Physical Heliacal Visibility Phase 4 Jones MYSTIC Holdout Checkpoint

Date: 2026-07-31

Status: corrected-v2 sealed 550 nm geometry holdout gate closed; lower-boundary decision subsequently closed on 2026-08-02; runtime model not admitted

Candidate identifier: jones_paranal_scattered_moonlight_2013_v1

Pilot identifier: jones_paranal_mystic_550nm_pilot_v2

Outcome

All three fresh geometry combinations reserved before the corrected-v2 pilot threshold freeze passed the frozen 550 nm holdout contract. A fourth execution repeated the interior cross-axis holdout with the same seed and was byte-identical across all five scientific repeat files. An independent validator reconstructed the complete input contract and reproduced every reported metric.

This closes the corrected bounded geometry holdout gate. It does not admit a spectral grid, interpolation surface, runtime model, production data pack, public API, or network dependency.

Correction Boundary

The original v1 pilot, threshold, and holdout evidence was invalidated after the lower-boundary investigation exposed incorrect libRadtran explicit-aerosol layer ownership. V1 listed physical files at upper rather than lower layer boundaries. The exact v1 receipts remain preserved in phase4_jones_mystic_v1_invalidation_checkpoint_2026-07-31.json, but none of their numerical results is admissible.

The replacement protocol used a corrected v2 pilot, newly frozen v2 threshold checkpoint, three previously unexecuted v2 holdout geometries, and a new seed. Both the artifact and compact checkpoint bind the layout top_marker_then_null_gap_then_layer_files_at_lower_boundaries.

Freeze Ordering

The ordering is machine-bound:

1. Fifteen corrected-v2 pilot cases were executed and independently validated.
2. V2 numerical thresholds were frozen and committed to a canonical receipt.
3. The threshold checkpoint stated that no v2 holdout had been executed or used to select a threshold.
4. Only then did the runner derive four executions from the three reserved v2 geometries and exact-repeat protocol.

No absolute holdout radiance was prefrozen. The gate tested finite positive output, the prefrozen relative Monte Carlo uncertainty ceiling, and exact fixed-seed repeatability.

Frozen Protocol

Each unique holdout used:

• wavelength: 550 nm;
• photon count: 1,000,000;
• random seed: 271828183;
• maximum relative standard error: 0.005; and
• the same checksum-bound libRadtran 2.0.6 MYSTIC, ESO source members, and corrected explicit atmosphere/aerosol construction as the v2 pilot.

holdout_v2_interior_cross_axis was executed a second time with identical inputs and seed.

Results

Holdout	Radiance (W m-2 nm-1 sr-1)	Standard error	Relative error
holdout_v2_low_altitude_oblique	3.09175e-08	3.71113e-11	0.0012003331
holdout_v2_interior_cross_axis	7.88475e-09	1.97965e-11	0.0025107327
holdout_v2_phase_boundary_waxing	9.54726e-09	2.84338e-11	0.0029782157

The worst relative error was 0.0029782157, below the frozen 0.005 ceiling. All radiances and standard errors were finite and positive. The repeat was byte-identical for mc.rad.spc, mc.rad.std.spc, mc0.rad, mc0.rad.std, and randomseed.

Artifact and Tool Receipts

The corrected-v2 external artifact manifest is locked at:

• bytes: 30349; and
• SHA-256: d01f9d4a984c011d04b2d3cd2d3bde6bfeb426db4ae8bf030537b7f55777b321.

The committed canonical checkpoint is 6,058 bytes with SHA-256 5038773b2a2ea05cf07887f0acd45abe5d2e3b8503f81523ff6f35de3421ba4b. It binds:

• holdout builder SHA-256: 03debaad43172953eb1cd2586739e18d27ba7cdbcc20ca12256b444c223cd9cc;
• holdout validator SHA-256: 304a94c2729e3dfcd058a0f009b482dc9f7512b9bb6adf44fc301afc51d5aa15;
• corrected pilot builder and validator receipts;
• the exact v2 threshold specification and pre-holdout checkpoint;
• the v1 invalidation lineage;
• the libRadtran source archive, binary, build files, and data tree; and
• external ESO identities without redistributing their bytes.

