Skip to content

OpenStack's capture misfit: three explanations tested, none survives

Calibrated against its own simulated null, the fit's maximum gap against the longest-horizon cohort rejects OpenStack at p = 0.017 and clears Qt at p = 0.580. The roadmap carried this as an open item with a named suspect. The suspect is innocent, and so are the next two.

  • A latency trend violating quasi-independence. Ruled out already: Tsai's conditional Kendall tau rejects quasi-independence for neither organization.
  • A mixture of project families. OpenStack's corpus spans 264 projects, and the two big communities settle at genuinely different speeds: F(0) is 0.594 for starlingx/* against 0.390 for openstack/*, a twenty-point gap in same-month settling. Their mix also drifts, from 32% starlingx in the oldest cohort to 48% in the newest. That is a real compositional effect, and it does not explain the misfit: fitting openstack/* alone still rejects at p = 0.035.
  • Qt being the more homogeneous corpus. False, and measurably so. Qt's per-project lag distributions spread further than OpenStack's family-level gap: F(0) runs from 0.512 for qt/qtdeclarative to 0.758 for qt-creator/qt-creator, a spread of 0.247 across five projects with at least 120 changes each. Qt's top project also drifts more across cohorts than OpenStack's, 6.2 points against 3.4. Whatever distinguishes the two organizations here, it is not that one is a cleaner single population.

So the misfit stays open, and the honest reading is that the fit is off for OpenStack for a reason not yet identified. The bound already recorded still holds: refitting on recent cohorts moves the dev figure about seven points toward less censoring and leaves the test window's figure smaller, so every fit points the same way on the decision that depends on it.

The side finding is worth more than the answer that was being chased. Review latency differs sharply between projects inside one organization. qt-creator settles 76% of its example-bearing changes in the month they were created; qtdeclarative settles 51%. That is a quarter of the whole distribution's range, between two projects under one organizational roof, measured on process rather than prose.

It corroborates the separability probe from a direction with nothing in common with it. The probe read reviewer word choice and found projects as distinguishable as organizations; this reads how long review takes and finds the same thing. Two measurements sharing no mechanism, agreeing that the project is where the variation lives, is a stronger argument for the granularity question than either alone.