CI diet: aprender itself burns 99 intel runner-hours/day — coverage on tag/nightly only, path-filter book jobs, perf/model suites nightly, mutants only on .rs diffs #3676

What (measured, not estimated)

Operator, 2026-09-21, verbatim: "ensure aprender releases are as fast as possible by: A. batching aprender PR B. keeping intel idle by telling out projects to batch PR or improve their CI/CD process (for example only code coverage on tagged release), etc. C. being relentless at linear and non-linear optimization…"

aprender itself is the largest intel consumer. Completed jobs on intel-* runners over the 24 h before 2026-09-21 08:26Z add up to 5,965 runner-minutes (99 h). By event: pull_request 3,812, merge_group 1,571, push 358, schedule 122.

Proposal

1. Coverage (ci / coverage) on the release tag and the nightly only, never on pull_request or merge_group. COV_FLOOR enforcement moves to the tag gate (T-2) and the nightly, where a regression still blocks the release. This is the operator's example, applied.
2. Book Chapter Examples and mdBook: path-filtered. They run when book/** or the crates a chapter imports change, and nightly always.
3. The perf gate case table and the model-tests falsification suites move to nightly. The PR path keeps a smoke row.
4. mutants runs only when the diff has .rs.
5. Every moved job still runs on the tag. check_workflow_path_filters.sh and the gate aggregator are updated so a skipped job is a pass only when the skip reason is the documented filter.

done_when

• intel runner-minutes per 24 h for aprender, measured the same way (sample the jobs API; a full sweep burns the shared API budget), drop by ≥ 40% at equal PR throughput
• the tag gate still runs coverage with COV_FLOOR, and a planted coverage drop fails the tag (control)
• a book-only PR runs the book jobs; a crate-only PR not imported by a chapter does not (both rows)
• gate accepts skipped only for jobs whose filter says skip; any other skipped state fails (mutation)

Workflow edits: authorized by the operator's instruction above. P0 for release speed. Depends on #3658 (in #3669) and pairs with #3668.

Folded sub-tickets (TRIAGE-100, 2026-10-14)

These open sub-tickets were folded here as checklist rows and closed as not planned so the open count stays at or under 100. Reopen one when its work resumes; an open PR for it reopens it.

• Docs-only PRs still spend ~45 min in guard-cargo (25) and guard-tree (15): gate them on a single docs_only decision so a roadmap PR's CI is < 10 min #3668 — Docs-only PRs still spend ~45 min in guard-cargo (25) and guard-tree (15): gate them on a single docs_only decision so a roadmap PR's CI is < 10 min (branch: ci/3668-unfold-neutral, PMAT-3668-docs-only-ci)
• Lever 2: one local entry point that runs the SAME guard scripts CI runs (no second copy, F1 #2640) — catch guard reds before push #4416 — Lever 2: one local entry point that runs the SAME guard scripts CI runs (no second copy, F1 ONE dogfood: two dogfood.sh runners share 61/61 gates and have drifted 86 lines — unify to a single runner with discovered repo gates #2640) — catch guard reds before push (branch: fb/4416-lever2, PMAT-4416-guards-local-wip)
• CI has no smart segmentation for docs-only PRs — full fat job + mandatory pr-review quorum run regardless of change type #4472 — CI has no smart segmentation for docs-only PRs — full fat job + mandatory pr-review quorum run regardless of change type
• check_hardcoded_paths.sh hp_row: printf | grep -q under pipefail flakes the R5 row (SIGPIPE) #4525 — check_hardcoded_paths.sh hp_row: printf / grep -q under pipefail flakes the R5 row (SIGPIPE)
• flow: signer-bot pushes leave CI action_required — approve held runs in-job #4579 — flow: signer-bot pushes leave CI action_required — approve held runs in-job
• check_crux_serve_code.sh: stubborn() TERM-ignoring helper leaks past the job (orphans in runner cgroups) #4637 — check_crux_serve_code.sh: stubborn() TERM-ignoring helper leaks past the job (orphans in runner cgroups)
• MUT-SCOPE: diff-scoped per-PR mutants + nightly full (sha-cached) + one full run per rc (post-0.70) #4642 — MUT-SCOPE: diff-scoped per-PR mutants + nightly full (sha-cached) + one full run per rc (post-0.70)
• C148: cancel-superseded.sh — cancel only runs whose head_sha differs from the PR head (DEVIATION-004) #4645 — C148: cancel-superseded.sh — cancel only runs whose head_sha differs from the PR head (DEVIATION-004)
• CI: fail-fast GPU lane on the Jetson Orin runner; a GPU test must not pass without a GPU #4879 — CI: fail-fast GPU lane on the Jetson Orin runner; a GPU test must not pass without a GPU
• Floor-priority lane lock for shared-host session builds (FIFO, no starvation) #4995 — Floor-priority lane lock for shared-host session builds (FIFO, no starvation)
• workspace-test nightly: widen the tests lane from aprender-core to the workspace after one measured run #5027 — workspace-test nightly: widen the tests lane from aprender-core to the workspace after one measured run
• nextest list-wrapper self-test: run nightly (cannot block merge or release) #5298 — nextest list-wrapper self-test: run nightly (cannot block merge or release)