ExperTrace™ · outcome evaluation
Find out which of your agent's calls were wrong before your customers do.
Your agent made the call. ExperTrace tells you whether it was right: it connects each judgment to what actually happened next, scores it, and feeds the result back.
what the agent decided.
whether that decision was actually right.
How it works
Outcome-scored judgment, not trace logs.
Observability shows what an agent did. ExperTrace adds the part that matters: whether it was right, measured against what happened next.
For agent teams, today
Production evaluation from your own decisions.
When Flow Resolver judges an escalated action, it returns a request ID. When the result is observable, report CORRECT, WRONG or PARTIAL against it. Reporting is free. ExperTrace scores the call against its falsifier and feeds the result back.
- Measure decisions against actual outcomes
- Find where judgment fails in production
- Build your own outcome record as you operate
curl -X POST https://resolve.flowinfo.co/api/v1/outcome \
-H "Authorization: Bearer fjp_live_..." \
-H "Content-Type: application/json" \
-d '{"request_id":"<from the judge response>",
"outcome":"CORRECT"}'A SCORED RECORD (ILLUSTRATIVE){
"signal": "Agent proposes a $12,000 refund; authority is $5,000",
"judgment": "Refund exceeds autonomous authority; approval absent",
"action": "ESCALATE",
"falsifier": "Manager approval recorded before execution",
"request_id": "req_...",
"outcome": "CORRECT"
}Flow's scoring and calibration methods remain proprietary.
Already running inside Flow
Scored in production, not in a lab.
Flow built this loop for its own judgments first. A live cohort generates judgments, actions, outcomes and corrections, and outcomes are scored nightly.
Confirmed, partial and wrong, once outcomes became observable. The first call came 13 days before the first public IPO report.
Restraint held as pressure rose. Netflix walked away with a $2.8 billion breakup fee and its shares rose more than 10%.
Talk to us
Score your agent's decisions.
Tell us what your agent decides and when the outcome becomes visible.
Also by request, for AI labs: outcome-scored judgment traces at $0.15 per trace (50,000 minimum).