Historical Solar Eclipse Validation — Framework Scene-Graph vs the Documented Record

This page is the empirical confirmation that the model’s predictions actually hold against the solar eclipse record. For the parallel lunar eclipse validation (which has higher temporal resolution because lunar eclipses don’t depend on geographic localization), see Lunar Eclipse Validation. For the closed-form ΔT formula itself, see Timekeeping & Delta-T; for the Moon polynomial’s implementation and parallax limit at modern epochs, see Sun, Moon & Planets; for the derivation of the lunar theory’s constants themselves, see The Derived Moon (DLT-1); for the broader validation tradition (Wells 1963, Cheng 2016, etc.), see Supporting Evidence.

1. Thesis

Pure-tidal Farhat-based Moon orbital evolution + an L1-orbital-coupled α(t) GIA correction from satellite gravimetry produces a ΔT formula that aligns with the documented solar-eclipse record across at least 2,700 years, using the model’s own predicted umbra track rather than fitting to the eclipse dataset itself.

The conventional interpretation of the long-term ΔT record holds that pure tidal physics alone cannot account for the observed eclipse data — that an additional non-tidal Earth-rotation component is required. Stephenson’s empirical polynomial reproduces the observations and is consistent with this non-tidal component being present. The conventional Munk-MacDonald magnitude for this component is ~5-6 ms/century, attributed to glacial isostatic adjustment plus core-mantle coupling.

The model takes a more careful position. Pure tidal physics from the Farhat 2022 deep-time evolution model, applied via angular momentum conservation, aligns with the solar-eclipse record on the 26-event eclipse alignment audit (§2 Layer 3). The model adds a single physically-derived non-tidal channel — an L1-orbital-coupled α(t) GIA correction anchored on independent satellite gravimetry (Cox & Chao 2002 dα/dt) and tied to the L1 orbital layer of the Climate Formula — that does apply, but with much smaller magnitude (-0.35 ms/century at J2000, from the Cox & Chao 2002 satellite anchor with the Peltier ICE-6G factor-2.0 J₂→α conversion) than the conventional Munk-MacDonald assumption. This is the one physical mechanism that couples climate, planetary gravity, and Earth rotation through ice-mass redistribution — see §4 for the derivation.

The higher-resolution lunar timing test additionally identifies a smaller fractional non-tidal channel ~0.5 ms/century (window-average) in the residual beyond α(t) — about 2× Cox-Chao’s satellite baseline; ~10% of full Munk-MacDonald. This channel is the era-localized signature of the documented millennial core–mantle LOD fluctuation (Stephenson–Morrison–Hohenkerk’s ~1,500-yr oscillation, mechanism per Dumberry & Bloxham 2006), modelled as the framework’s Core-mantle swing (Resonator driver — the 4th dLOD/dt channel, fitted jointly with the 4-flag stack). The full Munk-MacDonald-scale assumption is rejected; a uniform secular version of the fractional channel is independently disfavored at ~4σ by the solar-drift bound (see Lunar Eclipse Validation §9 and doc 104 ).

2. Three independent validations

The validation runs in three layers, each independently informative.

Layer 1 — Moon polynomial vs NASA in TT space (foundation)

The simulation’s Moon position uses Meeus Ch. 47 (60 longitude + 60 latitude perturbation terms) on top of a 5-layer geometric precession hierarchy. For 11 canonical eclipses from NASA’s Five Millennium Catalog of Solar Eclipses spanning -524 to 985 CE, the model’s computed time of Moon-Sun conjunction was compared to NASA’s published Terrestrial Dynamical Time (TD) of greatest eclipse.

The comparison is ΔT-independent: in TT-space, the astronomical event is fixed regardless of which ΔT either side assumes. Any residual is purely a Moon polynomial accuracy question.

This is the expected Meeus Ch. 47 polynomial residual at these timescales (≈ 0.13° in Moon ecliptic longitude at year 980, the worst case). The polynomial is sound at every epoch tested.

