## Summary Nine focused performance optimizations to stage-collision checking and lifecycle-state cleanup, extracted from project-specific performance work on a real, large snap build. Each commit is independently benchmarked with statistically significant results (Holm-corrected at family-wise alpha 0.05). Two related candidate changes were evaluated but excluded because they did not reach statistical significance on a targeted, isolated benchmark of the code they touch (a 2-line "reuse empty path" tweak measured 0.54% *slower*, p=0.796; an "avoid candidate directory stats" change measured 5.26% slower, p=0.149). ## Commits 1. `perf(executor): cache stage collision candidate inventories` — 86.18% faster collision checks (p=2.67e-7) 2. `perf(executor): validate changed stage candidates incrementally` — 45.13% faster on top of `#1` (p=1.75e-7) 3. `perf(executor): reuse resolved fileset paths` — 36.03% faster (p=5.49e-6) 4. `perf(executor): cache resolved relative path parents` — 18.71% faster on top of `#3` (p=0.000763) 5. `perf(executor): optimize fileset walk path construction` — 11.96% faster (p=0.00291) 6. `perf(executor): reuse lifecycle states during cleanup` — 26.04% faster stage cleanup (p=4.05e-6) 7. `perf(executor): copy only cleanup-mutated state sets` — 32.38% faster on top of `#6` (p=2.23e-6) 8. `perf(state_manager): use safe C loader for state YAML` — 52.67% faster (p=3.43e-7) 9. `perf(state_manager): cache parsed cleanup state YAML` — 81.10% faster, merged by hand with `#8` so both optimizations are retained together (p=1.26e-9) Each commit message includes full benchmarking methodology and figures (operation-level isolated benchmark plus the corresponding whole-build median clean-build/rebuild improvement). ## Testing - `pytest tests/unit` passes (only pre-existing, environment-specific failures remain: missing JDK for `ant`/`java`/`maven` plugin tests, missing `chisel` binary — unrelated to this change, present on `main` too). - `ruff check` and `ruff format --check` pass.