The Retraction That Strengthened the Thesis
5 min read · ~1,000 words
Michael Darius Eastwood · Independent AI alignment researcher
Published
The original alpha 2.24 single-model fit was retracted. The alpha ~0.49 blinded six-model re-analysis replaced it. Both are disclosed. The retraction is not hidden. This is how science is supposed to work, and how it almost never does in AI safety research.
Evidence Anchors: 8 December 2024 manuscript (SHA-256: f0d1f38f) · Infinite Architects published 2 January 2026 (ISBN 978-1806056200) · OSF programme registration 17 January 2026 (DOI 10.17605/OSF.IO/6C5XB) · 19 institutional convergences documented · All claims traceable to canonical numbers block
The verification culture is the product.
Every claim carries dated evidence. Every prediction has a kill condition. Independent verification is invited.
Verify any claim →
Every claim carries dated evidence. Every prediction has a kill condition. Independent verification is invited.
Verify any claim →