Trust and Confidence

Every scored Page can show two donuts. They use the same values and color rules as the public Page and Console; documentation examples never generate random scores.

Trust

Trust measures how strongly the available origin, identity and ecosystem evidence supports trust in the subject being assessed. It is subject-specific: an X account score is not an Agent code-safety score.

Confidence

Confidence measures how much weight to place on the assessment based on evidence quality, coverage, freshness and corroboration. Low Confidence is a prompt to gather or refresh evidence, not proof that the subject is untrustworthy.

Read the pair

Trust	Confidence	Practical reading
High	High	Strong positive evidence with good coverage. Continue under your policy.
High	Low	Positive evidence is incomplete. Require more evidence or narrower permissions.
Low	High	Available evidence consistently supports caution.
Low	Low	Evidence is weak or sparse. Treat the result as unresolved.

Scores are always bounded from 0 to 100 when present. A missing score must render as unavailable, not as zero.

Read the pair in order

First ask whether the Page is the intended subject. Then read Trust, Confidence and the score timestamp together. A high score for the wrong repository or account is still the wrong evidence.

Confirm the score state

complete means both canonical values are available. partial, pending, failed, stale and unavailable require different presentation and policy behavior. See Score states.

Interpret Trust

Trust summarizes the direction of available identity, origin and ecosystem evidence for this exact subject. It does not grant permissions or certify source code.

Interpret Confidence

Confidence describes the coverage, quality, freshness and corroboration behind the assessment. Low Confidence should usually narrow the action even when Trust is high.

Add operational context

Consider requested permissions, data sensitivity, network access, reversibility and blast radius. These facts belong to your decision policy, not the score itself.

Example policy input

Keep evidence fields explicit so they can be audited and refreshed independently:

type TrustEvidence = {
  pageId: string
  pageType: 'agent' | 'skill' | 'creator' | 'x_account' | 'tg_account'
  trust: number | null
  confidence: number | null
  scoredAt: string | null
  scoreStatus: 'pending' | 'complete' | 'partial' | 'failed' | 'stale' | 'unavailable'
}

function requiresReview(evidence: TrustEvidence) {
  if (evidence.scoreStatus !== 'complete') return true
  if (evidence.trust === null || evidence.confidence === null) return true
  return evidence.confidence < 70
}

This sample demonstrates data handling only. It is not a Clawdbase threshold recommendation.

Display requirements

Use the same two-donut component and score colors on registry cards, Console Page cards and public Pages. Preserve null as unavailable, include the last-scored timestamp, and never generate placeholder numbers while loading. A refresh may optimistically mark the Page as processing, but it should display new values only after the authoritative record is returned.

On this page