No source, no factual claim
17,896
sources stored
Candidate-specific claims must link to stored source metadata, quote/excerpt, date, confidence, and review status before they are treated as evidence.
Trust Center
The product has to earn trust claim by claim.
This page explains methodology, privacy, AI use, correction workflows, and source quality with live evidence metrics. It is deliberately visible early, not buried in a footer like legal confetti.
DB-backed evidence stats18602 claims trackedLatest source fetch: 2026-10-09
Unknown is a first-class answer
15,915
claims needing review
Claims that need more evidence stay visible in the review queue instead of being smoothed over by partisan assumptions or model guesswork.
Evidence quality is measurable
3,443
high-reliability sources
The trust layer tracks source reliability separately from candidate positions so weak evidence cannot quietly masquerade as strong evidence.
AI explains retrieved evidence
17,895
automated extractions
Automated extraction is labeled and counted; the assistant should explain stored claims and sources, not invent candidate facts from model memory.
Source reliability coverage
Know where the evidence is strong — and where it is still squishy.
This rolls up stored sources by reliability tier and shows how many public claims each tier actually supports. Weak sources cannot hide in the soup.
High reliability
3443 sources- Source-backed claims
- 1800
- Claims needing review
- 434
- Latest fetch
- 2026-07-10
Medium reliability
11885 sources- Source-backed claims
- 103
- Claims needing review
- 6828
- Latest fetch
- 2026-10-09
Low reliability
2568 sources- Source-backed claims
- 784
- Claims needing review
- 8653
- Latest fetch
- 2026-07-10
Source freshness
Spot stale evidence before it quietly rots.
Fresh is fetched in the last 30 days, aging is 31–180 days, stale is older. Counts come from stored source metadata and linked claims.
Fresh
3
sources • 0 claims
Aging
17893
sources • 18602 claims
Stale
0
sources • 0 claims
Not fetched
0
sources • 0 claims
Correction workflow
0 open- 1. User submits the claim, source, and correction context
- 2. The request is stored in Postgres with an open review status
- 3. Reviewer checks source fit against evidence standards
- 4. Public claim status is corrected, disputed, or removed
Investigation queue
0 pending0
Awaiting source scrape
0
Manual triage
0
Invalid source
Submit a correction
Add ?contact=you@example.com or sign in before submitting if you want this Trust Center to show your correction status later.
Flag unsupported, stale, or missing evidence without emailing a black hole. Revolutionary technology, apparently.