Sidereal general-precession reconciliation

Date: 6 October 2026. Scope: the PAC 1948 SE solar-ingress residual identified during VED-023 admission and the shared scalar that caused it.

Finding and correction

The initial lunar-month comparison found a solar ingress 64.4768 seconds later than PAC's printed 15 June 2026, 12:53 IST. The event solver correctly bracketed its own longitude boundary, but the shared ayanamsa precession scalar was the wrong astronomical quantity.

general_precession_in_longitude(jd_tt) used the Fukushima-Williams rotation angle psib, including its -0.041775 arcsecond frame-bias term. The caller adds this scalar to a stored J2000 mean ayanamsa reference. That operation requires accumulated general precession in longitude, p_A, which is zero at J2000. The rotation matrix needs the distinct FW angles.

The authority is the P03/IAU 2006 formulation, independently evaluated with SOFA-derived ERFA eraP06e, output pa. It explicitly distinguishes general precession from FW rotation angles and notes the published fifth-order sign misprint. In arcseconds, for TT Julian centuries T from J2000:

p_A = 5028.796195 T + 1.1054348 T² + 0.00007964 T³
      - 0.000023857 T⁴ - 0.0000000383 T⁵

Moira now evaluates this scalar in degrees. At the published June instant, the previous scalar exceeds p_A by 2.551926848 arcseconds. Before the correction, the sidereal Sun was 2.564347555 arcseconds short of 60°. The scalar error therefore accounts for the discrepancy to about 0.01242 arcseconds, consistent with the remaining fraction of a second in the comparison to a rounded printed minute.

This is a correction of the implemented astronomical object. No ayanamsa reference constant was fitted to an ingress. UT1-to-TT conversion, DE441 positions, apparent geocentric longitude, nutation, mean/true choices and event boundary ownership retain their existing contracts. The named Lahiri reference is still 23.857092317461543° at J2000 mean; the 1956 historical origin does not constitute an independently imposed exact apparent epoch constraint in this implementation.

Sources and finite comparison corpus

The institutional source is IMD/PAC's English 2026–27 archive, retrieved 6 October 2026. PDF RP 1948 SE Final.pdf SHA256:

a8816abe4fae7fc0f0e4349a3d91eef00cfc0044a5f050b1b3bfe826847f9eaa

Printed 158 / PDF 178 contains 13 Sun ingress rows. Every row was transcribed and checked against the rendered page, including the following year's second Mesha ingress. Printed vi–vii distinguishes variable ayanamsa nirayana ingresses from the fixed-23°15′ saura calendar; only the former is the comparison object. Printed vi states the geocentric event times can be converted from IST to UT by subtracting 5h30m. Tests use timezone-aware IST instants and canonical UTC-to-UT1 conversion rather than passing TT to a UT1 evaluator.

The versioned transcription records the publisher, checksum, page, precision, time conversion, event semantics and original 60-second acceptance limit. No threshold was relaxed after the failed comparison. The table is printed to minutes and does not establish subsecond external accuracy.

Additional angular corroboration: at 00:00 UTC on 22 May and 22 June 2026, corrected true Lahiri values are 24°13′39.063″ and 24°13′44.540″. These round to PAC's month-header values 24°13′39″ and 24°13′45″, printed 18 and 26 / PDF 38 and 46. Their headings specify a date rather than a precise epoch; this corroboration is not a separate subsecond or subarcsecond oracle.

Measured event results

The following reproducible measurements use installed DE441, apparent geocentric ecliptic-of-date Sun, Lahiri true mode, UT1 event JDs, and the default 0.1-second lunar-month bracket. Residuals are Moira minus the published instant, in seconds. Sankranti is a separate existing public event assembly with its own angular solver, sharing the corrected scalar.

Published IST	Lunar month	Sankranti
2026-04-14 09:33	-21.176	-21.259
2026-05-15 06:22	+1.978	+1.978
2026-06-15 12:53	+0.330	+0.330
2026-07-16 23:39	+22.577	+22.577
2026-08-17 07:59	-14.420	-14.502
2026-09-17 07:53	-2.472	-2.472
2026-10-17 19:52	-18.210	-18.292
2026-11-16 19:43	+12.524	+12.524
2026-12-16 10:25	+2.307	+2.307
2027-01-14 21:10	+25.296	+25.214
2027-02-13 10:09	+4.697	+4.614
2027-03-15 07:00	+5.603	+5.603
2027-04-14 15:28	+22.824	+22.742

All 13 pass the original one-minute institutional comparison. The maximum absolute residual is 25.296 seconds. The solvers differ by at most 0.0824 seconds, within their combined bracket/resolution allowance of 0.2 seconds; this latter check is internal consistency, not independent astronomical authority. Entered-side endpoints from different scan brackets can differ within the configured tolerance.

