LIVING-1: Does living a life help predict a person? Frozen protocols, sealed predictions and the feasibility gate

Preregistration bundle for LIVING-1, a longitudinal persona ablation test. The hypothesis: a person's future behaviour is better predicted by an architecture that lives a trajectory (structured memory consolidating over time, path dependent) than by the same information held statically or retrieved flat. Seven conditions separate information from architecture, with a shuffled sequence control and a wrong person control. The bundle contains the frozen feasibility spike protocol, the scored spike results under a verified seal (feasibility pass; the bare model scores below chance, so no pretraining memorisation; the wrong person control collapses, so the architectures measure the person), the frozen full Stage A protocol with two preregistered paired contrasts, and the frozen openESM substrate implementation. The full runs are in progress at deposit time on two substrates: 276 pseudonymous authors from a public web archive, and 601 participants from two open experience sampling datasets. Predictions are hashed into a run manifest before the sealed answer keys are opened. Results follow as a new version of this record, whatever they show. Version 2 (27 October 2026): internal infrastructure naming in the prose and code is replaced by neutral terms, with every replacement listed in the archive's sanitisation note and the executed originals anchored by the hashes in the included freeze manifest. The analysis addendum now carries its fourth section, declared on 26 October before the ESM manifest existed. The E2 replication protocol, its amendment and its freeze manifest are added. Status at this deposit: the ESM arm of E1 unsealed on 26 October (results follow with Stage 2); the Reddit arm of E1 and the whole of E2 remain sealed and running.