KILLER_TEST_RESULTS - TheDaniel166/moira GitHub Wiki

Killer Test Results

Status: verified in the project .venv

Purpose

This report records the measured results for the current ephemeris stress tests used to probe Moira's edge-case correctness.

These are not feature demonstrations. They are adversarial validation checks for:

  • DE441 segment-boundary continuity
  • Delta-T robustness
  • extreme historical and far-future in-coverage epochs
  • clean out-of-coverage failure behavior
  • conjunction solver precision
  • long-range search stability

Proof files

  • tests/unit/test_de441_segment_boundaries.py
  • tests/unit/test_ephemeris_stress_proofs.py
  • tests/unit/test_topocentric_multi_path_consistency.py
  • tests/unit/test_ephemeris_breadth_gauntlet.py
  • tests/unit/test_polar_house_breadth_gauntlet.py
  • tests/unit/test_polar_chart_public_gauntlet.py
  • tests/integration/test_topocentric_multi_path_horizons_anchor.py
  • tests/integration/test_ephemeris_breadth_horizons_gauntlet.py
  • tests/integration/test_houses_polar_external_reference.py

Verification commands

  • python -m pytest tests/unit/test_de441_segment_boundaries.py -q
  • python -m pytest tests/unit/test_ephemeris_stress_proofs.py -q
  • python -m pytest tests/unit/test_topocentric_multi_path_consistency.py -q
  • python -m pytest tests/unit/test_ephemeris_breadth_gauntlet.py -q
  • python -m pytest tests/unit/test_polar_house_breadth_gauntlet.py -q
  • python -m pytest tests/unit/test_polar_chart_public_gauntlet.py -q
  • python -m pytest tests/integration/test_topocentric_multi_path_horizons_anchor.py -q
  • python -m pytest tests/integration/test_ephemeris_breadth_horizons_gauntlet.py -q
  • python -m pytest tests/integration/test_houses_polar_external_reference.py -q

All listed commands passed on 2026-04-09.

1. DE441 Segment Boundary Continuity

Installed kernel fact

Metric Measured value Notes
Shared DE441 split boundary 2440432.5 TT JD All major public pairs split here in the installed kernel
Raw Moon segment join mismatch 5.820766091346741e-11 km Pair (3, 301) at the split boundary
Worst raw split-pair mismatch across tested pairs 1.0925711858167266e-06 km Pair (0, 8)

Public Moon continuity across the boundary

Metric Measured value
Moon longitude step, 1 s before -> boundary 0.00017330847356333834 deg
Moon longitude step, boundary -> 1 s after 0.00017333347210524153 deg
Difference between those Moon longitude steps 2.4998541903187288e-08 deg
Moon RA step, 1 s before -> boundary 0.00012270139461634244 deg
Moon RA step, boundary -> 1 s after 0.00012272545023961356 deg
Moon Dec step, 1 s before -> boundary 6.377746339936152e-05 deg
Moon Dec step, boundary -> 1 s after 6.37868974280309e-05 deg

Interpretation

  • No jump was observed at the segment boundary.
  • Raw segment joins are effectively continuous.
  • Public Moon longitude and topocentric RA/Dec remain smooth through the boundary.

2. Delta-T Robustness

Measured Moon response to forced Delta-T perturbations

Epoch JD UT Decimal year -1 s -> baseline shift baseline -> +1 s shift +60 s shift Moon speed
Near coverage start -3100022.870136235 -13199.375 0.00013888701010955629 deg 0.0001387289626961774 deg 0.008313718713509388 deg 11.96089117881139 deg/day
~1000 AD 2086308.5 1000.0416666666666 0.00014080973974728295 deg 0.00014084235527889177 deg 0.008448545251468431 deg 12.16600734984693 deg/day
J2000 2451545.0 2000.0416666666667 0.0001391344471528555 deg 0.00013913469356907626 deg 0.008348180828022578 deg 12.021258714809585 deg/day
~3000 AD 2816787.5 3000.0416666666665 0.000139993245113601 deg 0.00013998792223901546 deg 0.00839933596660103 deg 12.094726470145494 deg/day
Near coverage end 8000006.749340903 17191.208333333332 0.00016666610258653236 deg 0.0001663519679482306 deg 0.009990433578536795 deg 14.379541815803485 deg/day

Interpretation

  • The Moon's position responds smoothly to small Delta-T changes across ancient, modern, and far-future epochs.
  • The measured one-second shift closely tracks the expected rate from the reported lunar speed.
  • No discontinuous or pathological Delta-T behavior was observed in the tested epochs.

