The Priors, Measured (EA-SEI-BATTERY-01 v1.0) is the work plan and first empirical record of the program that extracts the SEI classifier-foreclosure family — the OAR Protocol (#931), the Collapse Synthesis (#932), the auditable-foreclosure architecture (#933), The Endogenous Sophon (#935), and the SIGAGNOSTIC disambiguation packet (#1436) — into three publishable papers: a science paper for SciPost Physics, a philosophy-of-science paper on measurable theory-ladenness, and an STS/political-economy paper on the double enclosure.
Its empirical core is inversion battery v0.1, a demonstration-scale execution of Protocol I from the OAR Protocol, run 2026-08-11 on two public community datasets: the top-quark tagging reference dataset and the LHC Olympics 2020 R&D dataset. Four process pairs were tested against three anomaly-score families — a reconstruction autoencoder, an encoder-side latent-norm VAE with latent dimension 8 (the AXOL1TL-class score), and a Gaussian-mixture density model — in both training directions, with thresholds calibrated on held-out own background at accepted-background rates of one percent and one per mille. The design was fixed in the executable script header before any training.
The battery's headline finding is that the direction-dependence first measured by Finke et al. (2021) replicates in an independent implementation and extends beyond reconstruction loss: on jet constituents, QCD-trained systems detect top jets at AUC 0.84–0.87 while top-trained systems rate QCD as more ordinary than their own training class (AUC 0.24–0.30), for the autoencoder and the density family alike. Direction-dependence is systematic across every background-versus-signal pair; the density family posts the largest Inversion Asymmetry Index values (up to 0.41); at trigger-like operating points every tested system assimilates at least 93 percent of the structurally distinct partner class as ordinary; and the near-structure control pair stays quiet, showing the battery does not manufacture asymmetry. The record states its bounds with equal care: single seed, demonstration scale, surrogates rather than deployed systems, and no bound of any kind on the open-world Ontological Assimilation Rate — the no-bounds discipline of the protocol paper reproduced rather than softened.
The record also fixes the program's literature position: the direction-dependence conversation exists (Finke 2021; the Heidelberg normalized autoencoder; the CMS Wasserstein-NAE inversion check), and the program's contribution is the estimand framework, the systematic pre-registered panel, the deployed-model Benchmark Assimilation Rate audit, and per-stage retention maps as model cards for triggers. Battery v0.2 requirements, per-paper task ledgers, and the open decisions reserved to the MANUS close the plan.