Confidence Scoring Kit

Assign and maintain confidence scores for every fact based on source quality and corroboration.

Configuration

Avg Score

0.72

High Conf

2

40% of total

Low Conf

1

20% of total

Fact Confidence Scores

API rate limit is 1000 req/min

3 sources · 2 corroborations · 2d

0.92

Auth uses JWT tokens

2 sources · 1 corroborations · 5d

0.88

Max payload size: 10MB

1 sources · 0 corroborations · 1d

0.75

Webhooks retry 3 times

1 sources · 0 corroborations · 12d

0.45

Supports GraphQL endpoint

2 sources · 1 corroborations · 8d

0.62

Scoring Pipeline

ScoreBayesian posterior: 0.89
Sources3 sources, avg reliability: 0.82
Corroborate2 independent confirmations
DecayAge penalty: -0.02

Integration Code

import { createConfidenceScoring } from 'agent-tools-kit/knowledge-evolution'

const scorer = createConfidenceScoring({
  model: 'bayesian',
  decayEnabled: true,
  decayRate: 0.01,
})

const facts = [
  { id: 'f1', text: 'API rate limit is 1000 req/min', sources: [...] },
  { id: 'f2', text: 'Auth uses JWT tokens', sources: [...] },
]

const results = await scorer.evaluate(facts)
// results: [{ id: 'f1', score: 0.92, breakdown: { sourceQuality: 0.88, corroboration: 0.95, decay: -0.01 } }]

scorer.on('score-updated', (fact) => {
  console.log(`${fact.id} confidence changed to ${fact.score}`)
})