AMSER pre-registration: twenty-one frozen tests of the 20 GeV Galactic halo excess — estimands, nulls, thresholds and falsification rules for Gaia DR4, Fermi-LAT, AMS-02 and multi-messenger data.

Pre-registration of twenty-one tests bearing on the 20 GeV Galactic halo excess and on a Gaia-tracer / gamma-residual alignment, frozen before the corresponding data exist. It fixes the region of interest, the estimator (partial rank correlation with |b| and θ_GC controlled), two null generators with a fixed seed, and a pre-declared threshold and falsification condition for every item — including a rule under which the author publicly withdraws her own finding, and an inference table that fixes the conclusion for each combination of outcomes in advance. Fifteen further analyses on data already in hand are listed separately as pre-specified rather than pre-registered. Procedural commitments cover blind construction of the DR4 tracer, negative controls, declared power, frozen quality cuts, a ban on rescue forks after an outcome is known, disclosure of failed calibrations, pinned external-dataset and software versions, and publication of all outcomes. The record contains the frozen analysis apparatus, a self-test reproducing all five published values of the v6 release, the target-map provenance script, and a SHA-256 manifest.Version 1.1 (same day): three corrections found by the author while executing the items, each marked inline in the document and explained in a separate erratum note — the campaign's halo profile is NFW with a 20 kpc scale radius rather than isothermal (a stale label in the configuration file had been trusted); the placebo region of item S-9 lies outside the data coverage and is replaced by controls of equal intent inside it; and the resolution-stability threshold of item S-13 was mis-specified because it ignored that pixel reliability falls with pixel size. None of the corrections relaxes a threshold in the author's favour. This version also adds the first execution report: seven items executed within hours of the freeze, of which two failed their declared expectation, one landed between the declared branches, and the leptonic closure criterion was met for one of two channels. Version 1.2 (11 September 2026): execution report 2e (all pre-registered items scored, statuses per the addendum's grading) and the addendum (grading of registered vs analysis-grade items) added; supplements article version 10.5281/zenodo.22695693.