Eclipse Catalog Comparison

This report summarizes the repository's selected eclipse catalog comparisons. The NASA fixtures identify the Five Millennium solar and lunar catalog source URLs in tests/fixtures/eclipse_nasa_reference.json and the published Besselian-element rows in tests/fixtures/nasa_solar_besselian_reference.json, plus the paired 2015 polar path/Besselian product in tests/fixtures/nasa_solar_polar_path_reference.json, and the 2003/2006 partial-visibility Table 2 products in tests/fixtures/nasa_solar_penumbral_footprint_reference.json. Cached Swiss rows in tests/fixtures/swe_t.exp are a separate, secondary cross-engine corpus.

This is a readable evidence summary, not a replacement for the executable tests and not a claim of full-catalog, exact-model, or atlas-grade path validation. The bounded Besselian per-field evidence admitted below applies only to its named events, fields, epochs, models, and tolerances. The polar path evidence is likewise limited to the named 2015 product and its explicit geographic gates. Partial-visibility-footprint evidence is limited to the published 2003/2006 contacts and named boundary anchors. Both rows are total solar eclipses whose penumbral visibility footprints exercise the two admitted topologies; this is not an external footprint oracle for a globally partial event or a dense track oracle.

Classification at catalog maxima

The following rows record native classification at the NASA catalog's published maximum. They establish selected at-instant classification only; they do not validate Moira's searched maximum or every field of the event. The displayed NASA timestamps are catalog TD; the Moira column is evaluated at the fixture's corresponding derived UT1 Julian Day.

Solar

Lunar

tests/integration/test_eclipse_nasa_reference.py contains a broader selected fixture slice than the representative rows displayed above.

Search timing on a common TT scale

Raw UT1 differences cannot rank the native and NASA products because their UT1-to-TT mappings are different. For each row below:

• NASA reference TT is catalog UT1 plus the catalog's published Delta T; and
• Moira TT is searched event UT1 transformed with Moira's active, kernel-coherent Delta-T policy.

tests/integration/test_eclipse_nasa_reference.py computes the residuals at runtime and enforces the following policy. Exact residual snapshots are not frozen in prose because they change when an admitted search or time-policy implementation changes.

The ancient 360-second envelope is cross-authority regression evidence for the combined search and model-basis difference. It is not an accuracy bound or an uncertainty estimate for ancient Earth rotation. The post-2150 60-second TT gate checks search geometry on a common dynamical scale; it does not validate Moira's future UT1 scenario as a forecast.

The separate ten-row modern lunar method comparison admits nasa_shadow_axis_apparent_sun_moon as the NASA-compatibility default. Its greatest-eclipse timing is bounded at 10 s, and signed gamma is bounded at 2e-4 Earth radii. Both older geometric and retarded method identifiers remain in the comparison as explicit experiments rather than hidden fallbacks.

Individual lunar contact-instant comparison

tests/integration/test_eclipse_lunar_contacts_nasa_reference.py now binds all 14 applicable P1, U1, U2, U3, U4, and P4 instants across the existing 2023 penumbral, 2024 partial, 2025 total, and limiting 2027 penumbral cases. The dedicated tests/fixtures/nasa_lunar_contact_instants_reference.json fixture records the official NASA/GSFC detailed figure URL and SHA-256 digest for each event, its published UT strings, adopted Delta T, printed model, and shadow rule. Greatest eclipse is retained only as the search seed and timeline reference; it is not classified as a seventh contact.

The figure products declare VSOP87/ELP2000-85 and CdT (Danjon), and print contacts to one second. That lineage is deliberately separate from the Five Millennium century catalog rows used by the phase-duration comparison. Source TT is formed as the published figure UT plus that figure's Delta T. Native contact UT1 crosses through the reader-bound DE441 clock before comparison; the NASA-compatibility result is compared directly in its stored TT scale.

