Research NoteJune 25, 20253 min read
Managing Randomness in Sandboxes
Why `Math.random` breaks receipt verifiability.
Problem
If an execution policy relies on unseeded random number generation (e.g., Math.random()), replaying the policy against the input will yield a different outcome. This breaks offline verification of EvidencePacks because the hashes will diverge.
Approach
All execution policies must be strictly deterministic. If pseudo-randomness is required for a heuristic algorithm, it must use a Deterministic Random Bit Generator (DRBG) seeded explicitly by the prevHash of the receipt chain, or by a cryptographically signed entropy injection proposed prior to execution.
Invariants
- Execution environments must hook and block unseeded RNG syscalls.
f(state, input) == f(state, input)perfectly, infinitely.
Artifacts
References
- NIST SP 800-90A Rev. 1 (DRBG mechanisms).
Pesquisa Mindburn Labs • June 25, 2025