Confidence rubric

AI signal confidence should be earned, not guessed.

VexASI scores source-grounded records by evidence strength, specificity, recency, false-positive risk, review state, and action readiness.

Signal Quality

The score explains why the record should or should not move forward.

A confidence label is only useful when the reviewer can see what drove it. VexASI keeps the source trail and the quality factors attached to the record.

Factor Strong Signal Weak Signal
Source strength Primary public source, direct quote, named entity, clear date. Secondhand claim, vague article, no direct evidence.
Specificity Named role, tool, project type, facility, procurement clue, or delivery change. Generic language such as innovation, transformation, or AI-enabled with no concrete detail.
Recency Recent enough for the workflow's decision window. Old or undated source with no current activity marker.
False-positive risk Evidence matches the defined signal pattern and survives exclusion rules. Could be marketing copy, evergreen hiring, stale activity, or unrelated context.
Review state Reviewer accepted, routed, or escalated with a note. Unreviewed, unsupported, contradicted, or missing required fields.
Action readiness Clear next step, owner, account context, and handoff field. Interesting but not yet actionable.

Confidence Levels

High

Action-ready with review

The source is specific, recent, entity-matched, and directly relevant. The record still carries its evidence trail and review state.

Medium

Useful but needs context

The signal has evidence, but one or more factors need review: recency, role clarity, account match, procurement implication, or action timing.

Low

Monitor or suppress

The record is weak, stale, generic, or likely to create false-positive work. Low-confidence records should not drive outreach or delivery without review.

Want a signal workflow that can explain itself?

VexASI can define the confidence rules, review states, and handoff fields for your market or technical workflow.