Independent Validation

The holdout validator independently derives the reserved-case execution matrix, verifies every file receipt and run-directory inventory, reconstructs the corrected explicit aerosol profile through the separately bound pilot validator, parses MYSTIC outputs, recomputes all frozen checks, and validates the compact checkpoint. Its final status is valid_corrected_v2_holdout_evidence.

Explicit Remaining Limitation at This Checkpoint

The atmosphere and ground still begin at the 2,640 m observer altitude, whereas the Jones construction uses a 2,000 m lower model boundary beneath the observer. The successful corrected-v2 pilot and holdouts do not erase this difference. The next gate must reconstruct and validate that lower layer or explicitly retain a separately versioned observer-bottom model before spectral admission is designed.

That follow-on gate is now closed by the Jones MYSTIC lower-boundary checkpoint, which selects and independently validates the source-faithful 2,000 m lower boundary with the observer retained at 2,640 m. This historical holdout receipt remains the exact observer-bottom regression control.

Machine-Checkable Files

• scripts/build_visibility_phase4_jones_mystic_holdouts.py
• scripts/validate_visibility_phase4_jones_mystic_holdouts.py
• tests/artifacts/visibility_reference_lab/phase4_jones_mystic_holdout_checkpoint_2026-07-31.json
• tests/artifacts/visibility_reference_lab/phase4_jones_mystic_v1_invalidation_checkpoint_2026-07-31.json
• tests/unit/test_visibility_phase4_jones_mystic_holdouts.py

Reproduction

python3 scripts/build_visibility_phase4_jones_mystic_holdouts.py \
  --uvspec /home/nilad/.cache/moira/visibility-reference-lab/work/libRadtran-2.0.6/bin/uvspec \
  --lib-radtran-archive /home/nilad/.cache/moira/visibility-reference-lab/source/libRadtran-2.0.6.tar.gz \
  --data-root /home/nilad/.cache/moira/visibility-reference-lab/work/libRadtran-2.0.6/data \
  --eso-archive /mnt/c/Users/nilad/AppData/Local/Temp/moira-jones-20260731/SM-01.tar.gz \
  --output-root /home/nilad/.cache/moira/visibility-reference-lab/work/jones-mystic-holdouts-v2-20260731-1

python3 scripts/validate_visibility_phase4_jones_mystic_holdouts.py \
  --artifact-root /home/nilad/.cache/moira/visibility-reference-lab/work/jones-mystic-holdouts-v2-20260731-1 \
  --uvspec /home/nilad/.cache/moira/visibility-reference-lab/work/libRadtran-2.0.6/bin/uvspec \
  --lib-radtran-archive /home/nilad/.cache/moira/visibility-reference-lab/source/libRadtran-2.0.6.tar.gz \
  --data-root /home/nilad/.cache/moira/visibility-reference-lab/work/libRadtran-2.0.6/data \
  --eso-archive /mnt/c/Users/nilad/AppData/Local/Temp/moira-jones-20260731/SM-01.tar.gz

Gate State

Closed:

• v1 invalidation with exact historical receipts;
• corrected lower-boundary ownership in the 550 nm pilot profile;
• v2 threshold freeze before any replacement holdout execution;
• all three fresh v2 geometry combinations;
• finite-positive output and frozen uncertainty checks;
• exact fixed-seed repeatability; and
• independent artifact and metric validation.

Still open:

• spectral wavelength/grid selection and interpolation thresholds;
• untouched spectral holdouts and response-integrated products;
• release/distribution disposition for any derived data artifact;
• engine implementation and typed component receipt;
• public-surface parity; and
• Phase 4 closure.