Directional assimilation in accelerator anomaly classifiers: the ACRB hypotheses

The Accelerator Classifier Retention Battery sets out seven standing hypotheses on directional assimilation in learned anomaly selection: reconstruction-specific failure, cross-family persistence, the normalized-autoencoder remedy, the low-complexity blind spot, blind-spot inheritance through distillation, drift under quantization and firmware, and the value of architectural plurality. Inversion Battery v0.2 reports that only 36 to 47 percent of a teacher's top one percent survives into the student's, with the loss in distillation, that architecture changes the orientation of the blind spot, and what is deferred to v0.3. The priors paper and the OAR protocol name the program's contribution and its pre-registered class-conditional inversion battery at matched background acceptance. The common experiment proposed is to calibrate each model at matched operating points and measure how much of a withheld class is assimilated as ordinary.

Each claim below is registered at the earliest deposit that states it, and quoted verbatim from the deposit text, with every later statement, its dated revisions, and the later records that claim it came true or failed. The claims are the deposits' own; this page records them and does not judge them.