Adoption review, October 2026: the review that opens wave 4
The entry review of plan 0004, Step 1: the six signals of NFR-ADOPT-01, and which funding outcome of plan 0004 § 1.2 applies. The baseline and the method are in docs/adoption.md.
The signals, 2026-10-10
Measured with scripts/adoption-signals.sh . --repo benbahrenburg/rulebearing --gate self-check --since 2026-09-01 --graph target/cargo-graph.json, plus the nightly's Stop-hook row.
| Signal | Target after two months | benbahrenburg/rulebearing, merged since 2026-09-01 |
|---|---|---|
| Agent-authored pull requests whose first gate run passes | above 90% | 100% of 1 (1 of 71 merged detected as agent-authored) |
| Median pushes from a failing gate run to green | one | none failed |
Rules added by agents that fail test or liveness before merge |
any number, as long as it is caught | 0 |
p95 of the Stop hook with --affected |
under 2 seconds | 1.875 s over 200 edits on aspnetcore, source mode (nightly run 38021235710); under 2 s on every scheduled night since 2026-10-03 |
Rules carrying fix |
above 80% | 100% of 5 |
| Merged budget raises | zero | 0 |
What the numbers can and cannot say:
- The first three signals measure almost nothing yet. The script detects agent work from the pull request's author or a
Co-Authored-Bytrailer, and this repository's agent work carries neither, so 70 of 71 merged pull requests read as human. One pull request is not a rate. docs/adoption.md names this gap as the first to close. - Two consecutive months have not been measured. The baseline is from 2026-09-23; this is the first review after it.
- The private monorepo, the baseline that matters most, can only be measured by the owner, and has not been.
- The last three signals are met. That includes the Stop-hook p95, which became measurable with wave 3.
The funding outcome
The entry criterion is not met on the evidence. Four signals meet their targets on one repository. The first two, the ones the decision rule turns on, have one data point between them, and nothing has been measured over two consecutive months.
The wave is funded by the owner's decision of 2026-10-10 ("move onto implementing plan 0004-wave-4-reach.md"), ahead of the entry criterion. The decision rule of § 1.2 is therefore not what funds it. It stays the record of what the signals would have decided. The closing review (Step 14) measures the signals again, and says whether the agent features earned their place.
| Sub-wave | Funded | By |
|---|---|---|
| 4A docs site, playground, cookbook, live tables | yes | the owner's decision |
| 4B pull-request app and Azure DevOps extension | no, deferred | the owner's rule of 2026-09-30 (§ 1.2): it waits until the product ships, meaning plans 0001 to 0003 are implemented and a non-placeholder release is on every registry. Publishing is the owner's, still to come |
4C fix --plan and the eval harness |
yes | the owner's decision, which the § 1.2 rule alone would not give: the first two signals have not moved because they have not been measured |
4D fleet and declared edges |
yes | the owner's decision |
| 4E usage counts and the closing review | yes | never cut (§ 3) |
Before the closing review
- Mark agent-authored work so the first three signals can see it: a
Co-Authored-Bytrailer naming the agent, or a label the script reads. Without one, Step 14 will read the same zero. - Measure the private monorepo with
scripts/adoption-signals.sh,--gatenaming its dependency-cruiser job.