The remaining twelve packs restated on the proven shape — core, kind-*, type-*, op-*
The rollout, once phase-deepen has proven the shape survives the cut. 2 828 lines across twelve packs, at these ratios:
| pack | lines | wrong | right | ratio |
|---|---|---|---|---|
core.md | 811 | 63 | 22 | 2.9x |
kind-leaf-deepen.md | 430 | 34 | 3 | 11.3x |
kind-story.md | 251 | 13 | 1 | 13.0x |
phase-skeleton.md | 246 | 10 | 1 | 10.0x |
op-replan.md | 238 | 18 | 0 | 18.0x |
kind-bug.md | 176 | 22 | 0 | 22.0x |
kind-leaf.md | 142 | 8 | 1 | 8.0x |
type-design.md | 114 | 6 | 0 | 6.0x |
type-test.md | 104 | 9 | 1 | 9.0x |
type-migration.md | 90 | 7 | 0 | 7.0x |
type-bars.md | 68 | 1 | 0 | 1.0x |
type-code.md | 59 | 1 | 0 | 1.0x |
core.md is the one to treat differently: it is always injected for every cell, it holds the 17-gate checklist, and at 2.9:1 it is the least inverted pack in the corpus — the gate checklist is genuinely prescriptive and most of what surrounds it is not. type-bars.md and type-code.md are already close to the target shape and are the cheapest confirmations that the measure is not just counting warrant tags.
This story is deliberately left unexpanded until MOTIR-3297 lands. Its children are one card per pack (or per pack family), and their acceptance criteria cannot be written honestly before the proof pack shows what a restated rule looks like and how much of a limb survives migration. Expanding it now would be planning against a shape nobody has seen.
Two things the mirror needs on the way through, both measured on origin/main: CORPUS_ORDER has no type-test, type-bars or op-sprint entry at all, so three source packs have no mirror pack to land in and the routing needs those slots; and the mirror files some packs' rules under a different pack than the source does, so the destination is found by a text probe over the constant, never by a pack-name lookup.
Acceptance criteria
- Every one of the twelve packs is restated or explicitly deferred with a reason, one child card per pack or pack family.
- Each child carries the same evidence its proof-pack predecessor did: a per-unit table, the migrated rows with their categories and
sourceRefs, and the before/after register counts. CORPUS_ORDERgains thetype-test,type-barsandop-sprintslots, or the child that would need them records why the routing stays as it is.- The corpus total lands materially below 3 513 lines and the epic's ratio table is re-measured and quoted.
- No child deletes guidance: every migrated paragraph is retrievable from the store.
Context refs
motir-metaprompts/plan-rules/— the twelve packs andMANIFEST.md's routing table.motir-aisrc/llm/planningRulePacks.ts—CORPUS_ORDER,PLANNING_RULE_PACKS,PACKS_BY_TYPE.- MOTIR-3297 — the proof pack whose shape this story rolls out.