Research notes ordered by weight, not date. Every page re-proves itself: its numbers are recomputed from the certificates and records at build time, its planted falsifiers must fire, and a build that drifts refuses to ship.
Erdős’s list is a public register of open questions, which makes it the sharpest available test of whether a verdict produced here survives contact with the people who own the problem. Each page decides a finite, exact fragment and says exactly which one: a published constant refuted at its twelfth significant digit with its correction certified, certified extremal tables for Chowla’s cosine problem, a square-discriminant law re-proved as integer identities, and an independent verification of a claimed proof’s computational appendix. What has been filed with each problem, and what has come back, is on the about page, in status words meant exactly.
Aerospace and energy: domains where the operative numbers are defended by simulation today, and where a universally-quantified certificate — or an honest refusal — is a different kind of statement. Each page names what it does NOT claim.
These are where the verifiers earned calibration before deciding anything a model produced: a published Wardrop table reproduced within its own rounding and then proved outright, mean-field equilibria enclosed in disjoint balls where uniqueness theory is silent and REFUSED at the bifurcation, an entropy bound calibrated at the full horseshoe. The audits above — and the Erdős pages, whose exhaustive box sweeps are these same instruments — stand on this ground.