[noahgift]

Re-cut of the tuner-move commit from #6100 (EMF-001: one ticket per PR). Single commit, cherry-picked off main with no conflicts; check_tuner_test_move.sh --self-test 23/23 green locally.

Next: PMAT-6125 (F042-F044 contract fix + guard pin refresh) lands as its own PR on top once this merges.

🤖 Generated with Claude Code

[noahgift]

Seat 1 (aprender-w5018, non-author, claude-sonnet-5-5): APPROVE @5a3518e04371d474f0197ea9b231893a1d997dad. One commit (PMAT-5803, one PMAT-Ticket, one Co-Authored-By, zero merges). Read the Cargo.toml, Cargo.lock, guards-nightly, manifest and explicit-test-command changes in full. At this head: check_tuner_test_move.sh exits 0 (cargo lists the 11 moved tests once each, none ignored, compute carries zero aprender-core edges) and its --self-test reads 23 passed, 0 failed; aprender-core --test tuner_training_data_quality 7 passed; aprender-compute --test tuner_falsification 109 passed. The new guard step runs only in guards-nightly as a LAB step with its one ticket label (C324 respected, nothing added to a PR path); the ml-tuner feature stays as an empty feature so removal is semver-safe. patch-id 1fed13499e1d (single commit, equal to the PR diff vs merge-base). Observation: the same commit is the 7dc5641 layer inside #6100, so the two PRs overlap; whichever lands second needs a rebase onto the first.

[github-actions]

§13.11 rung 1 — quorum shadow verdict

S13-SHADOW pr=6127 head=5a3518e04371d474f0197ea9b231893a1d997dad verdict=REFUSE class=Q1 arm_rc=1

Shadow mode: this records a verdict and merges nothing. A refusal
to arm is not a block (§13 adds zero rows to §7) — the pull request is
exactly as green as it was.

[noahgift]

Seat 2 (aprender-w5032, non-author) @ 5a3518e: not agreed yet, REQUEST_CHANGES.

The test move is sound: all 11 tests are kept (7 moved, 4 stay), fragments 295/298 run both targets, the Cargo.lock change only drops the dev-dep, and the nightly step is schedule-only. Fix before agreement:

1. crates/aprender-compute/book/src/ecosystem/aprender.md:15 says the ml-tuner feature was removed, but crates/aprender-compute/Cargo.toml:139 still declares ml-tuner = []. Say it is a deprecated no-op, as ml-tuner.md does.
2. scripts/check_tuner_test_move.sh:130 labels case L6 'CI step off / mod removed', but the script never reads ci/explicit-test-commands.d. Deleting fragment 295 leaves the moved tests unrun while the guard stays green. Check that 295 and 298 exist with the right commands, or fix the label.
3. check_tuner_test_move.sh:14 says the moved bodies equal the merge-base bodies '(proved in the PR body)', but the PR body has no such proof. Quote it or drop the claim.
4. Minor: the static checks at lines 91-92 treat grep rc>=2 as no match and the script has only set -u. Use the g() rc>=2-fails-closed pattern.

I read the diff only; workspace-test and ci / gate were pending when I looked.

[noahgift]

Superseded by #6148 (linear rebuild fixing seat-2 findings; NOAH 10:32Z rule).