moira

Generated Physical Heliacal-Visibility Capability Matrix

Generated from current Python exports, the FastAPI OpenAPI registry, and the Phase 7 evidence registry by scripts/generate_physical_visibility_inventory.py. Do not edit this file by hand.

Capability matrix

Capability Admission Availability Named evidence classes Boundary
Legacy generalized visibility and heliacal search admitted compatibility contract default legacy policy legacy_regression_fixtures, historical_event_corroboration Not relabelled as the physical model.
Physical single-epoch point-source assessment admitted explicit opt-in plus external pack primary_source_equation_validation, independent_libradtran_holdouts, modern_era_observational_comparison, property_and_invariant_testing, separated_numerical_tolerances, public_contract_and_openapi_parity Clear-sky, unresolved steady point sources inside the exact pack domain.
Physical four-phase first/last event search admitted explicit opt-in plus exact pack 1.2.0 external_ephemeris_event_goldens, property_and_invariant_testing, separated_numerical_tolerances, public_contract_and_openapi_parity Event ownership and geometry are validated; event goldens are not observed visibility dates.
External physical visibility data pack admitted immutable resource caller/server supplied; never downloaded independent_libradtran_holdouts, release_artifact_and_offline_install CC BY-SA 4.0 pack remains separate from the MIT Python distribution.
Two Phase 6 numerical kernels admitted private optimization no public API and Python-owned policy native_python_differential Differential/performance evidence is not scientific validation.
Site-specific experimental moonlight extensions quarantined research absent from product and release surfaces experimental_site_specific_moonlight_quarantine A separately authorized model and evidence program would be required to reopen this work.
Cloudy sky, live weather, telescopes, extended objects, Sun, Moon crescent, and observer-population confidence closed exclusion absent property_and_invariant_testing A new model identity and independent evidence are required.

Evidence registry

Fingerprints bind each class to the canonical LF-normalized UTF-8 content of every listed source, test, fixture, artifact, or receipt. They are drift indicators, not accuracy scores.

Evidence class Role Bound files Fingerprint Limitation
primary_source_equation_validation scientific_admission 3 11e022118455c163 Equation checks validate named components and source examples; they do not establish universal atmosphere or observer truth.
independent_libradtran_holdouts scientific_admission 4 af74b1da52bd1265 Holdouts are independent of LUT fitting but remain within the frozen libRadtran scenario family.
modern_era_observational_comparison bounded_observational_validation 2 8c110feaacdf67a6 The Tousey-Koomen 1953 Table I comparison is a source-bounded modern-era twilight observation, not a contemporary multi-site campaign.
historical_event_corroboration corroboration_only 3 27e95868627f7840 Historical and published-window cases govern only legacy bounded-window claims; they are not exact physical-model event oracles.
property_and_invariant_testing computational_assurance 4 c8134dd842e7066d Invariants prove contract and numerical behavior, not observational accuracy.
legacy_regression_fixtures compatibility_assurance 4 5e4f9608022411d7 Regression fixtures freeze compatibility; snapshots are not scientific authority.
public_contract_and_openapi_parity transport_assurance 3 49886c5922b89944 Transport parity proves that typed engine truth survives projection; it does not validate the underlying science.
external_ephemeris_event_goldens astronomical_and_event_contract_validation 11 aa346b80fd045c79 Twelve independently reconstructed event times cover every admitted target and phase, while four additional matrix cells are explicitly typed engine regressions rather than event-time oracles; none are observed first-visibility dates.
separated_numerical_tolerances numerical_admission 5 4cd707610d3eb87d Solver, storage, interpolation, root, and scenario sensitivity remain separate; there is no aggregate accuracy score.
native_python_differential implementation_admission 3 748eaff30dec6e3c Differential evidence admits two private numerical kernels only; it is not independent scientific validation.
release_artifact_and_offline_install release_assurance 4 4efce44519438e9b A local clean-artifact receipt proves the tested artifacts; publication and downstream deployment are separate release operations.
experimental_site_specific_moonlight_quarantine quarantined_research_boundary 2 48e0d8714052b997 The Jones/Paranal experiment is archived research only. It has no runtime module, public API, compatibility contract, packaged resource, release gate, or active roadmap dependency.

Tolerance law

Source-solver/Monte Carlo uncertainty, binary storage error, LUT interpolation error, root-time tolerance, root-margin residual, limiting-magnitude envelopes, and deterministic event-time sensitivity are retained as separate receipts. Moira publishes no single aggregate tolerance or probabilistic confidence score for this model.