modeling-discipline: add Practice 11 — parameterize, don't duplicate + concept-home boundary

[briansrls]

Summary

The Pass A/B unification (PR #3443) caught a recurring carrier-duplication / parametric-pair pattern that no existing modeling practice named — every Pass A/B correction (solve_constraints + coercion fold + LawfulRewriteWitness → find_witness; 7 witnesses → 3 generic carriers; BootstrapWitness + FixedPointWitness → PromotionWitness; etc.) had the same shape underneath: two or more operations/carriers/witnesses declared as distinct, where one parameterized declaration would have done.

Every Wave-1 worker correction (PRs #3439 / #3440 / #3441) had the same shape downstream — the briefs encoded domain-named "primitives" the workers faithfully implemented, then the unification caught them at review.

The existing practices catch the implementation-time symptoms:

• Practice 4 — coproduct dissolution (flat enum with parametric structure)
• Practice 5 — single-authority metadata (one fact, one location)
• Practice 10's "carrier dissolution" — local coproduct cloning a std/ carrier

None of them catches the design-time cause: a domain-named declaration in the design doc that's actually a parameterization of an existing primitive. That gap is what seeded Worker A/B/C with mis-shaped briefs.

Practice 11 (NEW) names the meta-pattern.

What Practice 11 adds

Two sub-rules, both applying at the design-PR / brief-authoring layer:

11a — Parameterize, don't duplicate

Is this an instance of an existing one, with the per-instance variation expressed as a typed parameter?

Worked examples (drawn from real v4 corrections in PR #3443):

Domain-named pair/triple (the wrong shape)	Parameterized form (the right shape)	What the parameter is
solve_constraints, coercion fold, LawfulRewriteWitness	find_witness(facts, candidates, predicate, multiplicity_policy)	preservation predicate + multiplicity policy
CanonicalGroundingWitness, AcyclicityWitness, ClosedWorldDependencyWitness	StructuralPropertyWitness<P>	structural property P
Witness<homomorphism>, LawfulRewriteWitness	HomomorphismWitness<R>	preservation rule R
BootstrapWitness + FixedPointWitness	PromotionWitness (composed gate)	which two sub-checks compose
traverse_node, sequence_node, bind_outcome as primitives	fold_node + Outcome-threading algebra	algebra-as-data
3 MVP terminuses	1 MVP	(collapsed)
ProgramSchedulingEdge, TargetPlanSchedulingEdge, ArtifactSchedulingEdge	one DependencyEdge with DependencyKind label	the kind
DependencyGraph separate carrier	Edge-on-Node parameterized by DependencyKind	(carrier dissolves entirely)
TopologicalPlan as authored ledger	lens output (fold_node over Node)	(carrier dissolves to derived view)

11b — Concept-home boundary discipline

Before adding a field to a substrate file, confirm the field doesn't cross a boundary the file's identity is supposed to keep separate.

Canonical violation: NodeFileBinding in extdeps/file_system.dag — extdeps shouldn't know about Node; Node↔File provenance belongs in artifact/projection/ingest.

Why this is design-time meta-practice

Practices 1–10 are file-scale — a reviewer reads one diff and checks whether the diff conforms. Practice 11 is design-scale — a reviewer reads the brief that produced the diff and asks whether the brief itself names duplicates as primitives.

A Practice 11 violation in a design doc propagates as N file-scale violations across every worker the brief dispatches. The cost is asymmetric: catching it in the design doc deletes the violation once; catching it after dispatch costs N rounds of corrections + N worker contexts that need to be re-briefed.

Changes

• ## Ten Modeling Practices → ## Modeling Practices (was miscounting after Remove node override mocking mechanism from test framework #11)
• New entry in the Practice → invariant mapping bullet list (Practice 11 implements P2 + M2)
• Practice 10's carrier dissolution sub-case cross-refs Practice 11 for the parametric generalization
• New Practice 11 section (between Practice 10 and Calibration), with both sub-rules, worked-examples tables, and standard 🔴/🟡/🟢 dispositions
• Calibration section adds: "Practice 11 findings are always BLOCKING at the design-PR layer even when no implementation hunks exist"
• For Reviewers: new step 11 — design-PR review applies Practice 11 per declaration

Pure docs change. No code touched. No src/v4/ paths.

Test plan

• grep for stray Ten Modeling Practices references — none should remain
• Practice 11 cross-references Practice 10's carrier dissolution correctly
• Calibration's "always BLOCKING" addition reads consistently with Practice 10's same wording
• For Reviewers step 11 is reachable from the existing step 1–10 flow

Anchor

Drawn from the Pass A/B unification (PR #3443) corrections + Wave-1 worker analysis (PRs #3439 / #3440 / #3441). All worked-example rows are real corrections, not hypothetical.

🤖 Generated with Claude Code