typed unavailable; Lagna preserves independent evidence without inventing a full strength.
Kalavela borrows the supplied reader across every dependency and restores the previous reader on success/failure. It never closes that reader.
Sade Sati validates the range, finite positive advancing step (at most 5 days) and budget before ephemeris access, then counts every sample/refinement. SadeSatiBudgetError/HTTP422 sade_sati_budget_exceeded returns no partial complete result. The default 10,000 and maximum 20,000 evaluations are work limits, not a hidden year cap. Smaller steps consume more budget.

Measured DE441 searches from J2000 at 5-day sampling:12 years uses 1,034 evaluations,30 years 2,557,100 years 8,450. At 1-day sampling,12 years uses 4,504; the 3 matched windows differ by at most 0.00048828125 days at their boundaries, within the 0.001-day bisection tolerance. This supports the default budget; it does not prove detection of every substep retrograde excursion. The response names that sampling limitation explicitly. Both direct/facade exports and REST use the same bounded owner. Whole-REST operational budgeting remains VED-023.

Eleven-row completion ledger

Finding	Owning repair	Acceptance witnesses	State
VA-01	dasha.py, dasha_systems.py, _dasha_intervals.py	Full-parent rational trees, alternate horizon, J2000 HTTP Rahu/Mars	Closed
VA-02	shadbala_context.py, _shadbala_components.py	Four Kala components, observer/season/timezone/polar, real HTTP	Closed
VA-03	_shadbala_components.py, vedic_dignities.py	Two scales, Sun 90, Moola prerequisite, source divisions,144 friendship cases	Closed
VA-04	shadbala.py WarResolution	Independent quotient, graph cases, permutation/conservation, May2000 Lagna	Closed
VA-05	strength models/serializers/network	December2020 actual amounts, full/direct Lagna roundtrip, corrupted ledger rejection	Closed
VA-06	shadbala.py Ojayugma	Four Mercury D1/D9 parity combinations	Closed
VA-07	varga.py exact partition	65,880 rational cases, extended divisors, all 8 HTTP surfaces	Closed
VA-08	varga.py shared finite normalization	Tiny negative Shodashvarga, named/batch paths, circle consistency	Closed
VA-09	upagrahas.py reader scope	No ambient reader, distinct outer reader, nested success/error restoration	Closed
VA-10	sade_sati.py bounds and counter	Non-advancing step before access, preflight/runtime exhaustion, measured searches	Closed
VA-11	alternate Dasha historical fixture	Independent general-precession/frame reconciliation at original tolerance	Closed

Execution receipts

The original failing review artifacts are preserved and are not overwritten. The first repair checkpoint outcome is 6,276 passed, one explained skip, zero unresolved failures/errors, covering 173 distinct test files. This is an indexed reconciliation of the runs below, not a claim that the repository's entire test suite was executed. acceptance-summary.json retains each test's latest receipt and maps all five initial broad-run failures to passing reruns.

Run	Scope	Result
Broad Vedic regression	118 files; includes every original112-file selection, new repair tests and facade clock consumer	4,661 passed,5 failed initially; all 5 subsequently resolved
Release Hardening	All 62 files selected from the current workflow, including added source/Einstein guards	2,353 passed,1 existing skip,0 failures/errors
Strength/resource supplement	Real borrowed readers, geometry, source/graph tests, HTTP consumers	255 passed
Boundary/context completion	All alternate clocks, strict metadata, direct/HTTP source evidence	247 passed
Final reconciliation	23 affected files, all 5 broad-run failures, exact root/facade/Vedic exports, docstring guards	1,215 passed,0 failures/errors/skips
Final normalization follow-through	Strength source/legacy unit tests and retained real HTTP cases	375 passed,0 failures/errors/skips

The five initial failures were two missing curated Vedic exports, two Lagna roundtrip fixtures that omitted new context/ledger evidence, and one winter polar fixture that expected a full strength without sunrise. Exports now agree; roundtrips send complete receipts and retain exact equality. The polar test checks both equinox Porphyry fallback with full strength and winter typed unavailability. No arithmetic tolerance or validator was weakened.

Across these invocations all 835 DE441 resource receipts ran with zero resource skips/failures. Release tests also exercised2 supplemental small-body receipts covering3 manifests,940 shards and11,720 bodies. The single existing skip is test_void_of_course.py:526: no qualifying no-aspect example was found in its 30-day scan, so RULE-06 was not exercised. It is unrelated to the Vedic repairs.

