Adoption
The six signals of NFR-ADOPT-01, measured monthly from wave 1 on the repositories where agent-authored pull requests are visible (design ยง How to know, rather than believe). scripts/adoption-signals.sh computes them for one repository from a checkout and the GitHub API, read-only. Before a repository switches, the gate is whatever boundary check it already runs, so the first row is the pre-switch baseline (plan 0001, Step 21).
| Signal | Target after two months |
|---|---|
| Agent-authored pull requests whose first gate run passes | above 90% |
| Median pushes from a failing gate run to green (GitHub's nearest record of agent turns) | one |
Rules added by agents that fail test or liveness before merge |
any number, as long as it is caught |
p95 of the Stop hook with --affected |
under 2 seconds |
Rules carrying fix |
above 80% |
| Merged budget raises | zero |
Baseline, 2026-09-23
| Signal | benbahrenburg/rulebearing (since 2026-09-01, gate self-check) |
benbahrenburg/ai-sdk-otel-logger (since 2026-07-25, no boundary gate) |
|---|---|---|
| First gate run passed | no agent-authored pull requests detected (0 of 11 merged) | no gate; 0 of 4 merged detected as agent-authored |
| Pushes from failing to green | none failed | no gate |
| Agent rule changes caught | 0 | 0 |
Stop hook p95 with --affected |
not measurable before wave 3 | not measurable before wave 3 |
Rules carrying fix |
100% of 5 | no Rulebearing rules yet |
| Merged budget raises | 0 (no ratchets) | 0 (no ratchets) |
What the baseline can and cannot see:
- Agent authorship is detected from the pull request's author login (Copilot, Devin, Codex, Claude, Cursor, Jules) or a
Co-Authored-Bytrailer naming an agent. Work an agent writes and a person commits under their own name, without a trailer, is invisible to it. This repository's commits are of that kind, so its first signal reads zero agent-authored pull requests. Until the maintainer's repositories mark agent work (a trailer or a label), the first three signals measure nothing there. That gap is the first thing to close before the two-month comparison means anything. - The private monorepo, where dependency-cruiser runs today and the design's evidence comes from, is the baseline that matters most. Only the maintainer can run the script there, with
--gatenaming its current dependency-cruiser job. - The Stop hook's p95 needs
--affected, which is wave 3.
No offers to other projects
Rulebearing makes no unsolicited offer to another project: no issue or pull request proposing it is opened on a test bed or any other repository (ADR-0065, NFR-ADOPT-02). The evidence is published here instead. Layer 5 shows zero differences on dependency-cruiser's own repository, langfuse and FluidFramework at their pinned commits, with equal exit codes (conformance/divergences.md lists no gate divergence). The agreement tables and the scale table are in the README.