Prepare the case
Capture company records with their sources and retrieval times. Add labelled simulated requests and agree the expected behaviour.
Get API keyAgent Evaluations · Enterprise pilots
Test the decision.
Inspect the evidence.
See how your AI investigates companies, handles missing records and follows approval rules before it acts in your business.
Already have pilot access? Open Agent EvaluationsIllustrative case · Fictional companies
The name matches.
The company does not.
A supplier asks your agent to update a record. Two businesses have the same name, but different jurisdictions and registration identifiers.
A correct investigation identifies the conflict, preserves the approved record and requests clarification.
- Approved supplier
- Northbridge Components Limited
GB · TEST-GB-001 - Simulated change request
- Northbridge Components Limited
IE · TEST-IE-002 - Behaviour to evaluate
- Distinguish the entities before updating the supplier.
This is an explanation of a test case, not a measured agent result.
Your agent. A repeatable investigation.
From a business request to a useful finding.
Run your agent
Connect through the evaluation API. Your agent searches frozen evidence, inspects documents and attempts actions inside the simulation.
Review and retest
Inspect the investigation and decision. Change your agent, repeat the test and reserve unseen cases for a final evaluation.
One product family. Two jobs.
Verification & monitoring
Check supported company claims during real work, preserve evidence and track subsequent changes.
View verification plansAgent Evaluations
Test an agent’s investigation, tool use, decisions and simulated actions before release and after updates.
Scope an evaluationSeparate enterprise add-on
Agent Evaluations
Test how your AI investigates companies, uses evidence and follows approval rules before deployment or after an update.
- Private scenarios and fixed evidence snapshots
- Recorded investigations and simulated actions
- Agreed review criteria, findings and a retest
Quoted per pilot
Scope, scenario review and fees are agreed separately. Verification plans and claim credits do not include agent evaluations.
Discuss a pilotExplore Agent EvaluationsWhat is available today
The workspace supports company-data capture, fixed evidence, organisation access, independent scenario review, agent suites, version comparisons and downloadable reports. A library of 30 synthetic drafts is available for review. Captured real-data cases remain unscored until their expected outcomes receive expert approval.
Business actions and bank checks are simulated. The environment evaluates your agent; your team uses the findings to improve it. Sign in with your CompanyProof account. Evaluation access is activated for your organisation separately from verification plans. External agents receive limited, expiring test tokens from that workspace.
The Grounding Index covers factual grounding methodology. Agent Evaluations extends testing to decisions and actions.