Components › Human Control & Oversight › Confidence Threshold
HC10 · Transparency

Confidence Threshold

Purpose A human-in-the-loop gate for AI automation pipelines. When model confidence falls below a configured threshold, this component intercepts the pipeline, surfaces the specific source of uncertainty, and requires a user decision before automation continues. The resolution UI is inline — the user diagnoses the problem and resolves it in context, without navigating away. This is not a confidence meter; it is a gate. See also HC01 for general human-in-the-loop controls.
Use when An AI pipeline would automate a downstream action — write, send, transform, classify — and confidence is both quantifiable and meaningful to the resolution; the source of uncertainty is something a human can diagnose and decide (ambiguous field mapping, conflicting data, policy edge case); a documented human decision is required before automation proceeds; the resolution can be completed in-context rather than requiring navigation to a separate screen.
Do not use When the automation has no downstream consequence that warrants interruption; when confidence is model-estimated rather than grounded in measurable uncertainty; when the resolution requires tools or context the user does not have at this point in the flow; as a passive confidence display (use TD04 for that); in consumer-facing UI where the concept of a threshold has no user-legible meaning.
Failure mode prevented Low-confidence automation proceeding silently without a human checkpoint, producing incorrect outputs that downstream systems treat as authoritative. The inverse failure: surfacing a threshold breach as a generic error message with no resolution path, forcing the user to abandon the task and seek help externally.
Example Dark Academia
Dark Academia
The Atelier
Neon Tide
The Archive
Candlelight
The Conservatory
Overview
Theming API

Override these CSS custom properties to apply your own visual treatment without touching the component's behavioral contracts.

Shared semantic tokens

Token Controls
--rad-surface-card Card background
--rad-color-text-secondary Reason body text
--rad-radius-card Card border radius

Component-scoped tokens

Token Controls
--rad-threshold-color Label, score, fill — the primary alert color
--rad-threshold-border Card border
--rad-threshold-header-bg Header band background
--rad-threshold-header-border Header band bottom border
--rad-threshold-config-color Supporting config text (confidence values)
--rad-threshold-marker-color Threshold marker line
--rad-threshold-track-bg Progress track background

"*The threshold breach is the headline. The number is the evidence. Never reverse these.*"

← Previous Uncertainty Indicators Next → Algorithmic Nudge Disclosure

Intellectual Property

RAD is the original work of Jackie Curry. All rights reserved. No portion may be reproduced, adapted, or incorporated into any product or system without express written permission.

Permitted: citation in academic or editorial contexts with full attribution.

Licensing

© 2025 Jackie Curry. All rights reserved. Publication date: 2025.

For licensing inquiries, connect on LinkedIn →