Truth Discovery Engine + AI Reliability
Two engines, one API. Reconcile conflicting data from multiple sources. Score AI model reliability and detect hallucinations.
5 sources disagree on a price. Which is right? Moresq finds the truth using academic algorithms.
Group records that refer to the same entity. 5-level dedup.
Filter placeholders, normalize units, currencies, country codes.
CATD truth discovery finds consensus with confidence scores.
Cross-field coherence checks catch impossible combinations.
Placeholders filtered. Currencies converted. Consensus computed. Coherence validated.
19,756
fields reconciled
8,992
junk filtered
<2ms
per resolution
const result = await moresq.pipeline({
entity_id: 'asset_001',
evidence: [
{ field: 'price', source: 'source_a', value: '$285,000' },
{ field: 'price', source: 'source_b', value: '€262,000' },
{ field: 'price', source: 'source_c', value: '$290,000' },
{ field: 'year', source: 'source_a', value: 1967 },
{ field: 'color', source: 'source_c', value: 'Unknown' },
]
})
// → { price: 267000, year: 1967, color: filtered }
// → coherence.score: 1.05-level: exact ID, name, phonetic, composite, fuzzy. Presets for vessel, vehicle, product, property, company.
GPS from multiple sources → geometric median. Weiszfeld algorithm + DBSCAN outlier detection.
$15M, €14.2M, £12.5M → auto-converts then reconciles. Handles every format and symbol.
Upload CSV or connect Supabase/Postgres. Reconcile in background. Download results as CSV or branded PDF.
Score AI models. Route questions to the best model. Detect hallucinations. All from response data — no LLM inside.
POST /v1/ai/scoreSend 5 AI responses to the same question. Get weighted consensus, agreement score, and per-model reliability.
POST /v1/ai/routeWhich model for this question? Based on historical reliability per domain. Save tokens, get better answers.
GET /v1/ai/reliabilityIf 4 models agree and 1 says something different → hallucination flagged. No external lookup needed.
Score AI responses
const score = await moresq.ai.score({
question: 'Average rent in Paris?',
domain: 'real_estate',
responses: [
{ model: 'claude', extracted_value: 1200 },
{ model: 'gpt-4o', extracted_value: 1350 },
{ model: 'gemini', extracted_value: 1280 },
{ model: 'perplexity', extracted_value: 1320, has_source: true },
]
})
// → consensus: 1291, confidence: 0.80
// → hallucinations: []
// → perplexity weighted higher (has_source)Intelligent routing
const route = await moresq.ai.route({ domain: 'legal' })
// → { recommended: 'claude', reliability: 0.92,
// fallback: 'mistral', skip: ['gemini'] }After 100 queries, Moresq knows:
Scrape 5 APIs, reconcile in one call. Replace your custom if/else code.
Score model responses, route to the best, flag hallucinations.
Product feeds from 10 suppliers — which price, name, description is correct?
Listings from 17 agencies — reconcile prices, surfaces, locations.
Noisy sensor data → clean consensus. Incremental O(1) updates.
Market data from multiple providers — which quote is right?
$200K/year, 6 months setup
300 lines of fragile if/else that breaks at 3am
Validates rules you define — doesn't discover truth
Entity matching only — no field-level consensus
Moresq is the first
Pricing
All features included. Both engines. Pay for volume.
All features. Both engines. Never expires.
5,000 resolutions
Packs
Popular14,500 or 39,500 resolutions
Business
500,000 resolutions/month