Proof + simulation: 20,138 command sequences verify a bank model with zero violations
samgoodwin89 · x · 2026-11-19
samgoodwin89 shows an Alchemy verification report for a Bank model: 6 invariants mathematically proved for all states and inputs, exhaustive checking of 20,138 sequences (9,725 states, up to 2 commands with duplicated events, reordered reactions, twice-run saga steps) covering money-loss and pending-transfer rules, plus 200 randomly sampled sequences up to 12 commands long — none breaks a rule. Mixing proof checking and simulation yields a high baseline of confidence without hand-written tests.
More from coding & agent
- Open-Source EggLab Brings Local Image Gen, Editing and an AI Design Agent to PhotoCraft — AdditionCorrect2929 · 2026-11-19
- 40+ Free AI Learning Resources: Videos, GitHub Repos and a Full Roadmap — nikola_mr64990 · 2026-11-19
- Alchemy merges Z3-based formal verification: declare invariants, get proofs of correctness — samgoodwin89 · 2026-11-19
- One test to prove your model never violates its invariants, via Z3 — samgoodwin89 · 2026-11-19
- Who Authorized the Agent? Reddit Debate on Proving AI Agent Actions in Court — icccpda · 2026-11-19
- GPT 6.2 Astra builds interactive 3D web app of Singapore Chinatown in one shot — OpenAIDevs · 2026-11-19