8 projects
protocol-bench
Ground-truth safety verdicts for 15 published IEEE 802.11 and 3GPP procedures, with machine-validated counterexamples.
failclosed
Default-deny ASGI middleware: a gated endpoint succeeds only on an affirmative machine-checked verdict.
polyfrac
Exact polynomial and rational-function arithmetic over Q, with Sturm-sequence real-root counting. No floating point.
minicheck
An explicit-state model checker in ~2900 lines. Shortest counterexamples, no required dependencies.
evidence-runner
Run the verification portfolio over your repo and get ONE verdict — the weakest leg, never the mean
certhead
An LLM output head that PROVES its token instead of guessing it, and falls back to dense rather than be wrong
abstain-bench
How often does a verifier claim success on input it cannot check? A benchmark for unearned passes.
tokencount-verify
A token count both parties can recompute. Refuses a claimed count that arithmetic rules out.
