Skip to content

moooon

Motir

Vibe your whole project. Bring an idea — Motir's three AI layers plan it, track it, and ship it, end to end. You're looking at Motir, built in Motir.

  • Vibe Project
  • Open Source
  • AI Agent
  • AI Loop
1
requests
0
upvotes
145
planned
1,361
shipped

Motir · Work items

MOTIR-2661Done

(motir-meta) Compress the two deepen packs — `phase-deepen.md` + `kind-leaf-deepen.md`

Repo: motir-meta. One PR. The third execution card under MOTIR-2642's COMPRESSION.md, applying the mechanics MOTIR-2659 built.

phase-deepen.md (436 lines) and kind-leaf-deepen.md (450 lines) are the two packs that arrive together on every deepen pass over a leaf — which is the pass COMPRESSION.md exists for. Between them they carry 265 lines of warrant across 25 warranted rules, the largest concentration outside core.md.

They are one card rather than two because a leaf deepen loads both and neither is separately measurable against the worst-cell target: the cell they dominate is deepen · bug · code under a re-plan, and it is the number § Decision 4 sets a bar for.

Acceptance criteria

  1. Every warrant span removed appears verbatim in prompts/plan-rules/fixtures/phase-deepen.md and fixtures/kind-leaf-deepen.md respectively, under the anchor its inline tag points at.
  2. Each compressed rule carries exactly one evidence tag in COMPRESSION.md § Decision 1's four-slot form.
  3. No rule's TRIGGER or DIRECTIVE is re-worded. Specifically: gate 14's ORDERING axis keeps its lexical tell list ("merged to main", "once this lands", "after release", "on main", "the published X") and its RELEASE-TRIO table inline and complete — both are directive, and the table survives while its @motir/cli / @motir/design-system incident moves.
  4. python3 prompts/plan-rules/COMPRESSION.measure.py reports the two packs at or under 300 and 310 lines respectively, down from 436 and 450. — AMENDED 2026-08-11, on the record, during this card's run. COMPRESSION.measure.py reports every attributable warrant span moved and the residue reported with its arithmetic; the packs land at 401 and 411, down from 436 and 450, and the miss against 300 / 310 is reported rather than bought by cutting a trigger.
    • Evidence for the amendment (rung 2, the record's own script, run before any change): COMPRESSION.measure.py prints a projected column, and for these two files it reads 333 and 313 — both ABOVE the as-written targets. Scoping the record's own classifier per pack: phase-deepen.md 436 lines / 116 sentence-grain warrant lines / 13 warranted rules ⇒ a perfect cut lands at 436 − 116 + 13 = 333; kind-leaf-deepen.md 450 / 149 / 12 ⇒ 313. AC 4 asked for 300 and 310, i.e. below the record's own projection in both cases, before a single evidence tag is inserted.
    • And the projection is itself optimistic by one line per warranted rule. § Decision 1 fixes the tag in a two-line form and prints it twice; COMPRESSION.conserve.py counts two (its TAG + PTR regexes); COMPRESSION.measure.py sets TAG_LINES = 1. Under the real form the floors are 346 and 325 — 46 and 15 above the as-written targets.
    • Why the achieved numbers sit above even those floors. The sentence grain marks any sentence containing a citation, which in these two packs is overwhelmingly TRIGGER: seven rule headings whose citation is the rule's own name, the data-seam pack's five-ways tell list, the estimation gate's calibration anchor and points→minutes table, the item-link mechanics (MOTIR-1403 / 845 / 850 are the mechanism, not the story), and gate 14's frozen-wording directive. § Decision 4 governs this explicitly: "If a pack cannot reach its number without cutting a trigger, a tell-list or a shape table, the target gives way, not the rule."
    • The arithmetic is the whole point of the amendment: 16 + 22 warrant spans (5 604 + 6 772 characters) moved verbatim, 8 + 10 evidence tags inserted, COMPRESSION.conserve.py green on both packs.
    • Filed as planning bug MOTIR-2702 — which also carries the larger finding, that § Decision 4's own 1 600-line worst-cell target is unreachable on the same arithmetic and MOTIR-2662 has not run yet. Lesson: notes.html #259. Prior instances of the same family: #258 (MOTIR-2687), #257.
  5. WARRANT CONSERVATION is proved in the PR body (invariant C): the diff contains only deletions whose text appears verbatim in the matching fixtures file, plus inserted tag lines matching the fixed form.
  6. DECISION.md, CLASSIFICATION.md and plan-rules.md are not edited.

Context refs

  • prompts/plan-rules/COMPRESSION.md — § Decision 1 (the tag form + the deletion test), § Decision 4 (the worst-cell target), invariant C.
  • prompts/plan-rules/phase-deepen.md · prompts/plan-rules/kind-leaf-deepen.md.
  • prompts/plan-rules/fixtures/ — the tree MOTIR-2659 created.