Operational evaluation platform
Run ColdStart.
Audit every recovery.
Deaimer’s control platform for qualifying private benchmark tasks, executing reproducible model-and-agent trials, reviewing fairness and publishing signed results.
Benchmark preparation in progress. Public results remain empty until an administrator approves and publishes a signed database snapshot.
PostgreSQL queue Dedicated workers Fairness audit Signed evidence
Control workflow
From private package
to published evidence.
Long-running workloads stay on isolated workers. The web application controls and observes execution without receiving a Docker socket.
Qualify tasks
Immutable ZIP intake, structural and security validation, oracle, NOP, determinism and independent QA.
Freeze systems
Exact model, agent, settings, limits, image and secret references—not raw credentials.
Execute trials
Transactional leasing, heartbeats, preserved attempts and automatic infrastructure retries.
Audit & publish
Fairness decisions, recalculated scores, Ed25519/KMS evidence and approved snapshots.