Layer 2 — Per-event same-day conjunction check

For 26 documented historical solar eclipses with known calendar date and observation site, the model’s Moon-Sun ecliptic longitude separation Δλ was evaluated at noon UT on the documented date. The question: is there a conjunction within ±12 hours of that noon (same calendar day)?

Result: 26/26 same-day matches. Every documented eclipse date corresponds to a conjunction in the model. The visibility-at-site question — does the model’s eclipse path actually reach the observation site? — is the subject of Layer 3 below.

Layer 3 — Scene-graph alignment audit (the headline)

For each event, the framework’s own simulation produces the model UT of greatest eclipse and the umbra-track geometry at that UT. The audit scans ±4 hours around the documented preset UT and measures the minimum umbra↔site distance across the scan (BestGap). Each event is then assigned one of five verdicts based on that BestGap and whether the framework’s own predicted UT agrees with the documented UT:

Tallies:

• 14/26 tight-alignment (confirmed + off-peak): the framework umbra centerline runs through the observation site.
• 20/26 umbra reaches the site region within the ±4h scan window (the four scan-reach verdicts combined).
• 6/26 pure geographic misses: framework places umbra >1,000 km from the documented site at every scanned moment. These are the four late-tenth-century Cairo observations (977 Dec 13, 978 Jun 8, 979 May 28, 985 Jul 20) plus the -647 Babylon early-diary partial — site-association artifacts, not path errors. The model’s central paths for the Cairo-dated eclipses match NASA’s greatest-eclipse positions to 65–290 km, and the records themselves are partial-phase observations: Ibn Yunus’s first-hand 977 report is a genuine 0.60-magnitude partial at Cairo, while the second-hand Said al-Andalusi entries are adjudicated as duplicates (978, 979 → the 977 Dec 13 eclipse) or misdated (985 → 982 Sep 20). Full adjudication: doc 107 .

Modern eclipses (1900+) all show BestGap ≤ 130 km. Deep antiquity, under the framework-native umbra chain (the derived Sun and the derived Moon) and the deep-branch lunar completions, has the entire first-hand ancient corpus at off-peak class — -708 Lu 9 km, -556 Nabonidus 104 km, -762 Nineveh 174 km, -584 Thales 211 km, -135 Babylon 366 km (§6.1) — with -309 Antigonus (790 km) and -430 Athens (853 km) regional. The geographic class is an umbra-centerline distance metric, not eclipse visibility: at high γ the shadow strikes the tilted Earth obliquely and the penumbral footprint spans thousands of km, so those sites can still observe a deep partial.

3. The ΔT gap — pre- and post-α(t) framing

Pre-α(t) pattern (the original finding)

When this validation was first performed (with pure-tidal-only physics, no GIA correction), the model showed a constant linear ΔT excess over Stephenson’s empirical fit, scaling at ~2 s/yr into the past:

The linear-in-time slope of ~1.96 s/yr corresponded geometrically to a constant Length-of-Day difference of ~5–6 ms between pure-tidal-only and Stephenson — the order of magnitude of canonical non-tidal Earth-rotation estimates (the Munk-MacDonald mechanism).

At the time, two readings of this gap were possible: either a real non-tidal Earth-rotation component (Munk-MacDonald-scale), or a phenomenological feature of Stephenson’s fit. The solar-eclipse visibility test couldn’t distinguish them.

Post-α(t) pattern (the resolved finding)

The higher-resolution lunar-timing test (doc 102 , see also Lunar Eclipse Validation) resolves the tension. Once the model includes an L1-orbital-coupled α(t) GIA correction from independent satellite gravimetry, the constant linear excess is absorbed in the ancient era — ancient-era ΔTs now agree with Stephenson to within ~50 s, and the residual signal is a bump-shaped structure in the medieval window (~1,000 s peak in the 840–1020 CE window; exact peak year is reference-conditional). The medieval bump is decomposed on the lunar validation page into the framework-native 4-flag 8H-lattice stack (Bond 8H/1830 + Hallstatt 8H/1104 + Jose5 8H/2989 + Jose4 8H/3749) plus a fractional non-tidal secular drift plus observation noise.

