Motir

The second-largest pack in the corpus after core.md, and the one every leaf deepen loads. 431 lines, 18 cautionary, 5 prescriptive (3.6x), measured with REGISTER.count.py kind-leaf-deepen at origin/main a9caaa1.

It is one card because it is one file with ten units, all about the same moment: a leaf card whose body is being written. Roughly, in file order —

• the data-seam rule (a card that consumes what another shipped verifies the READ) plus its name the exact call and the field discharge and its five ways this seam fails enumeration;
• the request-count claim (a criterion constraining how many requests something costs) plus its enumeration discharge;
• the unit of precondition-verification is the ACCEPTANCE CRITERION, not the card;
• the world-property AC rule — an AC asserting a property of the world rather than of the finished work is a precondition — with its why it is not merely untidy and its true/false resolution;
• the enumeration-card rule;
• a substrate method existing is not enough — verify it is callable in the plan/entity state;
• a method is not a path — discharge a presupposed mechanism by grepping for a CALLER;
• the manual / human classification by bulk;
• the artifact-obtainable completion gate;
• the estimation gate and its run-time ceiling (agent time excluding CI ≤ 1h).

Three of these are decision procedures and stay whole in the imperative, however the register scores them: the five ways this seam fails checklist, the artifact-obtainable gate's four limbs, and the estimation gate with its ceiling. Each is a thing a pass performs against the card in front of it, and each loses its function the moment it is compressed to a slogan. The estimation gate in particular is what every card in this very story was sized against.

Everything else follows the proven disposition discipline: keep means the imperative with no incident narrative, no > **Warrant:** block and no fixture pointer; migrate means the body goes to the lesson store and the one-line directive stays; a limb whose lesson row is not yet written is marked pending with its text left in place.

Note the cross-home mismatch and do not "fix" it. The estimation gate's source text lives here while its mirror segment is tagged kind-leaf in CORPUS_ORDER. Restate the wording; leave both placements alone, and record the mismatch in this pack's RESTATEMENT.md section so the sibling motir-ai cards know which of them owns that segment.

Acceptance criteria

• RESTATEMENT.md gains a kind-leaf-deepen.md section with one table row per unit and per named limb, carrying the disposition, the before/after line count, and for a migrate the destination sourceRef.
• The five-ways-this-seam-fails checklist, the artifact-obtainable gate's four limbs and the estimation gate with its run-time ceiling all survive as instructions, each still enumerating what it enumerated; the PR body quotes the restated estimation gate in full.
• Every migrated paragraph is verbatim under #restatement in prompts/plan-rules/fixtures/kind-leaf-deepen.md.
• The RESTATEMENT.md section records which mirror segments this pack's units correspond to, including any whose CORPUS_ORDER tag names a different pack.
• python3 prompts/plan-rules/REGISTER.count.py kind-leaf-deepen at this card's base and again on the result; both pairs of figures and the base ref in the PR body.
• No unit moves to another pack, and no unit is dropped without a row saying so.
• COMPRESSION.conserve.py kind-leaf-deepen is expected to FAIL; quote it, cite MOTIR-3377, and change no rule text to chase it.
• The change lands via a branch and a PR against main; no direct commit.

Context refs

• motir-meta prompts/plan-rules/kind-leaf-deepen.md — the ten units.
• motir-meta prompts/plan-rules/fixtures/kind-leaf-deepen.md — the conservation fixture to append to.
• motir-meta prompts/plan-rules/RESTATEMENT.md · REGISTER.count.py.
• motir-ai src/llm/planningRulePacks.ts — THE_DESIGN_GATE, THE_ARTIFACT_OBTAINABLE_GATE, THE_ESTIMATION_GATE — read-only, to record which segments this pack's units land in.
• MOTIR-3308 — the worked example of the table and the restated voice.