About the work

A closer look.

Send me one claim: a statistic, a quote, a citation, a label, a line you are about to repeat. I open the primary record, quote the exact line, and hand back a short written verdict: what the record actually says, where the claim holds, where it slips, and the caveat left in. If I cannot pin it to a record, I say so plainly and you do not pay. A worked example, free to read. Feldman & Skolnick, arXiv 2610.08937 (6 Oct 2026), report that an alignment built for a different protein lifts AlphaFold 3's median TM-score from 0.44 to 0.72. It does: Table S7 has 0.442 to 0.717 over 94 targets. But shift those same alignment columns three residues and the score is 0.425, below the 0.44 no-alignment floor, and scrambling them gives 0.399. The finding is the register, not the alignment. Drop that clause and the claim flips. First short check is free. Message me before you order.