I2995 — doctor does no work whose answer cannot change a finding: the roadmap lookups and the board's wide status
Remove the two readings that dominated doctor in the commit hook without changing a finding. (1) The roadmap doctrine reads the README's release list first, and when the section lists none it decides OK before looking up any milestone's release; a list is still compared with the release records both ways (ADR 0166, I2742: records, not tags). (2) The context builder's board asks `serve status --checkouts this` for the address it uses, and asks every checkout only when this one has no server, so its answer is the one the wide question gave.
DONE wave 0 · p1 · implementation profile · parallel safe
Objective
Remove the two readings that dominated doctor in the commit hook without changing a finding. (1) The roadmap doctrine reads the README's release list first, and when the section lists none it decides OK before looking up any milestone's release; a list is still compared with the release records both ways (ADR 0166, I2742: records, not tags). (2) The context builder's board asks `serve status --checkouts this` for the address it uses, and asks every checkout only when this one has no server, so its answer is the one the wide question gave.
Why
A profile of doctor in the commit-hook context (MJ_TIMING=1, 2026-10-18, after I2993) measured 244 s against the 100 s budget (policy benchmark.budget.doctor_ms). validate:roadmap took 64 s: mj_pj_m_released_in runs a `git log -S` pickaxe over the history for every DONE milestone, and in this repository the README's roadmap lists no releases, so every lookup was discarded. validate:budget took 63 s: it runs `majordomus context`, whose board lookup spent 11 s asking every checkout's server (`serve status` over the repository) to pick this checkout's address. Commit hooks took 14 to 37 minutes in every seat that day; the release seat named the hook the throughput floor.
Current state
The roadmap doctrine computes every DONE milestone's release before it reads the README; the board lookup asks the wide server status on every context.
Desired state
A README that lists no releases costs no release lookup; a worktree with a server costs one lease read and one probe for the board's address; the findings are unchanged.
Scope
- lib/doctor.sh
- lib/context.sh
- test/cases/3385_doctor_fits_its_budget_in_the_commit_hook.sh
Dependencies
None. This issue is a root of the graph.
Acceptance criteria
- A README whose roadmap lists no releases is judged OK with no pickaxe; with the change undone the case counts the lookups and fails
- A README that lists releases is accepted when it matches the records and refused when it omits one
- context asks only `serve status --checkouts this` when this checkout has a server; with the change undone the case sees the wide question and fails
- doctor's findings on this repository are the same before and after, and validate:roadmap and validate:budget take less time
Validation
- bash test/run.sh 3385_doctor_fits_its_budget_in_the_commit_hook 49_roadmap_doctrine 106_context_peers
Evidence required
- case_doctor_no_needless_work
Evidence
| covers | type | command | result | at commit |
|---|---|---|---|---|
| case_doctor_no_needless_work | ci | validate run 39438896504 at master 9a5a9b77520: suite shards 1-4 (jobs 118683336189 118683338077 118683336157 118683336457) run every case, incl. 3385_doctor_fits_its_budget_in_the_commit_hook (ed72c132539) | success (the run's only red job is pages, which runs no test) | a81b1e3 |
Timeline
- started
- 2026-10-20T16:19:24Z
- verified
- —
- completed
- 2026-10-20T16:33:32Z
Those three fields, the evidence above and the state of the dependencies are all the status is made of. There is no status field to disagree with them.
Canonical record: .ai/repo/project/issues/I2995.yaml. Read it back with majordomus plan show I2995.