Observed arrival · 2026-10-25
HypGraph — an evidence layer for AI research agents
HypGraph presents a hash-chained research record that checks agent-written claims against results, sample sizes, and review rules.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- Researchers and engineers building AI agents
- Worth noticing
- Its worked example computes that zero failures in 50 trials still leaves a 5.8% upper bound at 95% confidence.
Field notes
The evaluation walkthrough shows an agent moving from an “always beats” claim to progressively narrower claims, with refusals tied to arithmetic and test power. A separate Broad Street reconstruction follows competing explanations for cholera, including a skeptical curate and a decision linked to a claim that later changed. The page says the browser re-verifies the hash chain as the investigation loads.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
✦PrettyNotable craft visible
●ProPolished or operationally mature
◎NicheUnusually specific use
One card from the complete issue