The reading that survives: the non-tidal contribution IS real, and decomposes into a dominant GIA-scale channel (-0.35 ms/century at J2000, captured by α(t) via the Cox & Chao 2002 satellite anchor) plus a smaller fractional non-tidal secular rate (~0.5 ms/century — about 2× Cox-Chao — detected in the lunar-timing residual but not modelled) — not at the full Munk-MacDonald magnitude (~5-6 ms/century) the pre-α(t) framing originally suggested. The lunar timing test, which has ΔT resolution of minutes rather than hours, distinguishes them.

4. What the validation establishes

What the model can claim with confidence:

1. The Moon polynomial used in the simulation is validated against NASA’s JPL reference at ±15 min back to 2,500 years before J2000 (n = 11 events, -524 to 985 CE).
2. The model’s pure-tidal + α(t) GIA ΔT formula places the framework umbra within the ±4-hour scan window of the observation site for 20/26 documented events spanning -762 BCE to 2026 CE, with 14/26 tight-alignment (confirmed + off-peak) — see §2 Layer 3 for the full verdict breakdown.
3. The full Munk-MacDonald-scale (~5-6 ms/cy) non-tidal Earth-rotation postulate is rejected by the historical eclipse record. A dominant GIA-scale (-0.35 ms/cy, Cox & Chao 2002 satellite anchor with Peltier ICE-6G factor-2.0) channel is included via the α(t) correction, and a smaller fractional non-tidal secular rate ~0.5 ms/century (about 2× Cox-Chao) is detected in the residual by the higher-resolution lunar timing test (see §6 for the three-component decomposition).
4. The model’s deep-time grounding is independent of the eclipse record — anchored to Wells 1963 (Devonian coral growth bands), Wu et al. 2024 (650-Myr cyclostratigraphy), modern Lunar Laser Ranging, and Cox & Chao 2002 satellite gravimetry. None of this evidence is circular with the historical eclipses.
5. The stack (4 flags + Core-mantle swing) shows an open, unvalidated correspondence with an independent paleoclimate signal (Bond 2001 IRD) — see the intro-callout above and Timekeeping § The Bond 2001 IRD comparison.

What the model is not claiming:

• That Stephenson’s curve is wrong. It fits the eclipses well too.
• That all non-tidal mechanisms are absent. A dominant GIA-scale (-0.35 ms/cy at J2000, Cox & Chao 2002 satellite anchor) channel is included via the α(t) GIA correction, and a smaller fractional non-tidal (~0.5 ms/cy, about 2× Cox-Chao) is detected in the lunar timing test residual but not currently modelled. The rejected claim is specifically the full Munk-MacDonald magnitude (~5-6 ms/cy).
• That solar-eclipse data alone settles every question. Lunar-eclipse timing — which doesn’t depend on geographic localization and has minute-scale ΔT resolution — is the stronger constraint and is covered in the companion Lunar Eclipse Validation page.

5. Limits

Caveats the reader should keep in mind:

1. n = 26 historical solar eclipses is small on absolute terms. The preset list spans -762 BCE (Bur-Sagale) through 2026 CE (Burgos total) and includes both ancient primary-source events and modern reference eclipses. Statistical power for discriminating models with sub-100 s ΔT differences remains limited.
2. Geographic localization of ancient eclipse paths has irreducible uncertainty. The 4,500 km umbra reach and 7,500 km penumbra reach are approximations based on the model’s sub-solar-point distance.
3. Ancient observation sites can sit far from the central path (e.g., 977 Dec 13: Cairo is ~1,600 km north of the central path at closest approach, regardless of ΔT — the record is a partial-phase observation). For these events the test only tells us about penumbra visibility, not totality.
4. The ±4-hour scan-window audit cannot detect ΔT differences much smaller than ~50 s at the per-event level, because ancient site localization is coarse relative to sub-100-s ΔT precision. For finer ΔT discrimination, lunar-eclipse timing-at-site is the stronger constraint.

