decompose-gates

Split along verification boundaries rather than implementation convenience: each piece must have its own pass/fail check that does not depend on other pieces being correct. State each piece as a falsifiable claim with a concrete deciding test, such as a test slice, log observation, or measurement.

Each lane brief should be a sufficient restart brief on its own, carrying its gate (falsifiable claim plus check), exact paths and symbols, observed evidence, in-scope and out-of-scope surfaces, success criteria, required validation, and stop/fallback conditions. A fresh agent must be able to continue from the brief and on-disk report alone.

Order by information yield: test first the smallest shared assumption whose failure would invalidate repeated downstream units. Skip the ramp when prior evidence or a deterministic tool already proves the unit shape, and keep independent work moving in parallel.

Pieces that are too small to fail meaningfully create overhead; micro-slicing such as one-boolean extractions can grow the god-files it was meant to shrink. Each piece should own a whole responsibility end to end, including analysis, change, tests, and validation.

For compatibility work, decompose the transition into independently falsifiable prepare/read, activate/write, migrate/backfill, mixed-version or rollback, and contract/removal claims only where those phases are reachable. Each claim names its exact released baseline and old/new producer-consumer direction.