Skip to content

A secondary estimate pooled across organizations

Superseded on 2026-09-23. The pooled figure rested partly on Qt's lint bot (bot audit), and the secondary estimate is not part of the current design.

The conjunctive gate asks whether each organization shows the effect, which is what a generality claim needs and what costs the design its resolution. A pooled estimate answers the weaker question, whether organizations leave a fingerprint on average, and is the sharpest reading the same data supports. Over the three seeds, every change from both organizations as one cluster set:

reading estimate interval effect over half-width
OpenStack alone +0.0171 [-0.0057, +0.0407] 0.74
Qt alone +0.0312 [+0.0079, +0.0565] 1.28
both pooled, 697 changes +0.0260 [+0.0090, +0.0440] 1.48

It is a secondary, never the gate. Pooling treats the two organizations as one population of changes, so a single organization with a strong effect can carry it, which is the failure mode the conjunctive rule exists to prevent; and with two organizations there is no way to model between-organization heterogeneity rather than assume it away. Registered as a secondary estimate, reported beside the gate, and fixed now while the test window is sealed.