moira

Physical Heliacal Visibility Phase 1 Checkpoint

Date: 2026-07-29 Status: Phase 1 in progress; reproducible reference-lab checkpoint, not a runtime data pack and not a model-admission receipt Commit: Uncommitted working tree; no commit or push was requested for this checkpoint Governing plan: PHYSICAL_HELIACAL_VISIBILITY_IMPLEMENTATION_PLAN.md

Subsequent evidence: PHYSICAL_HELIACAL_VISIBILITY_PHASE1_ELEVATED_SITE_CHECKPOINT_2026-07-29.md closes the bounded named-profile elevated-site construction gate. Statements below that identify nonzero observer altitude as open describe this checkpoint’s historical boundary.

Outcome

The external libRadtran 2.0.6 reference laboratory now has a checksum-locked, offline generator and an independent artifact validator. The first convergence, geometry, and direct-transmission evidence sets were reproduced and bound to the exact source archive, executable, build configuration, scientific input trees, repository specification, builder, validator, and per-case outputs.

This checkpoint does not close Phase 1. It deliberately stops before a production grid, interpolation law, runtime table, or data pack.

Implemented Repository Surface

• scripts/visibility_reference_lab/phase1_lab_spec.json
  • fixes the research-only model and runtime boundary;
  • locks libRadtran 2.0.6 by bytes and SHA-256;
  • fixes fully spherical one-dimensional MYSTIC controls;
  • declares candidate domain envelopes without authorizing their Cartesian expansion;
  • defines a repeated convergence profile, a geometry smoke profile, and a repeated direct-transmission smoke profile, plus untouched holdouts; and
  • excludes exact-horizon and nonzero-observer-altitude generation until their constructions are admitted.
• scripts/build_visibility_radiance_lut.py
  • performs no download;
  • verifies source, build, executable, configuration, and input-data receipts;
  • renders deterministic path-neutral libRadtran inputs;
  • builds only three authorized serial profiles;
  • preserves failed cases for diagnosis;
  • resumes only complete, checksum-valid cases;
  • rejects unowned files and symlinks;
  • verifies a fixed-seed byte-identical repeat; and
  • rechecks immutable inputs before emitting a manifest.
• scripts/validate_visibility_radiance_lut.py
  • validates the exact current specification and tools;
  • validates the complete authorized case inventory;
  • verifies every bound file and checksum;
  • verifies build, source-data, environment, and runtime-boundary receipts; and
  • rejects stale, partial, extended, or tampered artifacts.
• tests/artifacts/visibility_reference_lab/phase1_reference_lab_checkpoint_2026-07-29.json
  • preserves the compact source-owned numerical and provenance receipt without adding libRadtran, CIE data, or a runtime table to the package.

No engine module, public Python contract, native calculation, facade, serializer, REST model, OpenAPI surface, dependency declaration, or default changed.

External Build Receipt

Field	Receipt
Source	libRadtran 2.0.6
Archive bytes	154147176
Archive SHA-256	64930cc40b6e4a37aa220520974d330fc1563796f466a649b2238131f2d69840
uvspec identity	uvspec, version 2.0.6-MYSTIC
uvspec SHA-256	d4e94259296a65f7700a0911f0dc7fc14aacde89985befac0266fe0a18531b7a
Platform	Ubuntu 26.04, WSL2, Linux 6.6.114.1
Compilers	GCC, G++, and GFortran 15.2.0
Supporting tools	GNU Make 4.4.1, Flex 2.6.4, netCDF 4.9.3, GSL 2.8
Flags	CFLAGS=-O2 CXXFLAGS=-O2 FFLAGS=-O2
Solver capabilities	MYSTIC, GSL, and netCDF4 enabled; MYSTIC 3D disabled; VROOM exercised
Upstream make check	0 failed uvspec calls; 0 failed difference checks
Upstream unit tests	458067 assertions in 17 test cases, all passed

The official archive contains stale generated src/.depend and libsrc_c/.depend files with machine-specific include paths. The documented clean-build procedure removes only those two generated files. No libRadtran source patch was applied.

Immutable Artifact Receipts

Profile	Cases	Files	Bytes	Root manifest SHA-256
Convergence	21	274	96837	f5246c0b54b7f1e1cb126a275df5a6c425cf71759d7afcf704dff14d991c2487
Geometry smoke	6	79	30883	736cb97e0f5e4218bb386f0290454cc15140438126ca1ded70b3b54f49a9d10a
Direct-transmission smoke	13	92	43528	5eaac5edd15081836ef1a57c62e360421a5bd73f71276095cb0cd6ea78b80aa1

All three artifacts:

• validate against the current repository specification and tool hashes;
• bind every output file;
• claim no network or runtime dependency; and
• carry phase1_reference_lab_evidence_not_runtime_data_pack status.

The convergence profile reproduces the first case byte for byte when its seed and inputs are repeated. The deterministic direct-transmission profile also reproduces every raw output byte for its repeated case.

Convergence Evidence

The convergence case is a sea-level, 550 nm, U.S. Standard atmosphere reference with:

• solar center altitude: -6 degrees;
• target true altitude: 5 degrees;
• relative solar azimuth: 180 degrees;
• AOD550: 0.1;
• Angstrom exponent: 1.3;
• ozone: 300 DU;
• surface pressure: 1013.25 hPa; and
• ground albedo: 0.2.

