Illustrative case / evidence graph
From AI incident to a defensible decision.
See how your team establishes what happened, assesses the exposure, and records the decision with supporting evidence and a clear account of what remains uncertain.
Follow the evidence
Examine the activity behind the claim.Conceptual illustration
Two sources, one question
Did the observed effect fit the task?
Synthetic example. The graph illustrates an attributed case; the live product must check links and provenance before making the same claim.
Agent reported
- 01User taskSummarize a support ticket
- 02Tool claimOpen customer record
- 03Reported resultSend brief to support lead
Host observed
- 01ProcessBrowser context spawned
- 02FileLocal export.csv read
- 03NetworkOutbound payload posted
Attribution check
Compare links, not timestamps alone
Known IDs and process ancestry can support a chain. Unmatched activity stays visible as context, not promoted into a causal claim.
Finding for review
Reported brief vs. observed outbound transferCase walkthrough
From detection to a defensible decision.
Select a stage to see what the investigator can verify in this synthetic example.
Stage 01
Start with the finding and its source.
An endpoint finding points to its sealed package. A session rule points to its triggering agent event. The distinction is visible before an analyst begins the case.
What stays checkable
- Finding type and severity
- Endpoint or session source
- Evidence package or event reference
The boundary matters
Evidence, inference, and verdict are different things.
Same time is not same cause
Co-occurring host events remain context until a supported link is found.
No claim is still evidence
Endpoint findings can reach the alert queue even without an instrumented agent session.
A candidate is not an identity
A model connection identifies a process candidate; a named agent needs a trusted binding.
Start a focused conversation
Bring your AI ambition.
Let’s discuss the accountability.
Tell us which agents touch your customers, data, or operations. We’ll discuss the evidence your teams need to evaluate AgenTrace.
- Your highest-priority AI workflow
- Your customer and compliance questions
- The sources needed to investigate it
Early access · Talk directly with the team