rmc-test-vs-prod

name	rmc-test-vs-prod
description	Rust test vs production split.
argument-hint	<qualified-symbol-name>
allowed-tools	Read, mcp__rust-code-mcp__*

Rust test vs production analysis

Layer 8 categorizes every reference as Read / Write / Test / Other. The Test split is the load-bearing one for "what's only used by tests?" Scope: single symbol or symbol family.

For broader refactor decisions, hand off to rmc-refactor-plan. For method-by-method audits on one type, use rmc-method-api.

Scope — single symbol or symbol family

Prerequisites

build_hypergraph(directory=<absolute-path>)

Workflow

Step 1. Pull the breakdown

who_uses_summary(directory=..., target=<qualified_name>)

Each row has a category_breakdown with Test and Other counts (some payload shapes also include Read and Write sub-counts).

Step 2. Classify

Pattern	Verdict
All rows 100% Test	Test fixture / builder. Demote to #[cfg(test)] or move to dev-deps.
All rows 100% Other (Test=0)	Production-only. Critical path. High refactor risk.
Mixed Test + Other	Legitimate API. Both tested and used.
Test >> Other	Either under-used in production or over-tested in isolation.

Step 3. Read vs Write encapsulation check

When payloads include Read and Write sub-counts:

• Many readers + few writers = good encapsulation.
• Many writers = diffuse invariants — the symbol's value flows through too many places.

Targeted recipes

Recipe — "Test-only constructor audit"

For each Type::new (and with_* / from_* constructors), run who_uses_summary. Rows 100% Test = builder used only by tests. Move to test fixtures.

Recipe — "Production-only methods"

Filter module_tree to methods. For each, who_uses_summary. All-Other rows = critical path. Annotate as "high touch risk" in PR descriptions.

Recipe — "Mostly-tested public API"

For pub items, who_uses_summary. Test >> Other (e.g. 30 Test, 2 Other) usually means the symbol is tested in isolation but barely consumed in production — under-used or over-tested.

Decision frames

Finding	Action
100% Test fan-in for a pub item	Demote to pub(crate) + #[cfg(test)]
100% Test fan-in for a pub constructor	Move to a test-fixtures crate / tests/common
100% Test for an entire trait's methods	Trait is test-only; consider deleting and using concrete types in tests
Heavy Write counts on a shared type	Diffuse invariants; consider &mut self API audit
Mixed but Test >> Other	Over-tested in isolation; verify production callers are real users

Pattern reference

Signal	Means
Test=N, Other=0 for Type::new	Constructor used only by tests
Test=0, Other=N for a trait method	Production-only API; refactor with care
Test=N, Other=M, Read=X, Write=Y with Y >> X	Many writers — invariants are diffuse
Test rows cluster in <crate>::tests::common	Existing test-fixtures module — good convergence target

Output format

Per symbol:

Symbol: <qualified_name>
Verdict: <test-only | production-only | mixed | over-tested>
Test rows: <n>; Other rows: <m>
Top consumers (Test): <list>
Top consumers (Other): <list>
Recommended action: <demote / cfg(test) / move-to-fixtures / keep>

Limitations

• The Test/Other category is heuristic — modules under tests/, *_test, or #[cfg(test)] items count as Test. Integration tests outside conventional paths may misclassify.
• Read/Write sub-counts are not present on every payload shape; rely on Test/Other for the core verdict.
• who_uses_summary excludes import-only references — pair with who_imports if you suspect a symbol is imported but never used.