6. Umbra-centerline tightening — the first-hand ancient corpus at conventional documented dates

An earlier diagnostic effort identified an ad-hoc visualization-layer overlay in the simulation’s Earth-rotation chain that had been adding a ΔT × 2π/86400 rotation correction on top of standard GMST(UT1) Earth rotation. The overlay made the visible umbra disc render at Stephenson 2016’s documented locations but introduced a systematic offset relative to the model’s underlying pure-tidal physics. With the overlay removed, the physics itself produces the centerline distances above.

The -135 Babylonian eclipse — case study

The 15 April 136 BCE (= -135 astronomical) Babylonian eclipse, recorded in the Babylonian astronomical diaries BM 45745 and LBAT 1285, is one of the most scholarly-secure attributions in the historical eclipse corpus (four-planet astronomical fingerprint, double-dated Arsacid/Seleucid eras, re-confirmed by Stephenson & Steele 2006) — and the framework agrees with it at the off-peak class.

This is not a ΔT-signal event. The framework’s own greatest-eclipse UT (06:01) sits within 9 minutes of the documented UT (06:14), and the umbra track reaches BestGap = 366 km from Babylon at scan offset BestΔUT = -0h60. The framework’s local-circumstance computation reads deep-totality-boundary magnitude at the site on the traditional date, and a ΔT-free identification cascade over the surrounding decades selects that date uniquely, with the implied ΔT inside Stephenson’s published totality window — the framework’s reduction and Stephenson’s overlap.

This accuracy class is carried by the deep-time lunar argument completions (the derived secular-ë and dynamical-precession carriers) and by the umbra chain’s framework-native Sun and Moon — the derived apparent solar longitude assembled with zero fitted solar constants, and the full derived lunar series. ΔT is exonerated (the framework’s value falls between NASA’s and Stephenson’s; ~2–3 km per 100 s), the α(t) constants are not load-bearing (tuning them across the full literature range moves the umbra by only tens of km), and the Meeus Ch. 47 Moon polynomial is exonerated (all modern lunar theories converge within 0.001° at year -135).

Full component-level decomposition and the identification-cascade evidence are in the simulation repo’s doc 103 .

7. Reproducing the validation

The complete validation suite is available in the simulation as developer-mode console tests at Console Tests (F12) > Historical Eclipses & ΔT (10 buttons): the NASA catalog cross-check (Layer 1 above), the per-event same-day check (Layer 2), the 26-event eclipse alignment audit (Layer 3), plus 7 supporting diagnostics. All 26 documented eclipses are exposed as planetStats nav-buttons under Moon → Historical Solar Eclipses (validation), where the user can step through each event and visually verify Moon-Sun alignment in the 3D scene.

A separate group of buttons under Console Tests (F12) > Lunar Eclipses & Validation runs the higher-resolution lunar-timing track — 20 buttons in 5 subgroups (foundation, predictive finders, NASA Canon cross-check, primary observation tests, residual investigation) covering 267 primary-source lunar observations + 89 primary-source solar observations from Stephenson 2016. See the Lunar Eclipse Validation page for the results.

Full reproducibility notes, methodology, and underlying numerical inputs are in the simulation repo’s doc 102  (α(t) GIA physics + 267-event lunar timing test + 89-event solar cross-validation) and doc 103  (-135 Babylonian eclipse case study).

Continue to Lunar Eclipse Validation for the higher-resolution test on 267 primary-source lunar observations, or Predictions for the model’s testable predictions across near-term, medium-term, and deep-time horizons.