Skip to main content
Insights are findings that find themselves. Enfors runs a detection engine on a daily schedule that flags meaningful shifts in behavior — so you catch errors and drift before your customers do.

​
Three tiers of detection

Drift is detected at three levels, each comparing against a different baseline:

Entity

The whole workspace vs. its own history — has the overall behavior of your AI workforce shifted?

Agent

An individual agent vs. its own past — is this agent behaving differently than it used to?

Session

A single session vs. its cohort — is this run anomalous compared to similar ones?
The engine combines several detectors — entity drift, agent shift, session anomaly, outliers, correlation, and graph-structure checks — and applies statistical controls (baselines, false-discovery-rate correction, and fingerprinting) so the findings that surface are real, not noise.
Insights and needs-attention findings

Insights surfaced in Explore — the "needs attention" view

​
Speculation ratio

Beyond standard metrics like confidence and routing, Enfors scores advanced quality signals. The headline one is the speculation ratio. For every decision in a session, Enfors checks whether it was backed by evidence. A decision with no supporting evidence is flagged as a potential speculation. The speculation ratio is the share of such decisions — a direct, structural signal that an agent may be making things up.
Example from the demo: a session with 28 decisions is checked decision-by-decision; any that lack backing evidence are flagged so you can monitor, alert, and investigate.

​
From findings to action

Insights surface directly in Explore and feed the “needs attention” section of generated reports — for example flagging a session as rework. They’re also queryable:

Session health

Metric trends, outliers, and drift indicators for any session.

Outcomes & Intent

The structured dimensions drift is measured against.