STONEHENGE S7 — DIMENSIONLESS TIMING AND THE LOCAL HORIZON
Revision v1.37 | 29 October 2026

REPRODUCTION
Use Python 3 with the NumPy and SciPy versions recorded in requirements.txt.
From this directory, run:
    python audit_dimensionless.py
    python verify_release.py
Paths are resolved relative to the scripts, so another working directory is also valid. The audit replaces the numerical files in results/. The verifier checks against verification/expected_summary.json and writes verification/release_check.json. No external data download is needed. Installing Python dependencies is separate from data access.

INPUTS AND PROVENANCE
inputs/hipparcos_panel.csv — archived catalogue entries for the regional five-brightest panels.
inputs/chronology_engine.py — unchanged S5/S6 precession and astrometry engine.
inputs/S6_roles.csv — archived geometric assignments, used only for regression checking after independent recomputation.
inputs/S5_free_scale_scores.csv — 5,280 archived free-scale role-class scores.
inputs/S6_fixed_unit_scores.csv — 5,280 archived fixed-unit role-class scores.
INPUT_PROVENANCE.json records each source location and SHA-256 digest. SHA256SUMS.txt covers the released files before rerunning the audit.

METHOD
All 88 regional groups receive the same geometric assignment rule. No four-arm-tree filter is imposed. The target is the ratio (7,18,4), and the score is the population standard deviation of interval/target divided by its mean. Common changes of angular unit or target scale leave this score unchanged. Searches cover astronomical years -8000 through +2000 (8001 BCE through 2000 CE), using both 0.5-year and 1-year grids, local refinement and endpoints.

For direct local observation, every star must have positive upper culmination at latitude 51.18 degrees N at the same candidate epoch. Every group's score is reoptimized over all numerically detected feasible epoch segments. A group is not simply deleted because its unconstrained optimum fails visibility. Boundary values can be limiting infima. This is a necessary geometric condition, not a model of darkness, extinction, terrain or prehistoric observation. Numerical grids and refinement are not a continuous-time proof.

RESULTS AND DISTINCT COMPARISON UNIVERSES
1. Fixed angular target (7,18,4) degrees, all 5,280 role classes: Cygnus ranks first near 2965.425 BCE. This is exactly the original minute calculation divided by four, including its unchanged 1.4208 runner-up ratio.
2. Free-scale ratios, all 5,280 role classes: 38 classes are strictly better than Cygnus, so its rank is 39. Its optimum is near 2985.482 BCE. Freeing scale changes the objective and channel weighting; it is not merely a change of units.
3. Free-scale ratios with one geometric assignment per group: Cygnus ranks second among 88 groups before the horizon condition, behind Tucana. After uniform horizon-constrained reoptimization, 71 groups have an admissible epoch and Cygnus ranks first. Tucana has none. These ranks answer a narrower question than the 5,280-class comparison.

Cygnus's relative dispersion is 0.649026%. Its best score in the later 2620–2480 BCE sarsen phase is 16.059 times the global minimum. The optimum at 2985.482 BCE lies in the 3000–2920 BCE initial-enclosure interval, but is about 5.5 years older than the separate 2980–2720 BCE 95% ditch interval. Neither epoch is a confidence bound or a direct construction date.

Writing the fixed angular target as W/D degrees with D=4 is an exact retrospective parametrization. It does not independently derive that scale or establish ancient use of a quarter degree. The test does not require the builders to have used sidereal minutes, but historical attribution still requires evidence for the numerical mapping and its transmission.

OUTPUTS
results/all_88_dimensionless.csv — all global and horizon-constrained fits.
results/all_88_checks.csv — assignment, archived-score, grid and unit-rescaling checks.
results/summary.json — full numerical summary, horizons and fixed-angular comparison.
verification/expected_summary.json — released reference summary for reproduction.
verification/release_check.json — structured verification outcome.
verification/reproduction.log — original execution output.

INFORMATION ORDER
PROTOCOL.txt records the retrospective development and the later angular-relabelling audit. The combination was specified after the prior results and preliminary Tucana failure were known. Ranks and error ratios are conditional comparisons, not probabilities of prehistoric intention or independent confirmations.
