Three binary verifiers: pairwise independence and decision risk
7 October 2026 Pacific (America/Los_Angeles)

A standard finite probability counterexample: class-conditional marginal error0.1 and pair joint error0.01 (error rho0, variance effective count3) coexist with different higher-order laws and balanced-prior Bayes risks3% versus1%, at the same3 abstract query units. Pairwise independence need not mean mutual independence. Exact source/tests/derivation/truth tables are unchanged from the independently reviewed authored diagnostic. No novelty, empirical verifier benefit, code-selection outcome, calibration guarantee or H1 amendment is claimed. The three-verifier result does not show insufficiency of the complete two-judge class-conditional pair law. Read DERIVATION.md for prior and error-versus-label correlation distinctions.

Python standard library only, after extraction:
  python3 -I -B diagnostic.py > reproduced-truth-tables.json
  python3 -B -m unittest test_diagnostic
  python3 -O -B -m unittest test_diagnostic

Compare parsed JSON to truth-tables.json. Seven authored tests verify a101-point rational grid, all eight observation likelihoods, endpoint and mutually independent interior laws, degenerate priors/null cells, nonbalanced-prior Bayes actions and strict schemas. These are exact mathematical checks, not real model measurements. No downloads, model/native-counter calls or paid work occur. manifest.json commits relative distributed files and meaningful reviewed-source hashes; no private operational paths, identities, keys or logs are distributed. ZIP byte identity is deterministic for this exact file set and stdlib compression implementation; scientific results are exact rationals rather than image or platform-dependent numeric estimates.