DNI · Integrity & novelty assessment

DNI's reference-integrity check now works on manuscript full text, not only on publisher-deposited reference lists. On a test specified in advance (RIC-PREREG-3: 205 PubPeer-listed papers, 410 same-journal controls), the 10% fault rule applied to DOIs read from open-access PDFs flagged 91.3% of PubPeer-listed papers and 4.9% of controls, and wrongly faulted 0.16% of correct references. That matches the deposited-list result already confirmed (RIC-PREREG-1: 93.9% vs 4.0%). The result holds only with the two new 8.9 settings switched on (parser v2, matcher v3), and both pre-registrations of this round were committed to git after their data, so they are weaker evidence than RIC-PREREG-1. An earlier, smaller test of the same settings (RIC-PREREG-2, 100 controls) was not confirmed. The novelty score was not re-tested in this round and remains weakly validated.