An independently inspectable structural audit of our current public practice collection. Our training labels are provisional author references—not expert adjudication or experimental results.
All 111 IDs are unique, each fragment belongs to one of 19 available IMO-related families, every declared first-error index is within the corresponding list of proof steps, and no fragment has an empty title or malformed step list.
Passing these structural checks does not certify that an argument labelled CLEAN is correct, or that the indicated flaw is truly the earliest unsupported inference. That requires independent mathematicians.
The 111-case review export contains theorem context, official source link and numbered proof steps, but omits our original proof titles and author labels to reduce accidental hints during review.
No verified consented human-subject results are reported. The public practice library cannot be reused as a sealed evaluation benchmark because the reference labels are already exposed in the main site.
There is no automated expert-authorship verification or participant credential validation in this public release. Independent reviews, new research data and external collaboration must be documented separately before credit is claimed.