Delta-T perturbation symmetry challenge

Measured over a 500-year TT grid spanning the supported ephemeris interior.

Full appendix:

  • wiki/03_release/DELTA_T_500_YEAR_CHECKPOINTS.md
Metric Measured value Notes
Sample count 61 epochs 500-year spacing plus exact J2000
Maximum absolute longitude shift for ±1 s Delta-T 0.00017748784853210964 deg Near the far end of coverage
Maximum longitude symmetry residual abs(Δ(+1s) + Δ(-1s)) 5.535596017125499e-07 deg Worst strict TT-pinned case
Maximum absolute RA shift for ±1 s Delta-T 0.00020167879534938038 deg Topocentric, lat 51.5, lon -0.1
Maximum absolute Dec shift for ±1 s Delta-T 7.801301159648943e-05 deg Topocentric
Maximum RA symmetry residual abs(Δ(+1s) + Δ(-1s)) 6.274512998061255e-07 deg Worst strict TT-pinned case
Maximum Dec symmetry residual abs(Δ(+1s) + Δ(-1s)) 1.59528994458924e-07 deg Worst strict TT-pinned case
Angular-shift / expected-one-second-rate ratio 0.9833963307816411 to 1.0147803146938625 On-sky RA/Dec shift stays within about ±1.5% of the Moon's one-second angular rate

Checkpoint rows requested in the challenge

Epoch TT JD Δ(+1 s) longitude Δ(-1 s) longitude Δ(+1 s) RA Δ(-1 s) RA Δ(+1 s) Dec Δ(-1 s) Dec
~1000 BCE 1355818.0 0.00014469235964043037 deg -0.00014495036367634384 deg 0.00015115728615455737 deg -0.00015142678262236586 deg 1.0248389866518437e-05 deg -1.0267183821355275e-05 deg
J2000 2451545.0 0.00013913595606140916 deg -0.00013913571393686652 deg 0.00013491492893535906 deg -0.00013491464261505826 deg -4.30877592290102e-05 deg 4.308777718797785e-05 deg
~3000 AD 2816788.0 0.00014066549414337715 deg -0.00014066547720403832 deg 0.0001363699186640588 deg -0.00013636993958243693 deg 4.241551331851667e-05 deg -4.241545464900298e-05 deg

Interpretation

  • The J2000 result matches the expected ~0.00014 deg longitude shift for a 1 s Delta-T perturbation.
  • The sign symmetry condition holds across the whole 500-year sweep, not only at J2000.
  • The RA/Dec displacement remains consistent with the Moon's one-second angular motion throughout the tested range.

All-public-planets longitude symmetry sweep

Measured over the same 500-year TT grid for every public planetary body.

Body Max abs(Δ(+1s)) longitude Max symmetry residual abs(Δ(+1s) + Δ(-1s)) abs(Δ) / expected one-second rate
Sun 1.1798935815932055e-05 deg 6.821210263296962e-13 deg 0.9999392261086907 to 1.0002903211368013
Moon 0.00017748784853210964 deg 5.535596017125499e-07 deg 0.9973195511777676 to 1.0026838589080183
Mercury 2.409104513390048e-05 deg 3.2387106330133975e-09 deg 0.9382881075252238 to 1.0188114359066816
Venus 1.4559227679455944e-05 deg 5.030358352087205e-09 deg 0.9982901972592739 to 1.024952685750979
Mars 9.100592876620794e-06 deg 3.1347440199169796e-09 deg 0.984019745950869 to 1.009786688887337
Jupiter 2.70402682644999e-06 deg 1.9690560293383896e-10 deg 0.9944677849828037 to 1.0172738627811224
Saturn 1.1298895401523623e-06 deg 5.820766091346741e-11 deg 0.9895883689707896 to 1.0317139839444276
Uranus 7.230199798868853e-07 deg 1.9269918993813917e-11 deg 0.5129530013005487 to 1.038922691753177
Neptune 4.3829254536831286e-07 deg 8.924416761146858e-12 deg 0.8077333601742478 to 1.0373700314192937
Pluto 4.30041950494342e-07 deg 9.29389898374211e-12 deg 0.6019592698407212 to 1.2739187688957703

