AutoFlow
Benchmarks
Engineering stress evidence—published with methodology.
These results demonstrate runtime reliability and fail-closed behavior. A real comparative benchmark is still required to quantify hallucination reduction.
Demonstrated now
Approved claims only
12
Rules
AutoFlow currently supports 12 deterministic financial and reconciliation rules.
1,000/1,000
Jobs
In a synthetic no-network stress test, 1,000/1,000 jobs completed successfully.
Byte-identical
Replay
The tested workload produced byte-identical replay artifacts.
100%
Schema
Certificate-schema validation in the tested tooling; gateway exercised under simulated 429s, timeouts, retries, and invalid outputs.
Methodology
How to read these results
- 01Stress jobs were synthetic and no-network for the 1,000/1,000 completion result.
- 02Replay identity refers to byte-identical artifacts for the tested workload given the same evidence, rule version, and claim.
- 03Gateway resilience covers simulated rate limits, timeouts, retries, and invalid outputs—not live production SLA claims.
Not yet published
Comparative hallucination-reduction benchmark
AutoFlow is designed to prevent unsupported financial claims from being released; a comparative real-model benchmark is in progress.
Do not treat stress results as quantified hallucination reduction.
Rule coverage detail: Verification Library
Validate against your own package.
Design partners bring a sanitized or synthetic recurring workflow. We measure coverage, accuracy, and review-time impact.