Benchmarks
Every number on this site, with the measurement behind it. Re-run any of them yourself. The numbers live in one file, facts.json, and a scheduled check compares them with the public repo.
| Metric | Evidence | Source |
|---|---|---|
| 1 of 3RUNS ACCEPTED | The same handoff dispatched 3 times, back to back, on real GitHub runners: exactly one ACCEPT flipped to COMPLETED; the other 2 were refused with duplicate_idempotency_key. |
Runs |
| 8 of 8ATTACKS REJECTED | The repo ships one valid signed packet and 8 attacks in examples/v03/packets/. Each attack lands on a named reason code; none pass. |
Examples |
| 1408PUBLIC TESTS | The public checker's test suite passes on main at commit 56c96f9, re-run on October 20, 2026. Command: python3 -m pytest -q. |
Tests |
| Ed25519SIGNATURES | Packets are signed with Ed25519 over domain-separated canonical JSON (synthe/handoff-signature/v1). The playground signs in your browser; the real checker verifies those signatures. |
synthe_crypto.py |
What is not here: throughput or latency percentiles under load. The checker is a local gate, not a hosted service, so there is no fleet to benchmark. If you measure it in your setup, send the numbers and they will be listed with your method.