Post-retraction citations in physics: 29% name the retraction, 15% still endorse
Measured by hand-coding the citing passages (arXiv LaTeX + OA PDFs + Semantic Scholar contexts) for four high-profile physics retractions: Dias CSH and Lu-N-H, the Delft quantized Majorana conductance paper, and the UCLA chiral Majorana paper. Of 136 texts written after the retraction (arXiv v1 after it, or journal-only and published at least 6 months after), 29% say 'retracted', 34% only hedge ('claimed', 'controversial'), 21% cite as background and 15% still cite the result as valid evidence. In small samples from before the retraction, valid-evidence citations were about 41-45%; texts written during the expression-of-concern window looked no different. Methodological tips: a blind second coder got kappa 0.85, with disagreements all at the endorse-vs-background boundary. Use publication date minus a lag, or arXiv v1, never the journal date alone, to decide whether a text was written after the event.
Correction from the author: not all coder disagreements were at the endorse-vs-background boundary. It was 10 of 12; the other two were about whether the passage concerned the target at all (X). Also the concern-window sample is n=56 after recoding one misdated record. Neither changes the headline numbers.