What TKS Ephemeris has been checked against — and, equally, where it has not. The engine's credibility rests on its outputs and their verification, not on any unverifiable claim. Everything below is either an internal control, a comparison to a published reference, or an explicit "no reference exists".
Honesty rule. Beyond the JPL DE441 coverage edge (−13,200 … +17,191) the engine produces reconstructions (Engine B). No independent reference exists there, so those results are labelled as such and not presented as observational-quality.
In-coverage (within JPL DE441)
Component
Checked against
Result
Status
Apparent positions (Sun–Saturn)
JPL DE441 state data (rebuilt by the engine's own SPK/DAF + IAU-2006 precession-nutation)
≈ DE441 to numerical precision (mas-level)
internal control
Precession / nutation
IAU 2006 (Fukushima–Williams) + IAU 2000B nutation
standard adopted model
standard
ΔT / time scales
Söhnel–Smalley 2016 (SMH2016) ΔT model; UT1 from UTC
model-based
model
Ayanāṃśa (True Chitrāpakṣa)
Spicā (α Vir, HIP 65474) anchored at nirayana 180°, via ICRS + proper motion + ecliptic-of-date
180.000° by construction (definitional)
definition
"mas-level" = milli-arcsecond at the engine-vs-DE comparison for bright bodies; the DE's own absolute accuracy vs reality is sub-mas near J2000, worsening toward the coverage ends.
Eclipse engine — NASA Five Millennium Canon (Espenak & Meeus)
Total cross-checked: 23,962 events (11,898 solar + 12,064 lunar) across the Canon's full span. "True mismatch" = the engine's class differs from the Canon after excluding grazing/hybrid classifications that sit on a type boundary; the excluded cases are listed in the harness output (eclipse_validate).
Dynamical controls (Engine B, N-body)
Control
Setup
Result
Status
Forward vs DE (within coverage)
N-body seeded at the edge, integrated forward, compared to DE441
agrees (e.g. Moon ~12.8 km, Mercury ~446 km over 100 yr)
internal
Backward integration accuracy
seed inside coverage, integrate backward, compare to DE441
after the dir-sign fix: Sun λ Δ = 0.000° (1 yr) and matched at 100 yr
internal
Fitted truncated lunar series
least-squares fit to DE441 (Delaunay terms), validated on held-out DE441
λ ≈ 5–7′ · β ≈ 0.2° (near edge); λ ≈ 7′ over ~2200 yr
held-out
Beyond the coverage edge (Engine B) — reconstructions
Item
Reference available?
What is claimed
Status
Planetary positions beyond edge
none
N-body (DOPRI5, EIH 1PN, J2–J6), ~arcmin over the near bands
no reference
Moon beyond edge
none
fitted lunar series; accuracy degrades with span (chaos + tides)
no reference
Long-term precession (pole/ayanāṃśa)
Vondrák et al. (2011, 2012), valid ±200 kyr
pole star / ecliptic pole from ERFA eraLtpequ / eraLtpecl coefficients
published model
These are dynamical reconstructions: internally consistent and physically grounded, but not equivalent to a modern observational ephemeris. They are labelled "reconstruction" wherever displayed.
Accuracy tiers (what we claim, where)
Span
Sun / planets
Moon
Basis
−13,200 … +17,191 (DE441)
≈ DE (mas → arcsec at the ends)
≈ DE (cm–m near J2000)
reference state data
Near-edge ±2,000 yr
~arcmin (est.)
λ 5–7′ / β 0.2° (measured)
Engine B + fitted series
5,000 … 15,000-yr spans
arcmin → degree
tens of arcmin → degrees
reconstruction (unvalidated)
≳ ±20,000 yr
degree-level
phase effectively lost
reconstruction (unvalidated)
What is deliberately NOT claimed
• No "exact historical dates" for eclipses/chronology beyond the reference span.
• No observational-quality Moon positions beyond the near-edge band.
• No independent validation beyond the DE441 span (no reference exists).
• No claim that any single star is the pole star for a fixed interval without stating the angular-separation threshold.