Five independent seeds were run at every photon count:

Photons	Mean radiance (mW m^-2 nm^-1 sr^-1)	Between-seed sample RSD	Mean MYSTIC-reported RSD
20000	0.014740458	22.90%	33.10%
100000	0.013240120	7.86%	15.13%
500000	0.013153260	4.84%	7.43%
2000000	0.013222160	2.32%	3.72%

The uncertainty contracts with the expected Monte Carlo scale, but two million photons still do not establish production precision.

Geometry Evidence

Every case below used 500000 photons, seed 32452843, and 550 nm:

Case	Sun altitude	Target altitude	Relative azimuth	Radiance (mW m^-2 nm^-1 sr^-1)	Reported RSD
Civil, near Sun	-6 deg	5 deg	0 deg	0.463911	2.88%
Civil, quadrature	-6 deg	5 deg	90 deg	0.0231493	5.95%
Civil, antisolar	-6 deg	5 deg	180 deg	0.0140619	6.97%
Nautical, near horizon	-12 deg	1 deg	90 deg	0.000040657	67.54%
Astronomical, near horizon	-18 deg	0.25 deg	90 deg	0 sampled	not estimable
Sunset, high target	0 deg	20 deg	90 deg	4.62224	0.44%

This is a decisive negative result for a fixed photon budget. The same budget ranges from sub-percent uncertainty to an unusable sparse-contribution estimate. The zero sampled at the deepest case is not physical-zero evidence and is not admitted as a table value.

Direct-Transmission Evidence

Direct transmission is a separate deterministic component. It uses libRadtran’s recommended double-precision DISORT solver with pseudo-spherical geometry, 16 streams, true geometric target altitude, and:

direct_spectral_transmission = (edir / E0) / sin(target_altitude)
extinction_magnitude = -2.5 log10(direct_spectral_transmission)

The sine division removes the horizontal irradiance projection from libRadtran’s documented output_quantity transmittance value. It is not an empirical air-mass fit.

At 550 nm:

Target altitude	Direct transmission	Extinction
0.25 deg	0.0003724155	8.57243 mag
0.5 deg	0.0008554776	7.66948 mag
1 deg	0.0029792195	6.31474 mag
2 deg	0.0141836458	4.62053 mag
5 deg	0.1032076570	2.46572 mag
10 deg	0.2895282903	1.34577 mag
20 deg	0.5237992659	0.70209 mag
45 deg	0.7295975852	0.34229 mag

At 5 degrees true altitude:

Wavelength	Direct transmission	Extinction
380 nm	0.0017691995	6.88056 mag
450 nm	0.0254750167	3.98471 mag
550 nm	0.1032076570	2.46572 mag
650 nm	0.2184704002	1.65152 mag
780 nm	0.3921759935	1.01630 mag

These results are smoke evidence, not a runtime table. The final spectral grid and a near-horizon solver/model error bound remain open.

Packaging Boundary Audit

A fresh local Moira 6.1.0 CPython 3.14 Windows wheel was built from the checkout. Its 632 entries contained:

• no libRadtran file or binary;
• no CIE spectral-response table;
• no visibility radiance LUT; and
• no Phase 1 lab artifact.

This was an unreleased local wheel audit, not a release-artifact identity. Project dependency metadata was unchanged.

Decisions Established by This Checkpoint

1. The final table cannot use one fixed photon count over the complete domain.
2. A full Cartesian expansion of the candidate envelopes is computationally and scientifically rejected. Before its 81-point spectral dimension it contains 653,184,000 combinations; with that dimension it contains 52,907,904,000 cases.
3. The runtime design must use a source-justified adaptive sparse design and a solver-uncertainty stopping or rejection law.
4. Deep-twilight solar radiance cannot be treated independently of the Phase 4 natural-background boundary. A Monte Carlo non-detection must not be serialized as zero truth.
5. Direct transmission now has an explicit deterministic solver and normalization law, but only as a smoke profile. It is not yet a spectral production grid.
6. Nonzero observer altitude remains unadmitted. libRadtran rejects the generic altitude option with the Monte Carlo solver; using zout alone would leave atmosphere below the observer and would not represent an elevated ground site.
7. Exact geometric horizon remains outside the pilot because libRadtran forbids umu=0.

Open Phase 1 Gates

• identify and validate an elevated-site construction, likely a source-derived atmosphere truncated at the site surface or an independently validated MYSTIC elevation construction;
• expand direct transmission to the admitted spectral design and quantify its near-horizon solver/model error;
• freeze the adaptive sparse sampling and Monte Carlo convergence law;
• determine whether the deep-twilight solar-only tail is generated, upper-bounded, or declared incomplete pending a measured/natural background;
• generate admitted spectral products;
• run the reserved off-grid cases only after the table design is frozen;
• compare storage representations and interpolation laws;
• propagate source-solver, storage, and interpolation error into limiting magnitude and event time;
• build, checksum, license, and validate the separate visibility data pack; and
• add the actual runtime loader only after that pack is admitted.

Phase 2 remains unauthorized until these gates and the Phase 1 exit criteria close.

This site is open source. Improve this page.