The initial comparison exposed a dominant compatibility defect rather than a large irreducible ephemeris residual: the previous default omitted the apparent reduction used by the detailed figure product. The repaired default derives one Earth reception state, applies reception light-time and then annual aberration to both the Sun and Moon, and omits gravitational deflection, topocentric parallax, and atmospheric refraction. Executable intermediate evidence independently compares the resulting 2025 apparent geocentric Sun and Moon right ascensions and declinations with the values printed on that NASA figure. The bounded remainder, rather than the former full discrepancy, now contains DE441/LE441 versus VSOP87/ELP2000-85, constants, and source-algorithm differences.

The ordinary per-instant cross-model ceilings are 120 s for the native path and 10 s for the NASA-compatibility path. The near-tangent 2027 event has separate 240 s native and 30 s compatibility ceilings and is admitted only as robustness evidence. NASA-compatible greatest eclipse is bounded at 10 s. The existing P4-P1 duration comparison remains independently enforced, so the two endpoint bounds cannot conceal an unbounded phase duration. These are cross-model regression envelopes, not source precision, uncertainties, UTC claims, or exact NASA-model parity.

Runtime Besselian per-field comparison

tests/integration/test_eclipse_besselian_nasa_reference.py compares the instantaneous EclipseCalculator.solar_besselian_elements(jd_ut1) result with four published NASA/GSFC Besselian products: the 2000 partial, 2024 total, 2031 hybrid, and 2032 annular solar eclipses. Each polynomial is evaluated at five TT/TDT epochs spanning t0 - 3 h through t0 + 3 h, for 20 element sets in total.

Every admitted numerical field is checked. The exact absolute envelopes are 1.0e-4 Earth equatorial radii for x, y, l1, and l2; 0.003 degrees for d; 0.007 degrees circular for mu; and 3.0e-6 for each dimensionless cone tangent. The test pins this tolerance map independently of the fixture so a data-only edit cannot silently weaken the gate.

This is primary-authority cross-model evidence, not exact-model parity. NASA's published elements use VSOP87/ELP2000-82 and the stated k1/k2 lunar-radius convention. Moira independently evaluates a content-identified DE441/LE441 Earth-reception shadow axis with physical mean-limb radii; NASA coefficients are never used by the runtime method.

Polar central-path comparison

tests/integration/test_eclipse_polar_path_nasa_reference.py compares the DE441-native path with the official NASA/GSFC 2015-03-20 total-eclipse product. The NASA fixture keeps its paired path and Besselian pages under the same declared model: JPL DE405, Delta T = 67.6 s, WGS 84 coordinates at the published 120-second cadence, published k1/k2 lunar-radius constants, and mean-limb center-of-mass geometry.

The independent gates are 1 s for searched greatest time, 3 km for the greatest point, five named central-line rows, and both axis/ellipsoid tangency endpoints, 3 km for greatest path width, 3 s for local central duration, 0.005 for greatest magnitude, and 3 km of physical cone clearance at each available published north/south limit. Five TT epochs also exercise the paired Besselian polynomial under the existing per-field envelopes.

This is one authoritative polar central-path fixture, not an atlas-wide claim. It does not assert per-row path-width parity away from greatest and does not reinterpret NASA's missing north limits at 10:16 and 10:18. At those epochs the ordinary closed-footprint solver is required to fail explicitly. A path whose other boundary closes on the terminator remains a distinct, unsolved public central-shadow width product.

Partial-visibility footprint comparison

tests/integration/test_eclipse_footprint_nasa_reference.py compares the first-class DE441-native footprint with two official NASA/GSFC Table 2 products: the 2003-11-23 one-limit connected case and the 2006-03-29 two-limit/two-loop case. Both products describe the penumbral footprint of a total solar eclipse, not a globally partial eclipse. The NASA tables declare DE200/LE200, k1 = 0.2725076, TDT, WGS 84 coordinates, and true east-positive longitude. Moira retains content-identified DE441/LE441 Earth-reception states, exact common-tangent cone geometry, and its physical spherical mean-limb radii.

