Date: 2026-07-30

Status: Environmental-parameter semantics gate passed. Phase 1 remains in progress. This is not a runtime-table, data-pack, engine, API, or release receipt.

Governing plan: PHYSICAL_HELIACAL_VISIBILITY_IMPLEMENTATION_PLAN.md

Predecessor: PHYSICAL_HELIACAL_VISIBILITY_PHASE1_NAMED_SPECTRAL_DIRECT_CHECKPOINT_2026-07-30.md

Compact source-owned receipt: tests/artifacts/visibility_reference_lab/phase1_environment_contract_checkpoint_2026-07-30.json

Checkpoint 4 admitted the clear molecular REPTRAN-fine direct-transmission reference but deliberately left environmental parameters open. This checkpoint resolves the control semantics required before an adaptive radiance design can be trusted:

• observer altitude and pressure ownership;
• ozone-column meaning;
• AOD550 and Angstrom binding;
• the complete named Shettle haze/season inventory;
• aerosol visibility, humidity, and temperature ownership;
• ground-albedo placement; and
• the distinction between physical line-of-sight extinction and delta-M solver bookkeeping.

The checkpoint freezes parameter roles, candidate nodes, reserved holdouts, and fail-closed boundaries. It does not claim that interpolation over those nodes is already accurate.

The probe binds:

• libRadtran 2.0.6 archive: 64930cc40b6e4a37aa220520974d330fc1563796f466a649b2238131f2d69840;
• built uvspec: d4e94259296a65f7700a0911f0dc7fc14aacde89985befac0266fe0a18531b7a;
• ten governing manual/source files covering aerosol, pressure, ozone, relative humidity, Beer-Lambert/Chapman geometry, and delta-M output semantics;
• all eight data/aerosol/shettle/tau550.<vulcan>.<season> files; and
• the exact Checkpoint 4 specification, builder, validator, and compact checkpoint.

The source declarations are in scripts/visibility_reference_lab/phase1_environment_contract_probe_spec.json. Every declared byte count and SHA-256 is rechecked before generation.

The first pack requires one of the six source-bound AFGL molecular profiles. Near-surface temperature, water vapor, and relative humidity are derived from that named profile.

Independent temperature or relative-humidity overrides are not admitted in the first pack. libRadtran derives the aerosol-grid relative humidity from the atmosphere's water-vapor density and temperature. Treating a single surface humidity value as a complete vertical humidity profile would invent missing atmosphere state.

An explicit humidity/temperature construction may be added later only under a new source-identified profile law with its own validation.

The admitted construction remains the Checkpoint 2 source-equivalent truncated atmosphere. The candidate nodes remain:

0, 500, 1500, 3000, 5000 m

The following values are reserved as untouched interpolation holdouts:

250, 1000, 2250, 4000 m

Interpolation across altitude remains unadmitted until those holdouts are executed against direct-transmission and radiance quantities.

Default surface pressure is derived from the named atmosphere at the observer altitude.

An explicit measured pressure override is admitted as an input policy, but the table coordinate is:

requested surface pressure
---------------------------------------------
named-profile pressure at observer altitude

Both the absolute pressure and the ratio must pass:

absolute pressure: 500 to 1100 hPa
profile-pressure ratio: 0.85 to 1.08

This prevents an absolute pressure that is numerically inside a global box but physically incoherent with the selected altitude/profile combination. The ratio candidates are 0.85, 0.925, 1.0, 1.04, and 1.08; separate midpoint holdouts are reserved.

libRadtran's pressure control scales the pressure profile and well-mixed gases. Ozone is then independently normalized to the requested above-observer column.

Ozone is the total column above the observer altitude in Dobson units and maps to:

mol_modify O3 <value> DU

Candidate values are 200, 250, 300, 350, 400, and 500 DU, with intermediate values reserved as holdouts.

AOD is authoritative at 550 nm. libRadtran's command accepts Angstrom alpha and beta, so the exact binding is:

beta = AOD550 * 0.55 ** alpha

The separate libRadtran visibility control is not exposed. Internally, libRadtran maps the requested optical depth back to a Shettle visibility to construct the vertical aerosol profile. Exposing both visibility and AOD would create conflicting authorities for the same component.

The frozen candidate values are:

AOD550: 0, 0.02, 0.05, 0.1, 0.2, 0.4, 0.7, 1.0
alpha:  0, 0.5, 1.0, 1.3, 1.8, 2.5

Intermediate holdouts are reserved and are not used to tune the future table.

