RADIUS-CALIBRATED STELLAR INTERVALS — v1.40
Companion to main section 3.18 and S8.10.

MODEL
Compare delta_RA/rho with W/(D*r)=(7/16,18/16,4/16). rho is the
maximum angular separation from the selected hub to its four companions.
Reference radii are fixed at the group-construction epoch; dynamic radii
are recomputed at each evaluated epoch. Units cancel. The count map,
factor D and radius correspondence remain specified structural hypotheses.
This additional cross-channel relation is retrospective.

SCOPE AND RESULTS
The complete nominal Hipparcos/J2000 family contains 19,939 archived
five-brightest circular-field quintets with fixed geometric roles. It is
not all possible five-star subsets. Cygnus has reference-radius rank 1
near 2989.281 BCE (runner-up/Cygnus error 1.31558) and dynamic-radius
rank 2 near 2987.992 BCE. Both branches rank it first in the 283/256
four-arm subsets. The sole superior unrestricted dynamic competitor
has no admissible epoch under the common necessary Stonehenge horizon
condition. Cygnus is feasible at its own minimum, implying local rank 1.
No complete-census local runner-up margin or feasible count was computed.

The v1.40 extension refits every archived cap-centred four-arm candidate
in four nominal catalogue/geometry panels and six targeted HYG-background
state panels. Both radius conventions retain unrestricted and local rank 1
in every panel. In the six targeted panels, the smallest local error ratio
is 7.9983 for reference radii and 5.0357 for dynamic radii. These checks
apply only to the four-arm family, not to the full 19,939-group census.
The state variants are not six independent surveys. No historical
probability, construction-date uncertainty or ancient intention follows.

REPRODUCTION
Python 3.12.14, numpy 2.3.5, scipy 1.17.0 were used. Keep this module
beside 03_Fields: robustness_panels.py imports its archived engine,
catalogues, candidate lists and targeted states. The nominal full-family
scripts use included reference/ and inputs/. No network is needed once
Python dependencies have been installed. Work in a copy, because commands
overwrite result files and runtime timing metadata can change hashes.
Allow several GB of RAM for complete half-year searches.

Run from this directory:
  python radius_normalized.py 1
  python radius_normalized.py 0.5
  python radius_dynamic.py 1
  python radius_dynamic.py 0.5
  python check_invariants.py
  python check_dynamic_invariants.py
  python check_horizon_winner.py
  python robustness_panels.py 1
  python robustness_panels.py 0.5
  python verify_radius.py

OUTPUTS AND VERIFICATION
radius_*_step_*.csv supply every nominal candidate and score.
radius_*summary_step_*.json report nominal ranks and leading fields.
The two *_invariance_check.json files check all fitted scores by a second
vector expression, common rotations, and degrees/radians conversion.
The astrometric engine is shared; this is not an independent astrometry
implementation. horizon_rank_check.json supplies the feasibility argument.
robustness_results/ holds all ten panel CSV files for each search grid,
their twenty objective-panel summaries, and cross-grid checks.
verify_radius.py validates both nominal and sensitivity outputs and writes
verification.json. It checks all per-candidate grid differences; the
archived two_grid_validation.json files record the initial comparisons.
Grid agreement does not certify continuous-time global optimality.

other_geometry_diagnostics.json retains adverse exploratory comparisons:
Cygnus is 2nd in absolute plateau width, 32nd in relative width and 12th
in tree margin in the nominal 256-group panel.

History/ preserves the initial stand-alone audit text and its original
manifest as provenance. Its statements that no sensitivity panels were
run and that v1.39 was unchanged describe that earlier release only.
SHA256SUMS.txt is the current manifest; the historical manifest is not
an instruction to verify the current release against obsolete paths.
