docs: drop verification sections from decision files

CLAUDE.md

@@ -24,7 +24,7 @@ go test ./internal/ -run TestGolden -update      # regenerate golden files after
 | `docs/SPEC.md` | behavior the user approved: commands, flags, JSON, safety levels, ecosystem catalog, config | behavior changes |
 | `docs/ARCHITECTURE.md` | module boundaries, cross-module contracts, adding an ecosystem | a boundary or contract changes |
 | `plugins/devclean/skills/devclean/SKILL.md` | the Claude Code skill that runs the binary for the user | commands, flags, JSON fields or safety gates change |
-| `docs/decisions/` | one costly-to-reverse choice per file: decision, verification, rationale, rejected alternatives | such a choice is made or reversed |
+| `docs/decisions/` | one costly-to-reverse choice per file: decision, rationale, rejected alternatives | such a choice is made or reversed |
 | `CHANGELOG.md` | release sections written by release-please from PR titles | never by hand (`docs/decisions/release-please.md`) |
 
 - Update docs in the same commit as the code.

docs/decisions/clone-aware-size.md

@@ -9,15 +9,6 @@
 
 The per-file block sum is kept as `allocated_size` when larger.
 
-## Verification
-
-```bash
-go test ./internal/scanner/ -run 'Clone|Pnpm' -v
-go test ./internal/output/ -run SharedBlocks -v
-```
-
-Measured 2026-10-02: Chrome copies 11.3 GB → 0.76 GB; a pnpm `node_modules` 448 MB → 3.9 MB; home scan time unchanged (~60 s).
-
 ## Rationale
 
 - Sizes that are mostly shared blocks (Chrome copies ~15×) lead users to delete for space that never comes back.

docs/decisions/hf-shared-blobs.md

@@ -4,13 +4,6 @@
 
 Shared blobs are sized and deleted by their `<hash>.refs` manifests with huggingface_hub's rules (`utils/_shared_blobs.py`): a reference counts only while its symlink still points at the blob, and a blob whose manifest is unreadable or whose `<hash>.lock` is held is never deleted.
 
-## Verification
-
-```bash
-go test ./internal/scanner/ -run 'TestLLMScanner_HF' -v
-go test ./internal/cleaner/ -run 'Release' -v
-```
-
 ## Rationale
 
 - Model directories hold only symlinks, so without the store GB of weights went unreported and deleting a model freed nothing.

docs/decisions/one-filesystem.md

@@ -4,13 +4,6 @@
 
 Finding, sizing and the pre-delete check walk through `fstree`, which stays on the root's filesystem (like `du -x`). A directory on another device, of unknown device, or unreadable is not entered and is returned as skipped. Path removal (including dry-run) is refused when `fstree` skipped any directory inside the item. To scan a mount, pass it as `--path`.
 
-## Verification
-
-```bash
-go test ./internal/fstree/ -run 'OtherFilesystem|WithoutDevice|CheckOneFilesystem' -v
-go test ./internal/cleaner/ -run 'TestRefusesUncheckedTree' -v
-```
-
 ## Rationale
 
 - Mounts under the root are not this machine's reclaimable disk. Example: Xcode CoreDevice's `~/Library/Developer/CoreDevice/DeviceFS`, measured at ~4 s per directory read.

docs/decisions/prune-non-project-trees.md

@@ -11,12 +11,6 @@ The walk engine skips these trees whole — no artifact matching, no project con
 
 When adding an ecosystem or rule, scan that tool's real SDK/toolchain checkout and verify what gets flagged. A scanner's own false positive is fixed in that scanner's PR.
 
-## Verification
-
-```bash
-go test ./internal/scanner/ -run 'TestWalkExcludesInstalledPackageTrees|TestFlutterWalk' -v
-```
-
 ## Rationale
 
 - Each package in an installed-package tree, including the runtime an app ships inside its bundle, carries its own marker (`package.json`, `setup.py`, `go.mod`). Without pruning it roots a project, and its shipped `dist/`, `__pycache__` or `vendor/` matches as `safe` output — deleting the `dist/` of the npm bundled in a GitHub Actions runner breaks the runner. A project's own `node_modules` matches as an artifact before the name check, so it is still reported.

docs/decisions/release-please.md

@@ -6,18 +6,9 @@ release-please keeps a release PR open on `main`. Merging it bumps the version,
 
 - Version: Conventional Commits, `feat` bumps minor, `fix`/`perf` bump patch; below 1.0 a breaking change bumps minor (`bump-minor-pre-major`).
 - Sections: `feat` → Added, `fix` → Fixed, `perf`/`revert` → Changed; other types are hidden.
+- The repository squash-merges only, with the PR title as the commit title, so the PR title is what release-please reads.
 - Token: the default `GITHUB_TOKEN`. Its tag does not trigger other workflows, so GoReleaser runs as a job of the release-please workflow. CI on the release PR waits for a manual approval because `GITHUB_TOKEN` opened it.
 
-## Verification
-
-```bash
-goreleaser check
-```
-
-CI runs the same check on every PR. The first release PR shows the generated version and section; its merge run shows the Release with assets.
-
-The repository squash-merges only, with the PR title as the commit title, so the PR title is what release-please reads.
-
 ## Rationale
 
 - Releases stalled for five months when version, changelog cut and tag were manual steps; an always-open release PR makes a release one merge.

docs/decisions/symlink-no-follow.md

@@ -4,14 +4,6 @@
 
 `fstree` never descends into a symlink and never matches one as an artifact, even when it is named like one (a symlinked `node_modules`, as pnpm and some monorepos produce). Only the scan root is read through, and only for `--path`. Deleting a symlink removes only the link; the cleaner's cross-filesystem trash copy recreates symlinks as links.
 
-## Verification
-
-```bash
-go test ./internal/fstree/ -run 'Symlink' -v
-go test ./internal/scanner/ -run 'Symlink' -v
-go test ./internal/cleaner/ -run 'TestCopyTree_PreservesSymlinkAsLink' -v
-```
-
 ## Rationale
 
 - A symlink's target is real content that lives on disk elsewhere, so following it would double-count that space and inflate the reported reclaimable total.