Reproducible project-runtime commands (PowerShell, from C:/dev/moira):

$env:MOIRA_NO_DOWNLOAD = "1"
$env:MOIRA_TEST_MODE = "1"
$env:MOIRA_STRICT_KNOWN_ISSUES = "1"
.\.venv\Scripts\python.exe C:/dev/outputs/vedic-adversarial-repair-2026-10-09/run_review_tests.py
.\.venv\Scripts\python.exe C:/dev/outputs/vedic-adversarial-repair-2026-10-09/run_release_tests.py
.\.venv\Scripts\python.exe C:/dev/outputs/vedic-adversarial-repair-2026-10-09/run_reconciliation.py
.\.venv\Scripts\python.exe -m pytest tests/unit/test_shadbala_source_repairs.py tests/unit/test_shadbala.py tests/server/test_server_vedic_repair_acceptance.py -q
.\.venv\Scripts\python.exe C:/dev/outputs/vedic-adversarial-repair-2026-10-09/probe_boundaries.py
.\.venv\Scripts\python.exe C:/dev/outputs/vedic-adversarial-repair-2026-10-09/measure_search.py
.\.venv\Scripts\python.exe -m ruff check moira moira_server scripts --select F401,F821,F822,F823 --no-fix

The saved manifests and XML preserve the executed selections; the broad runner discovers any future matching tests, so its future collection count can increase. Source arithmetic fixtures use independently transcribed or rational expectations. Corpus checks are invariants, DE441 HTTP checks are integration/regression evidence, and ERFA precession is separate astronomical authority corroboration. No new native implementation or parity claim is made.

Runtime: project .venv, Python 3.14.3, discovered DE441 at C:/Users/nilad/.moira/kernels/de441.bsp; downloads disabled and strict empty KNOWN_ISSUES ledger. All 46 changed Python files pass 3.10 grammar; this is grammar compatibility, not a second 3.10 runtime execution. Source Ruff and whitespace checks pass. Exact exports and existing docstring governance pass. REST remains 498 operations; models and receipts are additive where described above.

Generated artifacts are synchronized in order: REST reference, Hellenistic inventory, Git wiki, version-pinned website publication manifest. All six release-facing documentation checks passed at the final local checkpoint; documentation-gates.log records all ten generator/check exit codes as zero. Neither the website application nor Urania source was changed. No native code, dependency or release version was changed. Version remains 6.9.9. Source publication remains uncommitted/unpushed; VED-011 is the next separately authorized package.

Strength uses normalized context positions as the authoritative receipt after same-frame comparison within 1e-9 degrees. This prevents a tolerated rounding difference from selecting opposite sides of a discrete Varga boundary.

Second adversarial pass repairs — 9 October 2026

The second review found five additional composition/admission defects after the eleven repairs above. The user authorized all five repairs. Original failing witnesses remain in C:/dev/outputs/vedic-adversarial-second-pass-2026-10-09; new receipts are separate in C:/dev/outputs/vedic-second-pass-repair-2026-10-09.

Finding	Repair and targeted evidence
VA2-01: UTC/UT1 weekday mismatch	Derived context declares civil UTC midnight and carries its UTC JD. Both negative-DUT1 midnight dates, a positive-DUT1 date just before midnight, and a timezone-equivalent instant pass all six HTTP Shadbala products, facade and dated Lagna. Contradictory UTC/lord evidence rejects.
VA2-02: tolerated angles cross Lagna boundaries	Context positions govern Lagna before all discrete decisions, with D1/D9 nextafter tests, score/receipt equality, preserved missing bodies and rejected incompatible positions. Bhava receives the same composition repair.
VA2-03: balanced component corruption	Context-backed Uchcha, Ojayugma, Drekkana and Drig are recomputed through existing formula owners; ranges/discrete awards are checked. All seven planets reject balanced HTTP mutations under the real May 2000 war context while unchanged receipts round-trip.
VA2-04: full interval/year provenance	Shared engine child validation checks full containment and exact clipping at unchanged tolerances. Recursive REST admission uses the owner validator with explicit partial-child-list mode; generated trees still require complete coverage. Both modes enforce uniform year labels/lengths. Valid birth/horizon clips and unknown legacy provenance remain admitted.
VA2-05: Sade Sati sign 12	Status and sampled Saturn signs share finite half-open angle normalization with windows; result vessels reject invalid sign indices. Tiny negatives, signed zero and equivalent turns agree across classification and windows; HTTP emits valid signs.

The original focused witnesses reproduced the defects before repair. The initial focused new-test selection passed 65 tests, with all 33 DE441 resource receipts run; the final reconciliation also passes the added partial- tree regression, for 66 new tests. Two new files are included in Release Hardening. Original probe replay rejects every previously accepted balanced component mutation and foreign full interval; midnight facade/HTTP calls now succeed. The unchanged independent generated corpus validates 250 trees / 166,118 nodes. The time owner's UT1/UTC inverse round-trips exactly across 10,000 deterministic modern samples. These are numerical consistency and integration checks, not additional classical attribution or independent ephemeris authority.

Explicit limits: the UTC weekday is a Moira convention, not a classical sunrise-day claim. Supplied contexts remain caller evidence. Without house geometry, the validator checks Kendradi membership and Dig range rather than reconstructing their exact values. Context-free legacy strengths retain structural validation. Unknown full Dasha endpoints remain unknown; parent full containment is checked only when both full intervals are declared. Sade Sati retains its previously documented sampled-search limitation.

