- Who it helps
- Developers designing agent evaluations and completion checks.
- Availability
- The paper and MIT-licensed implementation are public. The full verifier needs Python 3.11 or later and a compatible model endpoint; provider charges apply.
A receipt for each verdict
Related criteria share an investigation. The verifier records supporting evidence so a reviewer can trace a judgment back to its source.
Evidence-First Agentic Verification ↗Sauron implementation and setup ↗
What the study measured
The authors report 0.96 F1 versus 0.92 for their strongest baseline across 2,586 labeled criteria from 100 assistant episodes, with lower cost in that configuration. This measures agreement on those cases.
What you can do
- Use the linked evidence-receipt exercise to practice separating a completion claim from its saved output. Open reference ↗
Where the evidence stops
Benchmark results are author-reported. The original helper in the guide was tested separately; Sauron itself was not run.
Sources read. Sources checked October 20, 2026 UTC. No external product workflow, model training or inference was executed.
Practical guide
Check an agent’s completion claim against saved evidence →
A finished review receipt identifies which claims the saved evidence supports, which it contradicts and which remain unknown.
Sources & further reading
Paper · paper
Evidence-First Agentic Verification ↗
Original research statement for the method, reported results and evaluation scope discussed in this article.
Sanayei and colleagues · page checked 2026-10-20 · Published 2026-10-13
GitHub · implementation
Sauron implementation and setup ↗
Project-owned implementation or interface showing availability, requirements and practical constraints.
Sanayei and colleagues · page checked 2026-10-20
Coverage note: The source bundle is a bounded sample. Independent replications and matching original discussion across all platforms were not established.
The record over time (1)
Dated source checks and published explanations. An observation date can differ from the original release date.
- · Added to Current
Sauron investigates an agent’s execution record and final artifacts before judging whether its task succeeded.