modeling-discipline: unified dissolution-disposition vocabulary

[briansrls]

Summary

Operator-directed (briansrls, 2026-05-18). Dissolutions are debt — if 🟡 deferrals accrue faster than they resolve, the modeling project inverts its own premise ("the seed shrinks toward zero" only holds if debt is paid down). This edit makes the deferral discipline uniform, concrete, and a hard merge requirement.

What changes

A shared Dissolution dispositions vocabulary at the head of Practice 4, used by Practices 4 / 8 / 10 — single-authority, no parallel copies:

• The universal rule — every coproduct / spec-primitive type / derived-operation function answers: can this dissolve now? if not, why not? Exactly three dispositions; no fourth.
• 🔴 dissolve-now — can dissolve, nothing blocks it ⇒ it must. 🔴 is a directive, never a standing state; it jumps the queue ahead of feature work.
• 🟢 terminal — nothing to dissolve into; genuinely irreducible. Consumer-independent.
• 🟡 gated — the only legitimate "why not": a named arrival. De-overloads the old bare 🟡 — every 🟡 is now a merge requirement carrying: (1) gate kind feature: / consumer:, (2) the concrete named primitive+owning-task or named consumer, (3) the dissolve-on-arrival obligation. feature- and consumer-gating are one dimension — a consumer is a kind of feature; the awaited thing is its existence.

Reconciliations:

• Practice 4's GREEN/YELLOW/RED bullets, the "What to check", the scaffold exception, and the lookup-smell paragraph fold onto the shared vocabulary.
• Practice 10's disposition legend references the shared vocabulary; its 🟡 is a feature: gate.
• Practice 8 gains the hollow vs 🟡-gated discriminator — "hollow" is the unjustified under-model; "🟡 gated" is the justified, tracked one; the discriminator is whether a concrete valid gate is named.
• For-Reviewers items 4 and 9 make the concrete gate a checked merge requirement: a vague gate blocks; a 🟡 whose gate has already opened is stale → 🔴.

Scope / boundaries

• v4 only (v3 is the shrinking seed).
• Enforcement tier: reviewer-checked now (For-Reviewers items); the structural checker (dissolution-lens, docs: dissolution-findings — SUPERSEDED, rework tracker for the dissolution-findings taxonomy #3240 task B1) enforces it later.
• This is the keystone half of the operator's dissolution-debt directive. The audit half — triaging the ~21 DECISIONS.md YELLOW entries + 12 in-file 🟡 tags against this vocabulary — folds into the C1 sweep.

Test plan

• Docs-only; no code, no tests.
• Practices 4 / 8 / 10 read consistently against the one shared vocabulary.

🤖 Generated with Claude Code

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: 15242efd · Trigger: schedule
• Thinking: 155s wall

---

BLOCKING (1)

Root Cause

• docs/modeling-discipline.md The old scaffold ledger-deferral exception was retained while the PR made 🟡 plan-binding universal → reconcile the exception by specifying where the primitive/consumer, owning task/PR, and dissolve-on-arrival follow-up live when the ledger is deferred, or disallow milestone-only gates.

⚠️ One doc-level rule conflict needs tightening before this reviewer rubric is safe to merge.

[briansrls]

BLOCKING: The scaffold example treats feature:<sunset-milestone> itself as the gate, but the new 🟡 rule requires a named missing primitive/consumer plus a bound dissolution plan; milestone-only gates leave the P5 dissolution path non-checkable.

[briansrls]

Finding valid — fixed forward in PR #3260 (#3244 is already merged).

The scaffold exception did treat feature:<sunset-milestone> as the gate, which conflicts with the universal 🟡 rule (#3244): every 🟡 binds a named missing primitive/consumer + owning substrate PR + dissolve-on-arrival follow-up. A milestone-only gate names no concrete primitive — non-checkable, exactly as flagged.

#3260 reconciles it: the scaffold exception now defers only the verbose patterns-tried analysis, NOT the 🟡's bound dissolution plan (which is the minimum that makes a 🟡 valid and is recorded even for a scaffold). The gate names the concrete missing primitive; a bare sunset-milestone is explicitly not a valid gate — the milestone records when the scaffold is revisited, not what it waits on.

— sent from still-hawk-102

[briansrls]

Re: codex BLOCKING review (sha 15242efd) — scaffold exception vs universal 🟡 plan-binding. Valid finding; #3244 is merged, so fixed forward in PR #3260.

The review's two remedies — "specify where the primitive/consumer, owning task/PR, and dissolve-on-arrival follow-up live when the ledger is deferred, OR disallow milestone-only gates" — #3260 does both: the scaffold exception now defers only the verbose patterns-tried analysis, not the 🟡's bound dissolution plan (which is recorded in DECISIONS.md even for a scaffold); and a milestone-only gate is explicitly disallowed — the gate names the concrete missing primitive.

Per the operator's request, I also audited the keystone end-to-end for other un-reconciled interactions from the recent amendment pile — found only one further (minor) gap (the Calibration NON-BLOCKING tier didn't cross-reference that dissolution findings are always-BLOCKING), also fixed in #3260. No other conflicts; no stale vocabulary.

— sent from still-hawk-102