What does the evidence support?

The research changed the question.

1. R-01A · dGR203

   The first hypothesis was not supported.

   In the evaluated seasonal pilot, nominal terrain and all five official clones retained the historical 6.5-day simultaneous-center classification. The separate Product 104 check also retained it.

   The negative result concerns this decision and tested configurations. Five local clones do not cover all terrain uncertainty or calibrate a probability. The selected fixed-start grid is a separate question.

   Read the findingRead the retained protocol

2. R-01B · dGR203 / dGR101

   The evidence changed the question.

   Small tested terrain effects at dGR203 did not generalize to dGR101. A separate nominal screen revealed a large response to a synthetic terrain representation at dGR101.

   The historical seven-versus-8.47-day discrepancy was partly reproduced numerically; its original cause remains unassigned. The observer convention also changed some threshold outcomes.

   Read the findingRead the retained protocol

3. R-01C · dGR101

   Native NASA products retained the response.

   The large complete simultaneous-visibility response survived native NASA terrain products at a common absolute observer. Synthetic averaging alone does not explain it.

   Removing reporting clipping changes the earlier threshold interpretation. Different product lineages, observer conventions and residual fine-grid differences prevent a terrain-truth or achieved-convergence claim.

   Read the findingRead the retained protocol

At dGR101, native Product 78 5 m and Product 90 80 m give 7.40 and 12.92 days of complete simultaneous visibility at the same absolute observer. That is a representation response, not an accuracy ranking.

Negative evidence / dGR203

A valid audit can preserve the decision.

The first local terrain-uncertainty hypothesis was not supported for the evaluated 6.5-day simultaneous-center decision. Duration changes still occurred. Inspect the limited pilot and the later modeling checks in their own domains.

Verifying the dGR203 control evidence…

Method & assumptions

Hold the question still. Change one modeling choice.

1. Record the question and controls. Retain the protocols, site, observer convention, visibility criterion, date domain and research benchmarks before interpreting outcomes.
2. Use traceable terrain representations. Compare native NASA grids within one common 20 × 20 km local footprint. Keep the 80 m / 240 m exterior continuation fixed. Pixel spacing does not mean uniform pole-wide coverage.
3. Construct the geometric horizon. Use bilinear DEM interpolation, spherical curvature and sampled horizon rays. Geometric visibility requires a body’s center above the represented skyline; it is not a power or radio-link prediction.
4. Refine event boundaries. Keep reporting-domain clipping separate from complete transition-bounded episodes intersecting that domain. Preserve the late solar episode separately from the longest solar episode.
5. Test numerics and retain counterexamples. Compare spatial, temporal and range refinements before interpreting model effects. Preserve the negative result and unresolved historical discrepancy.

Frame / ephemeris

MOON_ME_DE421 / DE421

SPICE aberration convention

NONE · geometric centers

Lunar reference sphere

1,737,400 m radius

dGR101 source point

-88.5134116385092° latitude, -68.01460926527753° east longitude

dGR203 source point

Published rounded point: 88.141° S, 71.696° W. Separate from the archived dGR101 point.

R-01C observer conventions

Common absolute elevation, or exactly 2 m above each represented ground. The latter also updates celestial observer position.

Registered domain

[2026-11-06 00:00, 2026-11-28 00:00 UTC)

Complete-episode diagnostic envelope

[2026-10-19 00:00, 2026-12-13 00:00 UTC). Only episodes intersecting the registered domain enter the reported complete results.

The archived protocols document pre-outcome controls. This page does not claim third-party preregistration certification. The 6.5-day simultaneous benchmark and 10-day solar screen are separate historical research criteria.

Saved scientific checks

Numerical agreement has a defined scope.

NASA data & provenance

Different representations. Traceable bytes.

Native NASA grids, the separate SDEM branch, and the synthetic control keep distinct identities. Raster hashes identify the evaluated bytes. Native-window and range hashes do not establish a whole remote-file digest.

Outside the replacement footprint: Unchanged native 80 m then 240 m nested continuation; missing data never zero. Original source pixels are retained without resampling into a new allegedly native grid. Native product lineages and alignments differ.

Interpretation limits

A precise number is not necessarily a stable decision.

• Representation response: different products and observer conventions can change a modeled answer. No product is designated terrain truth; finer pixel spacing does not certify accuracy.
• Scoped stability: dGR203’s tested decision can remain unchanged while durations and Earth-only results change. This does not establish stability at dGR101 or every lunar site.
• Censoring: the roughly 5.4-day registered nominal value is clipped. It must not become a claim of a physical 6.5-day mission failure; its complete R-01C simultaneous episode is about 7.40 days.
• Limited uncertainty coverage: the first-five local-clone pilot is not a calibrated distribution. Exterior terrain, shared biases and all systematic errors are not sampled.
• Geometry only: geometric center visibility omits mission power storage, radiance, eclipsing-source effects and communications link budgets. Finite-disc criteria in R-01B are changed observables with their own limits.
• Historical replication: 12-hour increment counting explains part of the seven-versus-8.47-day gap; a one-day residual and inconsistent archived raster semantics remain unresolved. A diagnostic that produces seven days is not proof of the original authors’ implementation.
• Product verification: automated checks do not establish terrain accuracy, accessibility certification or human comprehension. The uncoached twenty-second judge test and dedicated screen-reader review remain outstanding.

Reproducibility

Follow a claim back to its source.

The review packet lists frozen source hashes, exact downloads, numerical summaries, raster identity and kernel inputs. It supports checking the displayed evidence. It is not the complete research environment or the original DEM collection.

1. Inspect without rerunning science. Compare the exact CSV row or JSON pointer with the displayed value. Check the downloaded file’s SHA-256 against this packet. Retained hashes check byte identity, not independent physical validity.
2. Rebuild the product from retained evidence. In a complete research workspace, install the pinned web dependencies, run the build, and serve the static export. The build verifies the frozen baseline and packages derived display records. It does not rerun terrain horizons or event solutions.
3. Reproduce the research separately. Use a separate workspace copy with the verified raw rasters, kernel inputs, research environment and retained R-01A/B artifacts. Follow the original R-01C instructions; heavy raster processes run sequentially. Full raw data and environment files are required beyond this review packet.

npm ci
npm run build
npm run preview

# Verify retained evidence and product behavior
npm run verify:science
npm test
npm run test:e2e