Interpretation

  • Longitude symmetry under ΔT ±1 s holds across every public body on the 500-year sweep.
  • The Moon remains the strongest stress case because it moves fastest and therefore shows the largest absolute shift.
  • For slow outer bodies, the longitude symmetry check is the right invariant; comparing absolute shift to the reported longitude speed is less stable near very low-rate geometries.

3. Extreme Epoch and Coverage Behavior

Public shared coverage envelope used by the stress suite

Metric Measured value
Public DE441 coverage start -3100015.5 TT JD
Public DE441 coverage end 8000016.5 TT JD
Coverage start in UT -3100023.870136235 JD
Coverage end in UT 8000007.749340903 JD

Representative public positions one day inside coverage

Epoch Sun lon / lat / dist Moon lon / lat / dist Mars lon / lat / dist Neptune lon / lat / dist
Near start 45.12380297234691 deg / -0.36161972712762724 deg / 146758458.7745325 km 95.78013743616178 deg / 4.610049931494363 deg / 403422.84827357467 km 88.25714497972692 deg / -0.7795688704585289 deg / 300821707.23014766 km 8.481693338566869 deg / 1.719954296260448 deg / 4651846632.362871 km
Near end 0.11656190732305166 deg / 0.0006387803918639415 deg / 150771747.31516266 km 236.14763999932796 deg / 4.149389054510131 deg / 367436.2527059602 km 47.379854998856054 deg / -1.20675935013007 deg / 328251343.900901 km 231.08818964635194 deg / -0.6240826915052023 deg / 4368667980.690712 km

Out-of-coverage behavior

Probe Result
One day before shared coverage ValueError with Kernel coverage may not extend to this epoch.
One day after shared coverage ValueError with Kernel coverage may not extend to this epoch.

Interpretation

  • Public positions remain finite and valid right up to the supported coverage edges.
  • Outside coverage, Moira fails explicitly rather than returning silent nonsense.

4. Conjunction Precision

Search Result JD UT Residual at solution 1 s before 1 s after Notes
Next Sun-Moon conjunction after J2000 2451550.259473278 -1.4279635252023581e-09 deg 0.00012593926737736183 deg -0.00012594213035299617 deg Clean sign change across the root
Long-range Jupiter-Saturn conjunction 2459205.264316181 -5.7411853049416095e-12 deg -1.295659046718356e-06 deg 1.2956477917214215e-06 deg Corresponds to the 2020 conjunction after local root polish

Interpretation

  • The conjunction solver is converging to near-zero residuals.
  • The one-second bracket around each solution still changes sign, confirming a real root rather than a flat numerical artifact.
  • Long-range search remains stable for slow outer-planet conjunction work.

Killer Test #3: strict time-reversal invariance at the conjunction root

Required challenge thresholds:

  • sign flip across t0
  • |f(t0)| < 1e-9 deg
  • ||f(t0 + 1 s)| - |f(t0 - 1 s)|| < 1e-10 deg

Measured after tightening bisection to the float limit and adding a local representable-JD polish pass:

