STONEHENGE S8 — GEOMETRIC CONTROLS AND A FIELD CENSUS
Companion to manuscript v1.40, 29 October 2026

The first three modules integrate completed retrospective analyses. Their
original code, input files, full outputs, manifests and reproduction records
are copied byte for byte. The original Portuguese README files describe
their state before integration; statements that the manuscript was unchanged
refer to those earlier releases. The English S8 appendix and v1.40 manuscript
are the current integrated interpretation.

EXECUTION
Install the versions declared in the modules' requirements.txt files.
The tested environment is Python 3.12 with NumPy and SciPy. Scripts need no
network access after dependencies are installed.

Run within 01_Invariance:
  python audit_robustness.py

Run within 02_Axes:
  python audit_axes.py
  python audit_kinematics.py

Run within 03_Fields:
  python all_sky_caps.py
  python kinematic_fields.py
  python diagnose_inverse.py

The first script in each module is independent of execution of the other
modules; all necessary input bytes are included. Preserve the stated order
inside each module. Field enumeration needs approximately 210 MB for a
5,100-by-5,100 distance matrix, plus other working data.

SCOPE
01_Invariance uses the S7 regional panel: 88 geometric assignments, with
71 having at least one epoch admissible under the necessary horizon rule.
It supplies the vector form of right-ascension intervals, relative-loss
identity, conditional Lp and weight bounds, all target-order permutations
and the continuous two-side spatial ablation. Its rank-margin theorem does
not automatically transfer to the new candidate family.

02_Axes uses great-circle distances in the same regional panel. It includes
the near equality, all fifteen pair comparisons, epoch freedom, homogeneous
HYG comparison, six Cygnus state variants and failed direct Station-rectangle
templates. The near equality is not a second clock confirming 2985 BCE.

03_Fields enumerates every five-brightest set attained in a cap centred on
a catalogue star while that star remains in the set. Only sets whose cap
centre is the unique four-arm MST hub are retained. No constellation
boundary, target ratio, archaeological date or horizon enters generation.
The four-arm family was chosen retrospectively with Cygnus known, and is
neither all five-star subsets nor a superset of the earlier regional family.
The target 7:18:4, geometric roles, full time window and relative loss remain
specified choices.

Four nominal panels recover Cygnus first, globally and after separate
uniform horizon reoptimization. The unrestricted nominal margin is narrow.
All six targeted kinematic recensus variants preserve local first place;
global first place is lost in three. Inverse-construction and withheld-channel
controls include their adverse outcomes. No claim of historical intention,
archaeological dating precision or statistical significance follows from
these ranks or error ratios.

VERIFICATION
INPUT_ARCHIVES.json records original ZIP hashes and exact copied-file counts.
SHA256SUMS.txt covers all files beneath this root, excluding itself. Modules 01–04 retain their prior manifests; module 05 has a current manifest
and explicitly historical initial-audit documentation. Clean-extraction reproduction
records are included: the 03_Fields record verifies all 26 outputs byte for
byte. Integration validates archived bytes and document claims, and does not
claim a new complete rerun of S1–S7. See the v1.39 and v1.40 execution extensions below.
Catalogue provenance and the HYG licence are preserved within inputs/.

Navigation: main sections 2.9 and 3.16–3.18; appendix S8 sections 1–10.
The earlier S7 information-order reference 2.9 is now main section 2.10.

V1.39 EXTENSION — 04_FixedScale
The first three modules above remain unchanged. This fourth module adds the
complete nominal Hipparcos/J2000 census before any tree or horizon filter,
with fixed target (7,18,4) degrees and a matching free-scale comparison.
Keep 04_FixedScale beside 03_Fields, whose catalogue, precession engine and
selector it imports. From 04_FixedScale run:
  python audit_open_fields.py 1
  python audit_open_fields.py 0.5
  python verify_results.py
These commands rerun enumeration and overwrite their own result files;
use a copy to preserve release hashes. Allow several GB of RAM.

Cygnus ranks first among 19,939 groups under the fixed angular target, with
runner-up error 2.201 times larger. Fixed-scale horizon reoptimization retains
first place among 17,860 groups (margin 2.626). Free-scale rank is 68 in the
complete census and first in the 283 and 256 four-arm subsets. Section S8.9
and main Table 7 distinguish these objectives and candidate families.
The earlier catalogue/kinematic robustness results belong to the ratio
family in 03_Fields and are not transferred to the complete fixed ranking.
The old modules' execution records are retained; v1.39 additionally reruns
03_Fields (all 26 output bytes identical) and executes both complete grids.

V1.40 EXTENSION — 05_RadiusCalibration
The first four modules remain byte-identical to v1.39. The fifth module
tests interval/radius against W/(D*r), with both reference and dynamic
hub-centred radii. Its README gives commands, input dependencies and scope.
It includes both-grid complete nominal results, vector/unit checks, the
necessary-horizon rank argument, and newly refitted four-arm catalogue/
state sensitivities on both grids. Keep all five module directories together.

Reference-radius Cygnus rank is 1/19,939; dynamic-radius rank is 2/19,939.
The sole better dynamic candidate is infeasible throughout the search at
Stonehenge, while Cygnus is feasible at its minimum, establishing local rank
1 without computing a full feasible-group count or constrained margin.
Both radius branches retain rank 1 in every tested cap-centred four-arm
catalogue/state panel, globally and locally. These sensitivity results are
not transferred to the complete all-field census. The added relation is
retrospective and does not replace the earlier adverse free-scale controls.
The integration does not claim a new execution of S1–S7 or modules 01–04.
