This is the exact decision logic MandateGuard runs — in your browser, no LLM involved. Same inputs → same verdict, every time. The ledger below is chained like the real thing.
Budgets, allowlists, denylists, rate limits, and signed Ed25519 mandates evaluated by a pure function — with zero LLM sampling in the path. Try the loop: approve 0xGOOD until the per-window budget is spent, then watch the same call get DENIED. A prompt injection cannot change that verdict.