Tasks: Per-PR Sonar reuses CI Modules coverage
Mission: sonar-per-pr-coverage-reuse-01M2FR32 · Issue: #4334 Branch: issue-4334-sonar-reuse-shard-coverage · Created: 2026-09-14 Inputs: spec.md, plan.md, research.md, data-model.md, contracts/, quickstart.md Squad evidence: work/4334-sonar-coverage-reuse/squad-post-spec.md, work/4334-sonar-coverage-reuse/squad-post-plan.md
Lanes
| Lane | Work packages | Character |
|---|---|---|
| A — measurement | WP01 → WP02 | Changes what is measured. Touches the matrix runner and the test inventory. |
| B — topology & enforcement | WP03 → WP04 → WP05 → WP06 | Changes who reads the measurement. Touches the gate model, the reporting workflow, and the retirement. |
The two lanes own disjoint file sets and can run concurrently. Within a lane the order is strict — each WP consumes its predecessor's observable output.
The one ordering that is not negotiable: WP04 (the regression battery) must land before WP06 (the retirement), so the battery's strongest mutation — "reintroduce the retiring step" — is proven red against live source rather than a synthetic fixture. Retiring first would leave only a fixture, reproducing the exact blindness the battery exists to close.
Preconditions (outside this mission's control)
Verified live: the project currently holds no coverage metric at all. This blocks the mission's publication claims only; every other requirement is independently verifiable.
- #4350 — SonarCloud Automatic Analysis must be disabled before FR-015/SC-010 can be satisfied.
Subtask Index
| ID | Description | WP | Parallel |
|---|---|---|---|
| T001 | Red-first gate: every slice measures the full top-level package set | WP01 | |
| T002 | Broaden the coverage-target derivation in the matrix runner | WP01 | |
| T003 | Measure and record aggregate cost: runner-minutes, artefact bytes, parse duration | WP01 | |
| T004 | Record the per-file coverage baseline artefact | WP01 | |
| T005 | Derive and commit the marker-mismatch exception set (FR-016) | WP01 | |
| T006 | Prove no per-file coverage regression outside the declared exception set | WP01 | |
| T007 | Commit a reproducible shard-timings capture script | WP02 | |
| T008 | Measure durations for the two new directories; record the producing run | WP02 | |
| T009 | Add the two inventory rows plus matching recognised-group entries | WP02 | |
| T010 | Wire the router filter groups for the two new rows | WP02 | |
| T011 | Re-freeze the non-vacuity floor and run the inventory gate battery | WP02 | |
| T012 | Prove both directories collect >0 tests and balancing uses fresh data | WP02 | |
| T013 | Red-first: the gate model cannot see indirect suite invocations | WP03 | |
| T014 | Resolve run: steps transitively to a suite invocation | WP03 | |
| T015 | Reconcile the workflow allowlist with the widened detection | WP03 | |
| T016 | Rewrite the non-blocking rationale to describe what is actually detected | WP03 | |
| T017 | Register any new architectural test file in the shard map | WP03 | |
| T018 | Battery scaffold with a non-vacuity floor over discovered workflows | WP04 | |
| T019 | Mutations 1-3 proven RED against the live retiring step | WP04 | |
| T020 | Mutation 4: a net-new workflow file forces directory enumeration | WP04 | |
| T021 | Mutations 5-6: removal of the non-blocking and same-origin declarations | WP04 | |
| T022 | Assert the unmutated live tree is clean | WP04 | |
| T023 | Red-first: behavioural tests for each execution-condition conjunct | WP05 | |
| T024 | Extract source-tree delivery and identity resolution into a tested script | WP05 | |
| T025 | Add the reporting job to the aggregation workflow | WP05 | |
| T026 | Add the skipped-tolerant terminal verdict job and widen permissions | WP05 | |
| T027 | Re-check the downstream parse timeout against the enlarged artefact set | WP05 | |
| T028 | Commit the pinning-rule derivation, its artefact, and a re-derive gate | WP06 | |
| T029 | Retire the duplicate-measurement job | WP06 | |
| T030 | Execute every disposition in the derived inventory | WP06 | |
| T031 | Update the governance documents in lockstep | WP06 | |
| T032 | Record issue-matrix verdicts and the post-integration observation | WP06 |
WP01 — Broaden coverage measurement
Lane: A · Dependencies: none · Prompt: tasks/WP01-broaden-coverage-measurement.md · 236 lines
Goal: Make each matrix slice measure the full top-level package set, so a test that exercises another module's code stops having its measurement silently dropped. Establish the baselines every later acceptance claim reads.
Priority: P1 — everything downstream measures against this.
Independent test: Run one slice before and after; covered statements outside that slice's own module go from zero to substantial, and no file's covered set shrinks.
