Status: Accepted (draft v0.1)
Date: 2026-07-04
Decision: policy maps path globs → named scopes; per-scope floors compute from commits touching the scope (diff paths); effective(C) = min(own(C), min effective(deps)) over the workspace graph via pluggable graph adapters; dependency cycles collapse to their SCC’s minimum.
Rationale: Requirement 3. Propagation is what makes scoping safe rather than cosmetic — without it, risk launders into shared libraries while consumers’ scopes stay pristine (the pre-transitive-analysis failure mode of dependency scanners). Side effect with correct incentives: shared packages become the highest-ROI place to spend human review, since their level multiplies across consumers. Component tags (auth/v1.4.0-t2.1) align with Go nested-module conventions and existing monorepo release tooling.
Rejected: scalar repo-wide trust (original default — punishes monorepos); scope independence without propagation (laundering); commit-message or CODEOWNERS-declared scoping (subjective; diff paths are ground truth).