Leith built a fact-checking pipeline and ran it against 678 sessions of Claude building a real 21,000-line tool with no human-written code anywhere in its history. Checking a stated fact held up best, at 94.3% accurate.
In the News: September 26, 2026 (Morning)
A study of Claude's own coding-session claims finds roughly one in four to five responses contain a wrong claim, worst on proposed fixes.
Morning edition
Machine-readable
Download Markdown