The first-pack stratospheric background is fixed to libRadtran vulcan code 1. All eight supported lower-atmosphere haze/season combinations are named:

Haze affects aerosol single-scattering albedo and phase behavior and therefore belongs to directional radiance. With AOD550 and alpha fixed, the direct extinction dimensions are season, AOD550, and alpha. The radiance dimensions also include haze.

The first-pack surface law is explicitly a gray Lambertian approximation. Candidate albedos are 0, 0.02, 0.1, 0.2, 0.5, 0.8, and 1.0, with separate holdouts.

Ground albedo is excluded from the direct-transmission table and included in the radiance table. Spectral surface classes or BRDFs would require a separate source and complexity decision.

libRadtran's default delta-M treatment is appropriate for diffuse radiative-transfer efficiency, but a raw direct-beam output can inherit a small dependence on aerosol phase-function/haze type. That is not physical point-source extinction: the Beer-Lambert direct term depends on total extinction, not on how extinction is divided into forward scattering and absorption.

The admitted direct-extinction oracle therefore preserves aerosol optical depth while setting:

aerosol_modify ssa set 0

for the direct-only generator run. This converts the same aerosol extinction to absorption for solver bookkeeping, removing the aerosol delta-M term without changing total line-of-sight extinction. Directional-radiance runs do not use this override.

The 73-run artifact proves both sides:

• the raw delta-M diagnostic produces three distinct outputs across four haze codes; and
• the admitted direct-extinction oracle produces one byte-identical output within each season across all four haze codes.

This is an additive Checkpoint 5 solver identity. It does not rewrite the clear-molecular Checkpoint 4 result, where no aerosol delta-M term existed.

The immutable artifact contains 73 deterministic cases:

• 7 ground-albedo direct-invariance cases;
• 8 named aerosol profile cases;
• 4 raw delta-M haze diagnostics;
• 24 AOD cases across 0.25, 5, and 20 degrees;
• 18 Angstrom cases at 400, 550, and 780 nm;
• 6 ozone cases at 600 nm;
• 5 pressure-ratio cases at 550 nm; and
• 1 exact repeat.

Results:

• every albedo case has the same direct-output SHA-256;
• every admitted same-season haze case has the same direct-output SHA-256;
• summer and winter direct outputs differ;
• transmission decreases strictly with AOD at every tested altitude;
• near-horizon aerosol slant optical depth is materially nonlinear in AOD: the maximum/minimum tau/AOD ratio is 2.36273098116443 at 0.25 degrees, versus 1.08487839009281 at 5 degrees and 1.00578580135771 at 20 degrees;
• all alpha values are byte-identical at 550 nm, proving the AOD550 normalization;
• increasing alpha decreases 400 nm transmission and increases 780 nm transmission;
• increasing ozone decreases 600 nm transmission;
• increasing pressure ratio decreases 550 nm transmission; and
• the fixed repeat is byte-identical for input, output, seed, stderr, and syntax-check files.

The strong near-horizon AOD nonlinearity rejects an apparent shortcut: direct aerosol extinction cannot be represented as one unit-AOD surface and scaled linearly over the full 0-1 domain under the selected Shettle construction.

Admitted artifact:

external directory:
  phase1-environment-contract-2026-07-30-v2
manifest SHA-256:
  e79a250b01f00783f272bae409fa323a94b5c7811375760bf536eaa7de6b0580
generation fingerprint:
  882f23ac18053ca616b01f175c31cc26a4c68411021506d80d8d308659060cb4
run count:
  73

The artifact passed its independent validator under both WSL Python and the repository's Windows Python environment.

The discarded v1 experiment exposed a one-ULP platform difference in a derived logarithm. It is not admitted. The v2 builder serializes derived quantities to a fixed 15-significant-digit boundary and passes byte-level cross-platform reconstruction.

• no moira runtime code changed;
• no native code changed;
• no public Python export, facade, serializer, REST model, route, or OpenAPI contract changed;
• no package dependency changed;
• no physical table entered the wheel;
• no network or automatic download entered normal calculation;
• no production data pack was authorized; and
• Phase 2 did not begin.

Checkpoint 5 closes environmental parameter roles and source semantics. It does not close:

• altitude and pressure-ratio interpolation against reserved holdouts;
• solar/target/azimuth adaptive radiance nodes;
• a deep-twilight sampling and convergence law;
• versioned CIE response and target-spectrum inputs;
• response-integrated spectral products;
• storage precision and interpolation format;
• end-to-end error propagation;
• the separate checksummed visibility data pack; or
• the Phase 1 exit gate.

The next bounded gate is the altitude/pressure holdout and interpolation study, followed by adaptive directional-radiance design. Phase 2 remains inactive.