MIDPOINT ABLATION AND EQUAL-OPPORTUNITY RELATION CONTROL
Stonehenge S10.3 — v1.44 — 30 October 2026

Results
- Original specified-midpoint free-scale model: Cygnus 1/19,939, ratio 2.2234.
- Matched free-midpoint/free-scale ablation: Cygnus 119/19,939.
- Each field chooses among 66 descriptions, plus scale and epoch:
  Cygnus 16/19,939; first within the existing 283-field four-arm subset.
The 66 descriptions yield 62 distinct midpoint values. This is a retrospective
finite-family sensitivity control, not a global statistical significance test.
The original field generator, word, frame and family bounds remain fixed choices.

Reproduce with Python 3.12 and base_model/requirements.txt:
  python build_cache.py
  python audit.py --stage ablation
  python audit.py --stage family
  python verify.py
The cache is about 807 MB; it is generated locally and excluded from this archive.
An existing compatible cache can be supplied as --cache PATH to audit.py.
Five sampled coordinates and the array shape are checked before reuse.
Quick verification of archived outputs requires only:
  python verify.py

The original numerical model and pinned inputs are under base_model/.
All 19,939 ablation scores are retained. A nested numerical lower-bound screen
leaves 280 fields for 66 template fits each; all 18,480 template scores are retained.
Both annual and half-year grids are used over astronomical years -8000 to 2000.
This screen is not an interval-certified continuous global bound.
The independent verifier uses projected vectors and least-squares residuals,
while reusing the original precession engine. Ten noncompetitive evaluations
at orientation discontinuities are checked through one-sided limits.
No new archaeological observation or blind prediction is claimed.