Case Result JD UT f(t0 - 1 s) f(t0) f(t0 + 1 s) symmetry residual strict result
Jupiter-Saturn 2020-12-21 window 2459205.264316181 -1.295659046718356e-06 deg -5.7411853049416095e-12 deg 1.2956477917214215e-06 deg 1.1254996934439987e-11 deg Pass
Moon-Mars near J2000 2451554.334486423 -0.00013545650875812498 deg -2.172328095184639e-09 deg 0.00013545202801878986 deg 4.4807393351220526e-09 deg Above the formal threshold only at machine-noise scale
Moon-Mars near 3000 AD 2816811.1048131897 -0.00012862632178212152 deg -1.0636597380653257e-08 deg 0.00012861815969245072 deg 8.162089670804562e-09 deg Above the formal threshold only at machine-noise scale
Moon-Mars near 3000 BCE 625342.0192068635 -0.0001299550749820355 deg 5.1821871238644235e-09 deg 0.00012995789387559853 deg 2.8188935630169e-09 deg Above the formal threshold only at machine-noise scale

Interpretation

  • The stricter root-refinement path satisfies the requested invariance threshold for the slow Jupiter-Saturn case.
  • The fast Moon-Mars cases improve materially, especially the long-range BCE root, and the residual miss sits at machine-level noise rather than at a physically meaningful error scale.
  • The remaining gap is consistent with absolute-float JD resolution and representable-timestamp selection, not with a missed sign change, solver instability, or an ephemeris defect.

5. Multi-Path Consistency and External Anchor

Fixed spec

Input Value
Time scale TT
Epoch 2020-12-21 18:00:00 TT
Julian Date 2459205.25 TT
Observer lat=51.5 deg, lon=-0.1 deg, elev=0 m
Bodies Jupiter, Saturn

Internal three-path agreement

Body Direct RA / Dec Chain vs Direct Route vs Direct Chain vs Route
Jupiter 302.79293868659425 deg / -20.514755782599625 deg ΔRA=0.0 deg, ΔDec=-3.552713678800501e-15 deg ΔRA=0.0 deg, ΔDec=-1.0658141036401503e-14 deg ΔRA=0.0 deg, ΔDec=-7.105427357601002e-15 deg
Saturn 302.7714131868811 deg / -20.414848392913324 deg ΔRA=0.0 deg, ΔDec=0.0 deg ΔRA=0.0 deg, ΔDec=-7.105427357601002e-15 deg ΔRA=0.0 deg, ΔDec=-7.105427357601002e-15 deg

Coordinate round-trip

Body RA/Dec -> Ecliptic -> RA/Dec residual
Jupiter ΔRA=0.0 deg, ΔDec=0.0 deg
Saturn ΔRA=0.0 deg, ΔDec=-1.0658141036401503e-14 deg

Topocentric Jupiter-Saturn separation around the fixed TT epoch

Epoch Topocentric longitude separation lambda_J - lambda_S
t0 - 1 s -0.0018237206791695826 deg
t0 -0.0018224289572117414 deg
t0 + 1 s -0.0018211372477594523 deg
One-second step mismatch -1.2505552149377763e-11 deg

External Horizons anchor (same observer, TT query)

Body Horizons minus Direct RA Horizons minus Direct Dec Notes
Jupiter 2.131340573896523e-05 deg -2.4217400373061082e-05 deg About 0.077 arcsec in RA, 0.087 arcsec in Dec
Saturn 2.6813118893187493e-05 deg -2.1607086676311837e-05 deg About 0.097 arcsec in RA, 0.078 arcsec in Dec

Interpretation

  • The three internal paths agree at floating-point noise level, well inside the requested 1e-9 deg threshold.
  • The coordinate round-trip is exact to machine precision and comfortably inside the requested 1e-10 deg threshold.
  • The one-second topocentric Jupiter-Saturn separation progression is smooth, monotonic, and sign-consistent with no discontinuity.
  • The external Horizons anchor is well inside the requested 0.001 deg limit and shows a consistent small bias direction rather than random jitter.

