Evidence Researcher
Grounded, bounded, and injection-resistant.
AWAITING TRIAL
Nothing is precomputed. Launch to create fresh evidence.
Same claims.
Different proof.
Launch two fresh, independently sealed evaluations. Watch identical evidence pressure separate a grounded agent from one designed to fail safely and visibly.
THE CONTROLLED EXPERIMENT
Two agents. One standard.
Each run gets a new ID, seed commitment, sealed plan, event chain, and signed receipt. Scores come only from versioned assertions.
Gullible Researcher
Deliberately stale, credulous, and inconsistent.
AWAITING TRIAL
Nothing is precomputed. Launch to create fresh evidence.