Verification and consequences

Independent scalar checks compare against installed PyERFA 2.0.1.5 / ERFA 2.0.1 at 16 TT epochs, including J2000, the 1956 origin, the June ingress, BCE/modern samples and both ends of the ±50-century polynomial interval. All 16 failed before the fix and pass after it at the existing 0.001-arcsecond threshold. This proves the quantity independently of PAC solar timing or a secondary astrology engine. ERFA remains a development oracle; the runtime polynomial requires no new dependency.

Lunar-month integration checks exercise all 13 institutional rows through both existing event paths, retaining exact month identities, conjunction comparisons, root-side witnesses and explicit unavailable exceptional Purnimanta results. REST checks verify the formerly failing June ingress through the typed HTTP response as well.

Reproduction, with installed resources and the project's Python 3.14.3 runtime (Moira 6.9.9):

$env:MOIRA_NO_DOWNLOAD = "1"
$env:MOIRA_STRICT_KNOWN_ISSUES = "1"
.\.venv\Scripts\python.exe -m pytest tests/integration/test_erfa_validation.py -k general_precession -q
.\.venv\Scripts\python.exe scripts/reconcile_pac_solar_ingresses.py

The offline measurement script emits the source receipt, resolved installed kernel, policy, UT1 instants, complete event brackets and residuals as JSON; it exits unsuccessfully if the authority or internal-consistency checks fail. The installed resource for these measurements was C:\Users\nilad\.moira\kernels\de441.bsp, read by the required native SPK backend.

The correction affects every polynomial ayanamsa path, including explicit user-defined anchors and polynomial star fallbacks. Live true-star anchors remain separately evaluated. Derived sidereal engine and REST values receive the same correction through their existing calls. Historical changes can be larger than the modern 2.55-arcsecond example; this is not a tolerance-preserving refit of older outputs. The coordinate matrix and its admitted native counterpart use their own FW angles and receive no numerical modification.

The scalar polynomial continues its existing extrapolation outside ±50 centuries, which is a disclosed model limitation. The wider Vondrak matrix validity does not certify this scalar. This reconciliation closes the named PAC solar-ingress discrepancy; exceptional Purnimanta/Kshaya regional mapping and source-owned festival rules remain open in VED-023/024.

The selected broader command was:

.\.venv\Scripts\python.exe -m pytest tests/integration/test_lunar_month.py tests/integration/test_erfa_validation.py tests/integration/test_sidereal_external_reference.py tests/unit/test_sidereal.py tests/unit/test_sidereal_nakshatra_boundaries.py tests/unit/test_physics_layer.py tests/unit/test_panchanga.py tests/unit/test_daily_panchanga.py tests/integration/test_daily_panchanga.py tests/unit/test_lunar_month.py tests/server/test_server_lunar_month.py tests/server/test_server_daily_panchanga.py tests/server/test_server_panchanga_routes.py tests/server/test_server_panchanga_service.py tests/server/test_server_sidereal_routes.py tests/server/test_server_sidereal_context.py tests/test_native_sidereal_phase1.py -q --tb=short

That first run covered 901 cases: 900 passed and one failed. The failure was test_pbt_true_node_common_frame_intersection, whose witness applied apply_frame_bias before the already bias-inclusive precession matrix. An isolated replay at UT1 JD 2400000 using the original HEAD precession source also failed. Removing that extra bias application repairs the witness; both strict 1e-12 normalized plane-intersection assertions remain. No node runtime, coordinate matrix or native implementation was changed. The original command was not retrospectively classified as a clean run.

The affected/new follow-up commands were:

.\.venv\Scripts\python.exe -m pytest tests/unit/test_physics_layer.py::test_pbt_true_node_common_frame_intersection -q --tb=short --junitxml=tmp/pac-node-witness.xml
.\.venv\Scripts\python.exe -m pytest tests/server/test_server_lunar_month.py::test_http_mithuna_ingress_passes_original_pac_minute_gate tests/unit/test_vedic_facade.py tests/unit/test_vedic_surface_completeness.py -q --tb=short --junitxml=tmp/pac-reconciliation-followup.xml

These passed 1/1 and 22/22, with zero failures or skips. Across the broader run and corrected/new follow-ups, 923 distinct selected cases have passed their latest execution. The 16 scalar checks and an additional 86-case nakshatra-boundary rerun are already included in that coverage, not added again to its count. The broader harness admitted 78 planetary-resource receipts, all DE441, with zero skips or resource failures; the follow-ups reported 3 and 4 receipts respectively. External network access remained disabled; the server cases used explicitly marked loopback.

Scoped Ruff checks passed for the measurement script and lunar-month integration/server files. A wider lint check also reported the pre-existing E741 single-letter variable at test_erfa_validation.py:211, outside the changed lines; it was left untouched and is not reported as a clean lint run. git diff --check passed. tests/KNOWN_ISSUES.yml stayed empty and strict expiry checking remained enabled. No whole-suite, universal calendar, or historical regional-rule certification is claimed.