Re-audit `phase-skeleton` and `phase-deepen` on the act test — the design gate moves to skeleton
Opened by Zhu Yue ·
The mirror of the card before it. That one asks which core rules belong to a phase; this one asks whether the rules already IN a phase pack belong to the one they are in.
The case that names the card
THE_DESIGN_GATE (5,292 chars) sits in phase-deepen and reaches all 17 deepen cells. Read for its DISCHARGE rather than its subject, it is skeleton work throughout: "a UI code subtask is blocked_by a type: design subtask producing the three-file asset" is an edge and a card; "MUST spawn the matching type: design amendment subtask in the SAME pass" is a card; "link (relates_to) the subtasks that SPECIFY that workflow" is an edge. A deepen session can do none of it — by the time one card's body is being written, the design blocker exists or it does not, and the pass has no tool to change that.
It reached phase-deepen by a defensible route: MOTIR-3897 re-homed it out of the retired kind-leaf-deepen and phase-deepen was the minimal superset AVAILABLE, not the phase whose surface can act on it. The subject test could not see the difference.
THE_ARTIFACT_OBTAINABLE_GATE and the two grep/ref segments are the other candidates the measurement raises; the ACT decision's own criteria require the full list, and this card executes it.
Acceptance criteria
- Every
phase-skeletonandphase-deepensegment is classified by the ACT test, and the PR carries the table: segment, current pack, verdict, and the act that discharges it. A segment that STAYS carries a verdict too — "left where it was" and "never examined" are indistinguishable in a diff. THE_DESIGN_GATEis resolved on the record. If it moves tophase-skeleton, the PR states what a DEEPEN pass is left with for a UI card — because the deepen half of the design gate is real (write criteria against the asset) even though its discharge is not, and dropping it is a rule that stops firing.- For every segment that moves, the routing baseline's superset proof shows the cells carrying it are a superset of the cells where its situation can occur; any lost surviving cell is named with the act that cannot occur there.
- A composed DEEPEN cell of any type contains no rule whose discharge requires
propose_node,modify_node,remove_nodeorlink_work_items, asserted by pattern over the composed text — the act test made mechanical, the way MOTIR-3900 made the subject test mechanical. - A composed SKELETON cell does not GROW by more than the segments this card and its predecessor move into it, measured and stated. Moving a rule from deepen to skeleton is right and is not free; the skeleton union is what a pass composes when the kind is unknown.
- Text is unchanged except where a segment is DIVIDED, each division proved by a word-diff over its parts; conservation and register baselines refreshed deliberately with their diffs reviewed.
Context refs
motir-aisrc/llm/planningRulePacks.ts— thephase-skeletonandphase-deepensegments;THE_DESIGN_GATE,THE_ARTIFACT_OBTAINABLE_GATE,THE_REF_HALF_OF_A_GREP_TARGET,A_CONTEXT_REF_THAT_EXPLAINS_IS_A_CLAIM.motir-aisrc/llm/treeGeneration.ts—SKELETON_PHASE_TOOLS/DEEPEN_PHASE_TOOLS, the surfaces the test reads.motir-aitests/planningRuleRouting.test.ts— the superset proof, per moved segment.- MOTIR-3897 — where
THE_DESIGN_GATEacquired its current pack, and why that was the minimal superset available at the time rather than a classification.
Discussion
No comments yet.
Adding to this discussion signs you in on app.motir.co and brings you back to this request.