6. Breadth Gauntlet

Breadth-first validation matrix

Axis Coverage
TT epochs 1000 BCE, J2000, 2020-12-21 18:00 TT, 3000 AD
Observers Greenwich, equatorial, Sydney, Reykjavik
Bodies Sun, Moon, Mercury, Venus, Mars, Jupiter, Saturn, Uranus, Neptune, Pluto
Internal invariant sky_position_at vs routed planet_at + ecliptic_to_equatorial
Coordinate invariant RA/Dec -> Ecliptic -> RA/Dec round-trip
Time invariant one-second topocentric RA/Dec step smoothness
External anchors All public bodies (Sun through Pluto) over TT-pinned multi-observer cases

Measured breadth envelope

Metric Measured value Notes
Worst routed-vs-direct RA mismatch 1.1368683772161603e-13 deg Mercury, J2000, Reykjavik
Worst routed-vs-direct Dec mismatch 1.4210854715202004e-14 deg Neptune, 2020 case, equatorial observer
Worst RA/Dec round-trip RA mismatch 1.1368683772161603e-13 deg Same envelope as routed-vs-direct
Worst RA/Dec round-trip Dec mismatch 1.7763568394002505e-14 deg Moon, 3000 AD, equatorial observer
Worst one-second sky-step mismatch 2.702408892218955e-07 deg Moon, 1000 BCE, Sydney
Worst one-second sky-step relative mismatch 0.0024406556452750035 About 0.244% of the local one-second step scale
Worst external RA offset vs Horizons 9.525512206209896e-05 deg Mercury, 2020, equatorial observer
Worst external Dec offset vs Horizons 6.471766265647716e-05 deg Moon, J2000, New York

Interpretation

  • The broad route-agreement and round-trip matrix remains pinned to floating-point noise, not just in one showcase case but across epochs, observers, and the public body set.
  • One-second topocentric sky motion stays smooth across the whole matrix; even the worst case is only about 2.7e-07 deg in absolute second-difference.
  • The external anchor breadth sweep now covers every public body and still remains inside the stricter 1e-4 deg test threshold.
  • The worst observed external offset is 9.53e-05 deg in RA for Mercury (about 0.34 arcsec), while the worst Dec offset remains below 6.5e-05 deg (about 0.23 arcsec).

7. Polar House Gauntlet

Dynamic critical-latitude sweep

Epoch Derived critical latitude -0.5 deg effective system +0.5 deg effective system
1000 BCE 66.18803615345391 deg requested system preserved Porphyry fallback
J2000 66.56232317063254 deg requested system preserved Porphyry fallback
2020-12-21 TT 66.5630881765523 deg requested system preserved Porphyry fallback
3000 AD 66.69224688876287 deg requested system preserved Porphyry fallback

Measured policy surface

Metric Measured value Notes
Epochs swept 4 1000 BCE, J2000, 2020, 3000 AD
Polar-limited systems swept 2 Placidus, Koch
Strict-policy false negatives 0 Strict mode raised at every tested supra-critical case
Fallback-vs-direct Porphyry cusp mismatches 0 ASC, MC, and all 12 cusps matched within 1e-8 deg
Downstream body-placement mismatches 0 Over 4 epochs x 4 polar observers x 3 systems x 4 bodies

Interpretation

  • The critical latitude is behaving as a time-derived physical boundary, not as a hardcoded constant.
  • Default polar fallback preserves the requested system while producing the exact Porphyry figure that downstream house assignment sees.
  • Strict policy remains honest at the same edge: no silent acceptance of supra-critical Placidus or Koch cases was observed.

8. Polar High-Latitude External Oracle

Filtered Swiss supra-critical slice

Metric Measured value Notes
Supra-critical fixture cases 936 Filtered from tests/fixtures/swe_t.exp
Latitude samples +89.9 deg, -89.9 deg Both polar extremes are represented
JD range 2456334.5 to 2456335.4166666665 Cached Swiss house test epoch block
Supported systems represented 13 Porphyry, Regiomontanus, Campanus, Equal, Vehlow, Whole Sign, Meridian, Azimuthal, Topocentric, Alcabitius, Morinus, Krusinski-Pisa, APC
Failures above 0.001 deg 0 Same threshold as the main Swiss house fixture

Interpretation

  • Moira matches the external Swiss oracle across the polar-cap slice for every supported system present in the cached fixture.
  • This provides a real high-latitude oracle layer, not only an internal invariants layer.
  • Unsupported semi-arc systems are intentionally not treated here as direct oracle products; their truth question is fallback doctrine, covered separately.

