Core concepts - Failproof AI

Concept	What it means	What you can achieve
Session	One agent task or run	Reconstruct an individual outcome
Event	One recorded action in a session	Inspect a model call, tool use, error, human action, or policy decision
Trace	The ordered and nested session view	Understand causality instead of reading disconnected logs
Evaluation	A score or judgment on a session	Track quality, compliance, cost, or latency continuously
Audit	A review of a selected session population	Search for failure patterns with a stated goal and cadence
Finding	An evidence-backed failure discovered by an audit	See the failure mode, severity, and affected sessions
Issue	A durable response record	Assign, discuss, and resolve a finding or alert incident
Alert	A rule that detects recurrence	Notify responders when a known condition returns
Policy	A rule evaluated during agent activity	Observe, block, or redirect risky behavior
Deployment	A versioned policy rollout to machines	Control where a policy runs and roll it back safely

The reliability loop

Start from evidence. A session shows what happened. An audit determines whether it is an isolated mistake or a pattern. A finding identifies the failure mode; an issue owns the response. A policy prevents the same behavior, while alerts tell you if the condition returns.

A local failproofai audit scans local agent history. A recurring Cloud audit reviews sessions stored in Failproof AI Cloud. They are separate workflows with different scope and scheduling.

​The reliability loop

The reliability loop