OCS / Papers / Paper J

OCS Research Preprint · Paper J · v0.11 working draft

A Gaia Test of Published and Preprint Predictions for Stars Ejected from Milky Way Globular Clusters

Tim Swanson, Post Oak Labs / The Omega Centauri Society. Version 0.11, last revised 2026-11-02.

Status: methods-first skeleton with a preregistered design. No signal-region count has been computed, and Stage 1 has not run on real data.

Abstract

Two models predict populations of stars ejected from Milky Way globular clusters by encounters with stellar-mass compact objects: Evans et al. (2025) predict the population in aggregate with Gaia detectability applied, and Cabrera and Rodriguez (2023) publish per-cluster sky and proper-motion credible regions for the whole-lifetime ejecta.

We register a blinded count test of both predictions in Gaia DR3 and, under the pre-registered switch rule, in DR4. The test asks whether the pre-declared predicted regions hold the predicted excess over a field background estimated from rotated control regions. A blinded power gate, evaluated on control-region counts and template-derived quantities before any signal-region count is opened, treats a gate stop as a reportable outcome.

The first result is a bound that needs no signal model. Summed over clusters, the squared signal-to-background ratio is at most 1302/(2.0×106) = 8.4×10−3, so the expected Z0 under the alternative is at most 0.09, and at the registered Z0 ≥ 3 threshold the design has 0.18 per cent power against a 0.13 per cent null tail, for any distribution of the roughly 130 predicted Arm C stars across clusters.

The second result is a defect in the Cabrera and Rodriguez archive. Its per-cluster ejecta tables store a literal sentinel in every position and velocity field of the recent 500 Myr window, which the pipeline’s own transform maps to a single sky position that the mock sky exclusion removes, so the preregistered Arm C validation cannot be scored on the archive as published. Independently, the registered information floor asks for 118 detectable stars in a single cluster at the median background of the 148-row archive record against an arm-wide predicted total of about 130.

Arm E could not be run either. Its template model was never implemented, and its one testable region, the NGC 7099 box, fails the registered control-validity rule through the same absolute fdrop defect that affects Arm C. No power result exists for Arm E, and no signal-region count has been computed for either arm. The Methods section records the registered design with each number traced to its source line, and the deviations table records both Arm C findings in the eleven-field form.