moira

Physical Heliacal Visibility Broad Oracle - 2026-08-09

Status: accepted local Phase 7 engine release gate; no tag, package publication, downstream installation, deployment, or website change was performed.

Outcome

The original exact end-to-end event oracle contained one Jupiter morning-first rising and one Sirius morning-first rising. That evidence remains valid, but it was not broad enough to carry the four-target, four-phase public event surface by itself.

The expanded gate predeclares the complete product of:

• targets: Mars, Jupiter, Saturn, and Sirius; and
• phases: morning first rising, morning first setting, evening last rising, and evening last setting.

The resulting 16 cells contain:

• 12 independently reconstructed event times;
• 4 explicit typed negative/domain regression cells;
• 40 checksum-bound one-minute NASA/JPL Horizons source files;
• checksum-bound Hipparcos Sirius astrometry and a pinned offline Astropy/ERFA/IERS transform; and
• the exact external visibility pack 1.2.0, manifest cf93433a9f66a5ea92832271ce3c4b023fcc8693164803539a9f1be85b17468c.

All 12 event-time comparisons pass the fixed 60-second limit. The largest absolute engine/oracle difference is 5.8789461851119995 seconds. The live public-engine replay passes all 16 cells with the declared default search policy and content-admitted DE441.

Independence boundary

The source acquisition step uses an exact engine result only to center a bounded external query. It does not use that engine timestamp as oracle truth. The offline oracle then derives the event from:

1. JPL airless target and solar geometry;
2. JPL apparent planetary photometry, or Hipparcos/Astropy Sirius geometry and source-owned Sirius photometry;
3. a separate implementation of the immutable visibility-pack interpolation, CIE MES2, Crumey threshold, extinction, and target-color equations;
4. independently reconstructed target-boundary and solar-side ownership; and
5. linear root interpolation on the fixed one-minute external grid.

The builder and Phase 7 validator import neither Moira, the production event solver, nor a network client. They contain no direct NumPy import. The isolated Sirius authority-validation toolchain uses the declared development-only Astropy/ERFA stack and may load NumPy transitively; neither NumPy nor Astropy is part of the published engine, the live engine replay, or the base runtime. The acquisition script is the only networked step. The stored validator replay is offline and checksum locked.

Timed-event results

Case	Engine/oracle difference (s)	Guard-day independent basis
Mars, morning first setting, Babylon	0.352322	target setting occurs after sunrise on the adjacent day
Mars, evening last rising, Babylon	1.195729	target rising occurs before sunset on the adjacent day
Jupiter, evening last setting, Levant	0.679578	requested margin crossing is absent on the adjacent day
Jupiter, morning first rising, Levant	0.759038	requested margin crossing is absent on the adjacent day
Jupiter, morning first setting, Levant	0.144638	target setting occurs after sunrise on the adjacent day
Jupiter, evening last rising, Levant	0.194044	target rising occurs before sunset on the adjacent day
Saturn, morning first setting, Sydney	0.232548	target setting occurs after sunrise on the adjacent day
Saturn, evening last rising, Sydney	0.425063	target rising occurs before sunset on the adjacent day
Sirius, evening last setting, New Orleans	5.878946	requested margin crossing is absent on the adjacent day
Sirius, morning first rising, New Orleans	3.780989	requested margin crossing is absent on the adjacent day
Sirius, morning first setting, New Orleans	1.019588	target setting occurs after sunrise on the adjacent day
Sirius, evening last rising, New Orleans	0.414683	target rising occurs before sunset on the adjacent day

The adjacent-day trace matters. A raw visibility-margin zero can exist while the requested target boundary lies on the wrong side of sunrise or sunset. Such a day does not own the requested phase. The broadened oracle reconstructs that geometry rather than accepting only the selected event timestamp.

Typed negative/domain cells

These four matrix cells are intentionally not described as independent event times:

Case	Public result	Reason
Mars, evening last setting, Greenwich	not_found	no_phase_transition_in_search_window
Mars, morning first rising, Greenwich	not_found	no_phase_transition_in_search_window
Saturn, evening last setting, Greenwich	not_evaluable	target_altitude_out_of_domain
Saturn, morning first rising, Greenwich	not_found	no_phase_transition_in_search_window

They are full-window public-engine regressions that verify fail-closed status, reason, evidence state, and absence of fabricated event fields. They must not be counted as four additional event-time oracles.

Source-controlled evidence

• Frozen acquisition specification: scripts/visibility_reference_lab/physical_visibility_phase7_source_acquisition_spec_v1.json
• Matrix specification: scripts/visibility_reference_lab/physical_visibility_phase7_broad_oracle_matrix_v1.json
• Networked acquisition tool: scripts/acquire_visibility_phase7_oracle_sources.py
• Engine-only window discovery tool: scripts/discover_visibility_phase7_oracle_windows.py
• Offline independent builder: scripts/build_visibility_phase7_broad_oracle.py
• Offline independent validator: scripts/validate_visibility_phase7_broad_oracle.py
• Immutable golden: tests/golden/physical_visibility_phase7_broad_oracle.json
• External-source recovery receipt: tests/artifacts/release/physical_visibility_phase7_source_recovery_2026-08-11.json
• Governance: tests/unit/test_visibility_phase7_broad_oracle_governance.py
• Live public-engine replay: tests/integration/test_physical_visibility_phase7_broad_oracle.py

The golden SHA-256 is 29d96b8eb1187c013357039df8c224e6a41381d83bb81fd961e24057a563ede9. It records the exact source-file names, byte counts, SHA-256 checksums, query identities, pack identity, toolchain receipt, independent brackets, event roots, target/solar boundary roots, guard-day basis, captured engine results, and residuals.

Acquisition provenance recovery - 2026-08-11

The external sources and the admitted matrix were frozen in two deliberate steps. The network acquisition used the 5,311-byte pre-admission specification now preserved at scripts/visibility_reference_lab/physical_visibility_phase7_source_acquisition_spec_v1.json. Its raw LF SHA-256 is ba061013c6e6258475baab4442b072c82c887aabc840cc19f5b0126542eb9323, which is the exact digest recorded by the archived NASA/JPL source manifest.

The later admitted matrix adds the release gate and physical-policy receipt. Its raw LF SHA-256 is 9237a3829ec97121845a51827fd7013b71f8802fee02005e1d26c4222aec0f9a. The acquisition-driving fields - schema, status, selection policy, five sites, and all 16 cases - are identical in both files. Governance tests bind the two files and reject drift. This separation removes the former self-reference in which the acquisition tool defaulted to the later file containing the earlier file’s digest.

The 47 external source files were also preserved in the verified 48-entry archive physical-visibility-phase7-source-bundle-2026-08-11.zip, SHA-256 ebad4060250702c2dff378c960a67de657dab9a5366856825907e93f88b4e103. The archive remains external; its machine-independent identity and complete verification checks are recorded in the source-controlled recovery receipt.

Limits of the claim

This gate validates the admitted clear-sky, naked-eye, unresolved point-source model and its public event semantics. It does not control or predict clouds, smoke, local light pollution, campfires, obstructions, observer physiology, or other real-world conditions outside the explicit policy. It also does not admit Mercury or Venus event search, additional fixed stars, telescopic visibility, extended objects, or the quarantined Jones/Paranal experiment.

This site is open source. Improve this page.