The executable comparison covers P1/P4, optional P2/P3, and the published north/south boundary anchors on a common TT scale. Named boundary coordinates are matched to the nearest lawful footprint point. The independently pinned ceilings are 5 s for time and 40 km for coordinates. These are cross-model regression bounds covering the declared ephemeris and lunar-radius convention difference; they are not physical uncertainties.

The fixture and tests also enforce one_limit_connected versus two_limit_two_loop, contact ordering, boundary-family presence, and WGS 84 coordinate invariants. Separate DE441 regressions enforce folded connected components as strictly time-ordered segments sharing a refined fold endpoint, including stable graph identity at public sample counts 9, 99, 181, 257, and 721. Each admitted penumbral kind is one component with component_id=0; its fold segments use contiguous segment_id values and its two horizon incidences belong to the sunrise/sunset graph. A globally partial event's greatest point is covered separately by physical/geometric invariants, not by these external Table 2 rows. NASA's maps establish the topology, but no dense numerical penumbral-track table is published for these products. The evidence therefore does not claim dense-track, full-atlas, atmospheric-refraction, observer-elevation, magnitude-contour, or local apparent-circumstance parity.

Native and NASA-compatible products

The native model is Moira's DE441-backed physical geometry in TT. The nasa_compat path is a translation for catalog comparison and does not replace native truth. Its default method is nasa_shadow_axis_apparent_sun_moon; compatibility canon_method, source_model, and numerical values intentionally reflect that repaired policy. Existing facade signatures and REST paths and schemas are unchanged.

For the focused ancient lunar total case, the native result is transformed with Moira's active Delta-T policy and the nasa_compat result is transformed through ut_to_tt_nasa_canon(). Both are required to remain inside the same 360-second cross-authority regression envelope. Their exact residuals are runtime evidence rather than frozen documentation, and their raw UT1 residuals are intentionally not used to declare one model more accurate because that would compare unlike Earth-rotation policies.

tests/integration/test_eclipse_ancient_residual_diagnosis.py separately checks that the Delta-T branch change is a larger contributor than the sampled retarded/geometric Moon change and that aligned objectives identify the same physical TT minimum to within one second. It does not establish historical event time to one-second accuracy.

Swiss corroboration boundary

tests/integration/test_eclipse_external_reference.py compares selected classification and searched maxima with cached Swiss rows. This is secondary cross-engine corroboration, not primary authority validation.

tests/integration/test_eclipse_occultation_where_reference.py checks one cached Swiss solar where row at a one-degree latitude/longitude tolerance. That row does not validate magnitude, path width, central duration, or the sampled track against NASA Besselian data.

Interpretation

• Selected NASA catalog maxima support classification evidence across ancient, modern, and future rows.
• Search timing must be compared in TT while retaining each product's declared Delta-T basis.
• Native and compatibility modes are distinct products and must remain explicitly labeled at public boundaries.
• The separate runtime Besselian test establishes bounded per-field cross-model evidence for four modern event classes; it does not establish exact NASA model parity or atlas-wide coverage.
• The paired 2015 fixture establishes bounded DE441-versus-DE405 evidence for one polar central line, its ellipsoid tangencies, greatest width, local durations, and available published limit coordinates as cone-boundary points; it does not establish one-limit or full-atlas path-width parity.
• The 2003/2006 footprint fixture establishes bounded DE441-versus-DE200/LE200 contact and sparse boundary-anchor evidence for both admitted topology classes of total-eclipse penumbral footprints; it does not establish a globally partial-event footprint oracle or dense-track parity.
• Cached Swiss comparisons corroborate a bounded slice but do not govern Moira's native model.

2026-07 record note

This record retains the earlier removal of stale raw-UT rankings and unsupported small-angle claims. It now also records the independently derived runtime Besselian surface, its executable pinned per-field NASA/GSFC comparison, and the bounded 2015 polar central-path authority fixture. The 2026-07-18 update adds the first-class partial-visibility footprint and its bounded 2003/2006 NASA/GSFC Table 2 anchor evidence. Exact test-command outcomes belong in the change's completion receipt; this document states only the durable evidence scope and acceptance envelopes.