Sub-Patterns

01

Model Error Recovery

When a model fails

Every AI error must surface the error type, a human-readable cause, and at least one concrete recovery action. Never show a generic error for a model-specific failure. Hallucination, refusal, rate limit, and timeout are distinct states that require distinct UI responses.

Required components

02

Agent Pause & Override

When an agent stops unexpectedly

When an agent pauses due to an error, the recovery surface must show what completed before the pause, what the error was, and what the human can do next — override and continue, roll back and stop, or escalate to senior review.

Required components

03

Confidence Failure Escalation

When model confidence drops below threshold

Confidence failures are not errors — they are governance events. When model confidence drops below the operator-configured threshold, the UI must surface a visual breach state, identify the source of uncertainty, and offer concrete recovery options: send for review, adjust threshold, or override with documented justification.

Required components

04

Scope Escalation Recovery

When an agent exceeds authorized scope

When an agent attempts to access systems, tools, data, or authority outside its approved scope, the system must pause execution and require renewed human consent before proceeding. The scope violation must be named explicitly — not surfaced as a generic error.

Required components

05

Bias & Statistical Risk Alerting

When output involves demographic or statistical risk

When output involves demographic data, low-representation categories, or known bias signals, the interface must surface the statistical risk clearly and provide review or escalation paths. Bias surfaces are governance events, not errors — they require documentation, not just dismissal.

Required components