Start with a question about the system

A recipe turns structural evidence into a decision: choose the owner of a change, identify affected consumers, distinguish an alias from another implementation, or decide what needs independent verification.

Run the examples from the root of the repository you are studying. Replace example paths and names with real ones. Start with loct --version; consult the installed command's help when its grammar differs. Commands build or refresh Loctree analysis artifacts; they do not run your application or prove its behavior.

Choose your question

Your questionRecipeDecision it supports
Who owns this behavior?Map ownershipChoose the module and its boundaries before reading individual hits.
What will this change touch?Assess change impactIdentify represented consumers and plan verification.
One name, one entity, or several?Resolve identitySeparate definitions, re-exports, platform variants and copies.
What looks alive but has no consumers?Investigate dead codeBuild evidence for a removal candidate without deleting blindly.
Where can a dependency cycle be cut?Understand cyclesFind the actual coupling before choosing a refactor.
Which contracts connect the two sides?Follow runtime contractsInspect commands, events and routes within supported families.
Where does configuration come from?Investigate environment driftSeparate candidate conflicts from proven override relationships.
What has the team silenced?Read suppressionsIdentify exceptions and understand their reasons.

Read the evidence before acting

Validate an answer explains scope, counterexamples, test selection and how to assess a recipe. Evidence and open questions records where these recipes came from and which questions remain conditional.

Each recipe includes an expected answer shape, a decision, and a counterexample. Example output describes the information to look for; it is not a new benchmark result. A detector label is evidence to investigate, not permission to delete or deploy.

Recipe, raw material, silence

The original September 2026 exercise used three outcomes:

  • Recipe (PRZEPIS): the surface gives a useful answer, and verification supports it within a stated scope.
  • Raw material (SUROWIEC): relevant facts exist, but the answer requires joining artifacts or another query.
  • Silence (CISZA): the question is valid, but the available surface cannot answer it.

A recipe can become raw material on a different stack. A result with zero rows can be useful when coverage is known; it cannot establish absence outside that coverage.

A short working loop

  1. State the question and the decision you need to make.
  2. Use a map or relationship query to discover the relevant owner.
  3. Read definitions and literal references to check the interpretation.
  4. Look for one counterexample, especially across runtime or language boundaries.
  5. Record the answer, its scope, the decision it supports and what remains unknown.

Use structural tools because the relationship answers your question. Literal search remains useful for checking the map and reading the local detail.