bench — benchmarks

Benchmarks that measure AI assistants and implementations against the same yardstick. Problems are deterministic, and answers are verified by the reference implementation (a self-check without an LLM).

Bench	What it measures
AI-assist bench v4 — unstated premises	Whether an AI assistant declares the premises a requirement leaves unstated, instead of filling them in silently. 21 problems, answers, grading rules and runner (published 2026-11-04)