# Falsification protocol — ADI EM Geometry v1.0.0

This document defines exactly what would refute the construction. If any
external agent (human or machine) computes a value outside the stated
tolerance, the corresponding claim is wrong.

## Procedure

1. Take any positive circumradius `R`.
2. Construct two regular octagons sharing centre and `R`:
   - edge-up: vertex angles at 22.5° + k·45°
   - apex-up: vertex angles at 90° + k·45°
3. Compute the three radii:
   - `r_e  = R · cos(22.5°)`
   - `r_em = R · cos(22.5°) / cos(11.25°)`
   - `r_m  = R`
4. Verify against `tolerances.json`.

## Pass criteria

All 5 checks in `tolerances.json` must pass with the absolute tolerance
specified. Reproduce against every row of `benchmark.csv`.

## Refutation

Any of the following refutes the v1.0.0 claim:

- A check fails on a conformant double-precision implementation.
- An octagonal construction with the same axioms produces a different
  ratio `r_em / r_e ≠ 1 / cos(11.25°)`.
- An n-gon with n ≠ 8 reproduces the electric/magnetic symmetry properties
  claimed in axiom A2.

If you refute, please open an issue against the canonical URL and cite
the version (`1.0.0`) and the check that failed.