Skip to content

The confirmation pass: Apache and LLVM qualify for an H1 cell, .NET and Grafana for RQ2, OpenJDK and HashiCorp are borderline

github-sizing-report.json, from the 200 draws an organization planned above (github-sizing-draws-<org>.jsonl). The design checks out: the same estimator applied to the number of PRs recovers the counts the search API reports to within 1% to 7% (Apache 108,600 against 107,888; HashiCorp 24,531 against 22,870, the widest). No draw hit the search frame cap. The thread-to-example conversion read from the built corpora is 0.217 (Wikimedia) to 0.282 (OpenStack); the example range below is the interval's low end at the low rate to its high end at the high rate.

org reviewer threads (95%) projected examples H1 cell RQ2 bot/AI-authored PRs
Apache 87,609 [51,535, 129,328] 11,176 to 36,518 yes yes 19.6%
LLVM 51,914 [35,618, 70,468] 7,724 to 19,898 yes yes 0.3%
.NET 36,763 [11,294, 70,216] 2,449 to 19,827 borderline yes 43.2%
Grafana 25,108 [9,310, 46,325] 2,019 to 13,081 borderline yes 26.5%
OpenJDK 11,803 [6,525, 18,106] 1,415 to 5,113 borderline borderline 0.0%
HashiCorp 6,356 [912, 15,101] 198 to 4,264 borderline borderline 20.2%

Suggestion-block threads, excluded above, are 13% to 23% of reviewer threads. The 50-PR first pass had OpenJDK at about 26,600 threads; the weighted pass puts it at 11,803, and its first figure came from a few heavily reviewed PRs.

What it gives the study. With OpenStack and Wikimedia, six organizations reach RQ2's bar in six languages' worth of codebases (Python, PHP, Java and others, C++, C#, Go and TypeScript): OpenStack, Wikimedia, Apache, LLVM, .NET, Grafana. Four have the size for an H1 cell (OpenStack, Wikimedia, Apache, LLVM), with .NET and Grafana possible. LLVM's top-level directories behave as projects (llvm, clang, mlir, lldb, bolt, flang and others), though llvm itself carries about 40% of the sampled threads, so whether its halves meet the split criteria is read before it is admitted.

Not yet settled, each a ROADMAP item: a GitHub collection route (review threads to hunk, comment and next revision, suggestion blocks and bot or AI authors removed, identities pseudonymised at ingestion); open access for every resulting publication, which GitHub's terms require (MSR's ACM proceedings are open access; the registered report's Stage 2 appears in EMSE, so its open-access option and an arXiv copy); and what more cells do to H1, which passes only if every admitted organization's cell does, each cell then needing power 0.95^(1/k).