Why this is not a tidy-first step: broadening changes which files appear in coverage reports, which changes the blocking change-coverage gate's absent set and its denominator. This WP changes merge-blocking behaviour for every pull request in the repository. Treat it as the riskiest package, not the safest.
Included subtasks: T001 T002 T003 T004 T005 T006
Risks: the aggregate cost could exceed the saving (measured +9.9% on one slice — must be confirmed across the matrix); artefact size grows ~6×, and the set is downloaded twice.
WP02 — Declare the undeclared test directories
Lane: A · Dependencies: WP01 · Prompt: tasks/WP02-declare-test-directories.md · 174 lines
Goal: Give tests/unit and tests/specify_cli/runtime a declared home in the test inventory, so retiring the duplicate step does not remove their per-change execution.
Priority: P2 — they retain scheduled coverage, so nothing becomes wholly unrun.
Independent test: the inventory names both directories; a pull request runs their tests; the balancing guard passes against freshly measured data.
Included subtasks: T007 T008 T009 T010 T011 T012
Risks: a length mismatch between measured durations and collected tests silently degrades balancing to a test-count split and no gate notices; the original capture tooling no longer exists in-tree, so it must be re-authored rather than re-run.
WP03 — Teach the gate model to see indirect suite invocations
Lane: B · Dependencies: none · Prompt: tasks/WP03-detector-substrate.md · 150 lines
Goal: The duplicate-detection substrate is documented as structurally blind to the retiring step's invocation form. Close that blindness while the step still exists, so the regression battery can be proven against reality.
Priority: P1 — a battery built on the blind substrate is vacuous by construction.
Independent test: the model resolves a suite invocation reached through a make target or a script, and the workflow that carries the retiring step enters the detected set.
Included subtasks: T013 T014 T015 T016 T017
Risks: widening detection pulls a new workflow into a set-equality invariant, cascading into the coherence model — that cascade is the point, but it must be landed deliberately.
WP04 — Permanent regression battery
Lane: B · Dependencies: WP03 · Prompt: tasks/WP04-regression-battery.md · 168 lines
Goal: A permanent in-tree fault-injection battery that fails on every way the duplicate could return, and on removal of the two declarations that keep the report advisory.
Priority: P1 — this is the mission's close-the-defect-class-by-construction obligation.
Independent test: each mutation is demonstrated red; the unmutated live tree is clean; the rule's workflow set is asserted non-empty.
Included subtasks: T018 T019 T020 T021 T022
Risks: a battery fed only synthetic fixtures reproduces the blindness it was written to close — mutations 1-3 must run against the live tree while the retiring step is still present.
WP05 — The reporting job
Lane: B · Dependencies: WP04 · Prompt: tasks/WP05-reporting-job.md · 201 lines
Goal: Add the per-change reporting job to the aggregation workflow, consuming the measurement the matrix already produced, under an execution condition whose every conjunct is independently asserted.
Priority: P1 — the mission's functional core.
Independent test: the extracted condition evaluator refuses each of the five ways the job must not run, driven by synthetic payloads, before the job exists.
Included subtasks: T023 T024 T025 T026 T027
Risks: the analysis configuration must resolve from trusted content — an earlier design that placed sources in a subdirectory was refuted because the configuration file is located relative to the analysis base directory. The production trigger cannot execute before integration; that limit is declared, and the first post-integration run is observed, not assumed.
WP06 — Retire the duplicate, execute dispositions, move the documents
Lane: B · Dependencies: WP05 · Prompt: tasks/WP06-retire-and-disposition.md · 181 lines
Goal: Remove the duplicate-measurement job, and relocate, rewrite or deliberately retire every rule that pinned the old arrangement — via a derived inventory, never a transcribed one.
Priority: P1 — this is where the cost saving is realised.
Independent test: re-running the derivation reproduces the committed inventory; every rule carries a disposition; the duplicate no longer runs.
Included subtasks: T028 T029 T030 T031 T032
Risks: rules that go red are found by CI; rules that go greener by deletion are found by nobody. The earlier hand-built inventory was ~4× understated with four false positives — which is exactly why this WP derives rather than transcribes.
Parallelisation
manufacture parallelism that does not exist.
WP03 + WP04 alone deliver the regression guard. The cost saving needs the full Lane B chain.
- Lane A and Lane B run concurrently. Disjoint file sets; neither reads the other's surfaces.
- Within lanes the order is strict. The mission is genuinely mostly sequential and the plan does not
- MVP scope: WP01 alone delivers the measurement-accuracy half and is independently valuable.
Recording progress
Subtask completion is event-sourced. Use:
spec-kitty agent tasks mark-status T001 T002 --status doneThere are no checkboxes to tick; the reduced event log is the authority.