All five findings are closed locally. The indexed final outcome is 6,342 distinct passes, zero unresolved failures/errors and one explained skip, across 175 test files. This reconciles the completed runs below; it does not claim a single clean full-repository invocation. The original failure outputs are preserved, with every failure mapped to a final passing receipt in acceptance-summary.json.

Run	Completed scope	Outcome
Broad Vedic	121 files, including every prior adversarial selection	4,792 passed; 3 compatibility regressions subsequently repaired
Release Hardening	All 64 files selected by the current workflow	2,471 passed; same Dasha diagnostic regression subsequently repaired; 1 existing skip
Final reconciliation	17 affected files, including every failing test/file	1,104 passed; no failures, errors or skips

The compatibility regressions were one changed Dasha diagnostic precedence and two existing partial-child-list REST admission contracts. Diagnostic precedence was restored in the engine. The shared alternate-tree validator now exposes explicit complete/partial coverage modes; both modes retain full-interval and year checks. Existing test assertions and numerical tolerances were not weakened. Across these three runs all 814 DE441 resource receipts ran, with zero resource skips/failures. Release coverage also ran two supplemental small-body receipts covering three manifests, 940 shards and 11,720 bodies. The one skip remains test_void_of_course.py:526, whose 30-day scan found no no-aspect example for RULE-06; it is unrelated to these repairs.

Commands use C:/dev/moira/.venv/Scripts/python.exe, from C:/dev/moira, with MOIRA_NO_DOWNLOAD=1, MOIRA_TEST_MODE=1, and MOIRA_STRICT_KNOWN_ISSUES=1:

.\.venv\Scripts\python.exe C:/dev/outputs/vedic-second-pass-repair-2026-10-09/run_review_tests.py
.\.venv\Scripts\python.exe C:/dev/outputs/vedic-second-pass-repair-2026-10-09/run_release_tests.py
.\.venv\Scripts\python.exe C:/dev/outputs/vedic-second-pass-repair-2026-10-09/run_reconciliation.py
.\.venv\Scripts\python.exe C:/dev/outputs/vedic-second-pass-repair-2026-10-09/verify_properties.py
.\.venv\Scripts\python.exe C:/dev/outputs/vedic-second-pass-repair-2026-10-09/probe_review.py
.\.venv\Scripts\python.exe C:/dev/outputs/vedic-second-pass-repair-2026-10-09/probe_lagna_boundary.py
.\.venv\Scripts\python.exe C:/dev/outputs/vedic-second-pass-repair-2026-10-09/probe_transport_followup.py
.\.venv\Scripts\python.exe C:/dev/outputs/vedic-second-pass-repair-2026-10-09/finalize.py
.\.venv\Scripts\python.exe -m ruff check moira moira_server scripts --select F401,F821,F822,F823 --no-fix
.\.venv\Scripts\python.exe C:/dev/outputs/vedic-second-pass-repair-2026-10-09/documentation_gates.py
git diff --check
git -C moira.wiki diff --check

The saved JSON manifests and JUnit XML record exact test selections. The 48 changed Python files pass Python 3.10 grammar checks. All six final release-facing document checks pass after canonical generation; Ruff and parent/nested-wiki whitespace checks pass. No source oracle, arithmetic tolerance or native parity expectation was changed. REST remains 498 operations with additive weekday context evidence; version remains 6.9.9.

Changed owners in this second repair pass:

moira/shadbala_context.py, _facade_vedic.py, muhurta_lagna_dated.py, moira_server/services/shadbala.py, and the Shadbala/Lagna transport models: UTC Vara basis, epoch evidence and consistent dated propagation.
moira/_shadbala_components.py, shadbala.py, muhurta_lagna.py: shared positional formulas, component receipt checks and canonical consumer angles.
moira/_dasha_intervals.py, dasha.py, dasha_systems.py, and alternate Dasha transport model/service: shared full provenance, year consistency, explicit partial-list mode and preserved diagnostic precedence.
moira/sade_sati.py: shared longitude normalization and valid sign vessels.
tests/unit/test_vedic_second_pass_repairs.py, tests/server/test_server_vedic_second_pass_repairs.py, and .github/workflows/release-hardening.yml: regression witnesses and CI selection.
Four owning standards, this validation record and the remaining-work register, followed by canonical documentation generators and their outputs.

The prior eleven-repair work remains intact. The native implementation, astronomical time substrate, dependencies, release identity, website/Urania applications, unrelated work and VED-011 were not changed by this pass. This is a local repair checkpoint; no commit, push, release or deployment is included. Python 3.10 validation is grammar-only; execution uses the project Python 3.14.3 runtime and discovered DE441, with downloads disabled and the strict empty known-issues ledger.