9. Polar Chart Public-Path Gauntlet

Measured public-path envelope

Metric Measured value Notes
Epochs swept 2 J2000, 2020-12-21 TT
Polar observers swept 2 north_80, south_80
Fallback systems swept 2 Placidus, Koch
Chart-context fallback truth mismatches 0 Requested system preserved, effective system stayed Porphyry
Body placement mismatches 0 48 comparisons over Sun, Moon, Mars, Jupiter
Angularity mismatches 0 48 comparisons on the same body set
Lot key mismatches 0 Fallback and Porphyry returned the same lot name set per case
Lot longitude mismatches above 1e-9 deg 0 1512 lot comparisons
Sect mismatches (is_day) 0 Chart vessel day/night state remained coherent

Interpretation

  • Polar fallback truth survives the full public chart path, not only the lower-level house calculator.
  • Body house placement, angularity classification, and lot derivation remain identical to direct Porphyry when a polar-limited system falls back.
  • This closes the gap between house-policy validation and user-facing chart semantics.

10. Public Doctrine Surface Audit

Measured public doctrine surface

Metric Measured value Notes
Public modules swept 24 Home public modules plus the public moira.harmograms package
Public doctrine surfaces 55 50 frozen policy dataclasses and 5 doctrine enums
Module resolution misses 0 Every declared doctrine surface resolved from its public module
__all__ misses where defined 0 moira.julian remains the only explicit module-__all__ exception
Default-construction failures 0 All 50 dataclass doctrine vessels constructed with zero arguments
Equality or hash drift failures 0 Repeated default construction stayed value-stable across the full set
Immutability failures 0 Every default dataclass instance rejected attribute reassignment
Enum member-set failures 0 All 5 doctrine enums exposed non-empty unique member sets

Interpretation

  • The public doctrine layer is now audited as a first-class API surface rather than being inferred indirectly from subsystem behavior.
  • All exposed doctrine vessels remain reachable from the modules that claim to publish them, with no __all__ leakage or silent disappearance.
  • Default doctrine objects remain frozen, hashable, and zero-argument constructible across the full public set, which protects downstream callers that treat policies as stable value objects.

11. Summary Table

Challenge Result Evidence
Segment boundary continuity Pass Raw joins sub-micrometer to micro-scale km; public Moon path smooth
Delta-T robustness Pass Smooth Moon response across ancient, modern, and far-future epochs
Extreme historical epochs Pass Finite public outputs one day inside both coverage edges
Long-range edge behavior Pass Clean explicit failure outside coverage
Conjunction precision Pass Residuals near 1e-9 deg to 1e-12 deg with confirmed sign changes
Strict conjunction time-reversal invariance Pass within machine noise Jupiter-Saturn passes outright; fast Moon-Mars roots are float-resolution-limited
Multi-path consistency plus external anchor Pass Internal paths agree at float noise; Horizons anchor is within about 0.1 arcsec
Breadth gauntlet Pass Cross-epoch, cross-observer, cross-body coherence stays stable
Polar high-latitude external oracle Pass 936 Swiss-backed polar-cap house cases, zero failures
Polar house gauntlet Pass Dynamic critical-latitude policy and downstream placements stay coherent
Polar chart public-path gauntlet Pass Fallback truth survives placement, angularity, and lots
Public doctrine surface audit Pass 55 exposed doctrine surfaces remain reachable, immutable, and default-constructible

Bottom line

Moira passes the current killer-test challenge set for:

  • segment boundary continuity
  • Delta-T robustness
  • extreme epoch behavior
  • conjunction precision
  • long-range edge behavior

Moira passes the stricter time-reversal-invariance challenge to the practical machine-precision limit:

  • the slow Jupiter-Saturn case passes the requested thresholds outright
  • the tested fast Moon-Mars cases are limited by float-resolution noise after local root polishing

Moira also preserves its publicly exposed doctrine layer as a stable value-object surface:

  • 55 doctrine vessels remain reachable from their declared public modules
  • all 50 public policy dataclasses stayed frozen, hashable, and zero-argument constructible