Story: Run every check locally in one report

1. Goal
2. Status
3. Acceptance
4. Tasks
5. Decisions
6. Out of scope
7. Result

This page documents a story in Sprint 27. It captures the goal, current status, acceptance criteria, and the tasks that compose it.

1. Goal

One local command replaces the GitHub pull-request wait. The command classifies the change against a documented decision tree, runs only the checks that class needs, and writes one report that names each failing check and the fix. A docs-only change runs no C++ build and no C++ test.

2. Status

Field	Value
State	DONE
Parent sprint	Sprint 27
Now	Nothing.
Waiting on	Nothing.
Next	Nothing.
Last touched	2026-10-07

3. Acceptance

One command classifies the change and runs every check the change needs.
The report names each failing check, the evidence, and the command that fixes it.
The decision tree maps each change class to its checks and lives in the repo.
A documents-only change runs no C++ build and no C++ test.

4. Tasks

Task	State	Start	End	Description
Scaffold story: Run every check locally in one report	DONE	2026-10-06	2026-10-06	Story scaffolding rides this task: documents, sprint wiring, and the scaffold PR. Close it before merging that PR.
Build the local check runner, its report and its decision tree	DONE	2026-10-06	2026-10-07	Initial task for: Run every check locally in one report
Retire the GitHub pull-request checks	DONE	2026-10-06	2026-10-06	Delete every workflow that gates a pull request, once the local runner covers it. GitHub keeps only the builds that are not pull-request checks.
Order the generating checks before the deriving ones	DONE		2026-10-08	tangle-shell runs before component-drift in the catalogue, so it tangles the recipes as committed and component-drift then rewrites them: the scripts are stale again the moment the recipes are committed, which is why the shell library needs a re-tangle catch-up commit every few days.
A dependency may not run this time	DONE		2026-10-08	The dependency check compared against the selected checks rather than the catalogue, so a narrow selection that picked site-page without component-drift raised a ValueError and crashed the runner. Any docs-only change would have hit it.

5. Decisions

The decision tree is data, and it lives beside the checks it selects. A table of change class to path glob in build/scripts/local_checks.py, not a chain of conditionals, so --list prints it and the suite pins it.
A path no rule claims selects every check. A change the table cannot classify is one it cannot vouch for; the cost of the extra run is smaller than the cost of the break it hides.
The checks that write the tree run serially, before the checks that read it. component-drift --all regenerates every artefact in place, and it was rewriting generated C++ while format read it.
A workflow may not carry a pull_request trigger. The retirement removed twelve of them, and nothing read the workflow files, so one grew back within hours. A guard now fails on it.
The change is measured from the merge base with the trunk, so another author's merge never reads as this branch's change.

6. Out of scope

Merging the twelve retired workflows into one. They remain twelve near-identical copies of the same drift gates, running on a main push.
Checking that every check in the catalogue has a counterpart in the retired workflow set.

7. Result

GitHub runs no check on a pull request. The twelve workflows that gated one lost the trigger, and compass check is the gate: it classifies the branch diff against a decision tree of 10 change classes and runs the checks the matching classes need — 38 in the catalogue — writing one report that names every failure and its fix. A documents-only change runs the site page and the doc corpus checks and never reaches the compiler.

Verified on the implementation branch at c35caff642: 38 of 38 checks passed in 3920s with a clean tree afterwards, the first run to exercise the whole catalogue including the build, the database recreation and ctest.