CI Job Timings — `ci-test-topology-performance-01KXBJRT`

This is the mission's E4 timings artifact (data-model.md E4, mirroring tests/architectural/_gate_coverage.py's _TIMINGS_BASELINE shape). Per IC-07 / the project's flakiness policy (docs/guides/testing-flakiness.md): budgets here are measured and recorded, never asserted live. No pytest guard reads this file; a human/agent re-measures and updates it when the topology changes again.

1. Pre-mission baseline (the grounding evidence)

Source: GitHub Actions run 29196440986, main @ db904ebc5, workflow_dispatch, overall FAILURE (unrelated Sonar quality-gate failure — see spec.md's Evidence section). This run is fully serial (no -n auto, no --durations) and is the sole empirical basis for every NFR-001..008 budget below.

Job / leg	Minutes	`run_id`	Note
`integration-tests-next`	69.2	29196440986	441 tests fully serial; the dominant cost this mission targets.
`fast-tests-core-misc (specify-cli-rest)`	12.2	29196440986	Already a shard matrix but imbalanced vs `core-misc` (4.7 min).
`fast-tests-charter`	12.0	29196440986	Already `-n auto --dist loadfile`, but two sequential steps (charter + agent) summed on one job's wall-clock.
`slow-tests`	10.7	29196440986	~4 min of this was full-tree collection tax.
`fast-tests-cli`	8.3	29196440986	`--dist loadfile` single-file tail on `test_charter_activate_commands.py`.
`fast-tests-sync`	6.3	29196440986	The serial orphan-sweep step was ~130 s of this job's wall-clock.

2. Post-change topology — budgets recorded, not gated

The topology below is what actually shipped (WP01–WP06, verified against .github/workflows/ci-quality.yml on this branch). No real post-change CI run of this shipped topology exists yet — this mission's branch (feat/ci-test-topology-performance) has not been pushed to a remote that runs ci-quality.yml, and no PR has been opened. Recording fabricated wall-clock numbers here would violate the project's flakiness/honesty policy, so every minutes cell below is PENDING until a real run exists. Re-run this section (see §5, "How to refresh this artifact") the first time ci-quality.yml executes on this branch (or its PR), and replace the PENDING rows with the real --durations/job-summary figures plus that run's id.

Job / leg	NFR target	Pre-change (§1)	Post-change minutes	`run_id`	Met?
`integration-tests-next / next_shard_1`	NFR-001 ≤ 7 min	69.2 (whole job, serial)	PENDING	(none yet)	PENDING
`integration-tests-next / next_shard_2`	NFR-001 ≤ 7 min	—	PENDING	(none yet)	PENDING
`integration-tests-next / next_shard_3`	NFR-001 ≤ 7 min	—	PENDING	(none yet)	PENDING
`slow-tests`	NFR-002 ≤ 4 min	10.7	PENDING	(none yet)	PENDING
`fast-tests-sync-orphan-sweep` (new job, WP06 T016)	NFR-003 ≥ ~130 s off `fast-tests-sync`'s critical path	~130 s embedded in the 6.3 min above	PENDING	(none yet)	PENDING
`fast-tests-cli`	NFR-004 ≤ 5.5 min	8.3	PENDING	(none yet)	PENDING
`fast-tests-core-misc / specify-cli-rest`	NFR-007 ≤ 7 min	12.2	PENDING	(none yet)	PENDING
`fast-tests-core-misc / specify-cli-rest-2` (new shard, WP06 T017)	NFR-007 ≤ 7 min	—	PENDING	(none yet)	PENDING
`fast-tests-core-misc / core-misc`	NFR-007 ≤ 7 min	4.7 (pre-mission, unchanged shard)	PENDING	(none yet)	PENDING
`fast-tests-charter`	NFR-008 ≤ 7 min	12.0 (summed with agent)	PENDING	(none yet)	PENDING
`fast-tests-agent` (de-serialized from `fast-tests-charter`, WP06 T018)	NFR-008 (charter's split partner)	included in the 12.0 above	PENDING	(none yet)	PENDING
13 serial `integration-tests-*` single-dir jobs swept to parallel (WP06 T019: `agent`, `charter`, `cli`, `dashboard`, `doctrine`, `lanes`, `merge`, `missions`, `post-merge`, `release`, `review`, `status`, `sync`, `upgrade` — see §3 for the exact 14-job enumeration incl. `integration-tests-next`)	not separately budgeted (NFR-001..008 only name the headline jobs)	not separately measured pre-mission	PENDING	(none yet)	n/a

Shard-leg balance skew (NFR-006, ≤ 20% wall-clock spread): cannot be computed until real per-leg minutes exist for next_shard_1/2/3 and the fast-tests-core-misc legs (§4 also flags that the next_shard_N split itself is still WP01's placeholder bin-pack, not durations-rebalanced — this NFR is doubly pending).

3. Runner core-count context (plan.md Technical Context)

plan.md's Technical Context states: "GitHub Actions ubuntu-latest runners (assume ≥4 cores for the interim budgets; confirmed by the FR-001 --durations run)." As of this writing GitHub-hosted ubuntu-latest standard runners provide 4 vCPUs. Record the actual observed core count from the real run's --durations=25 / job-summary output alongside the run_id once §2 is filled in — do not assume it stayed at 4 without checking, since GitHub's standard-runner spec can change between mission authoring and the real run landing.

4. Known gaps affecting this artifact (do not silently omit)

next_shard_N balance is unmeasured. tests/_next_shard_map.py's own docstring says its 3-way split is "a placeholder — a greedy bin-pack over a def test_ count proxy... provisional pending WP06's real --durations=25 evidence... which rebalances this table from measured wall-clock." WP06's landed commit (f132ee611) parallelized integration-tests-next (T014) and rebalanced fast-tests-core-misc (T017), but did not touch tests/_next_shard_map.py to rebalance the next_shard_N split from real durations. NFR-006's ≤20% skew claim for the next tier is therefore still open — flagging here rather than silently marking it satisfied. Follow-up: once §2's real run exists, if any next_shard_N leg's minutes diverge from its siblings by >20%, rebalance tests/_next_shard_map.py's file assignment from that run's per-file --durations=25 output (same method _arch_shard_map.py documents using), not by re-guessing.
Coverage-union audit found a real (non-timing) gap — see ci-coverage-union-audit.md for the full evidence: GC-2b is currently RED for fast-tests-core-misc in this environment, filed as Priivacy-ai/spec-kitty#2607. It is a pre-existing baseline-portability defect unrelated to any test actually being dropped or double-run — see that document for the quantitative proof.

5. How to refresh this artifact

Push feat/ci-test-topology-performance (or its PR) and let ci-quality.yml run for real.
Pull each job's --durations=25/--durations=50 output (uploaded as part of each job's out/reports/ artifact) or the Actions run summary.
Replace every PENDING cell in §2 with the measured minutes and that run's numeric id; compute NFR-006 skew from the next_shard_N and fast-tests-core-misc rows.
If any NFR target is missed, do not silently adjust the target — record the miss plainly (this file is a measured record, not a gate) and open a follow-up if a rebalance or further split is warranted.
This file is a single point-in-time record, updated in place on re-measurement (per the mission's own risk mitigation) — do not append a second, conflicting timings table for a later run; supersede this one.