Fix flaky cut-isolation test; bring docs current on the hark relationship

test_cut_pending_isolates_per_episode_failures matched a bare "2" against
the full audio file path to target episode 2's failure — but pytest's
auto-numbered tmp_path can itself contain that digit, so the test failed
depending on run order. Now matches the deterministic per-episode filename.

README/CLAUDE.md/PLAN.md still described the hark relationship as an open
question; M5 was actually decided (library dependency, not a merge) — bring
them in line with what hark's own docs already say.

CHANGELOG.md

@@ -7,6 +7,18 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+### Fixed
+
+- `__version__` in `src/adscrub/__init__.py` was stuck at 0.1.0 across every
+  milestone release since `pyproject.toml`'s version was bumped each time but
+  this constant never was — the CLI's `--version` and every outbound
+  User-Agent header have claimed to be `adscrub/0.1.0` this whole time.
+- `test_cut_pending_isolates_per_episode_failures` matched a substring ("2")
+  against the full audio file path to distinguish episode 2's failure from
+  episode 1 — but pytest's auto-numbered `tmp_path` ("pytest-26", "pytest-102",
+  ...) can itself contain that digit, making the test flaky depending on run
+  order. Now matches the deterministic filename (episode id) instead.
+
 ## [0.4.0] - 2026-07-10
 
 ### Added

CLAUDE.md

@@ -54,19 +54,19 @@ episode works on any feed, at the cost of real per-episode compute.
 
 ## Relationship to hark
 
-Explicitly a candidate for later merging into `flan/hark` as a module — both projects
-fetch feeds, upsert episodes into SQLite, and re-host derived feeds for the same
-AntennaPod-stays-unmodified loop. Kept as a separate repo for now because the two
-pipelines (topic extraction from metadata vs. audio transcription/cutting) don't share
-meaningful code yet. Don't build permanence into this separateness (e.g. don't invent
-a distinct auth/web layer, deployment identity, or feed-registration UX that would
-just get thrown away on merge) — keep it a thin CLI + SQLite + pipeline, matching
-hark's own M0/M1 shape, so a later merge is a module import, not a rewrite.
+Resolved 2026-07-11 (PLAN.md's M5): stays a separate repo, separate CHANGELOG/SemVer,
+separate test suite. `flan/hark` depends on this one as a library (`uv` path
+dependency, editable) — hark's `episodes`/`shows`/`ad_segments` schema was shaped to
+match this project's own, so this package's schema-coupled functions
+(`pending_episodes`, `transcribe_episode`, `detect_pending`, `cut_pending`, ...) work
+unchanged when called with a `conn` from hark's database. hark's CLI
+(`chapters`/`transcribe`/`detect-ads`/`cut`) is a thin wrapper calling straight into
+this package — no code here is duplicated into hark.
 
-The full pipeline (ingest → chapters → transcribe → detect → cut → serve) is now
-built end-to-end as of 0.4.0, which is exactly the trigger PLAN.md's M5 names for
-actually making the merge-or-stay-standalone call — **that's the owner's decision, not
-something to resolve unprompted.** Don't preemptively start merging repos.
+Don't build features assuming this will get absorbed into hark's own source later —
+that already got tried once (a full copy-merge into `src/hark/`, wrong shape, reverted
+same day) and isn't the direction. Keep this repo's own CLI/pipeline/tests
+self-sufficient and runnable standalone, same as before.
 
 ## Conventions
 

README.md

@@ -72,13 +72,16 @@ Tests use local feed fixtures — no network.
 
 ## Relationship to hark
 
-This may end up merged into [hark](https://git.onetick.ninja/flan/hark) as a
-module rather than staying a permanently separate service — both projects fetch
-feeds, upsert episodes into SQLite, and re-host derived RSS feeds for the same
-AntennaPod-stays-unmodified loop. Kept separate for now since the two pipelines
-(topic extraction vs. audio transcription/cutting) don't share much code yet;
-don't over-invest in infrastructure that would need to be thrown away if they
-merge later.
+Resolved 2026-07-11 (see M5 in docs/PLAN.md): this stays a separate product.
+[hark](https://git.onetick.ninja/flan/hark) depends on it as a library (a `uv`
+path dependency, editable) rather than folding its source in — hark's own
+`episodes`/`ad_segments` schema was shaped to match this project's, so
+adscrub's schema-coupled functions (`pending_episodes`, `transcribe_episode`,
+`detect_pending`, `cut_pending`, ...) work unchanged against hark's database.
+`hark chapters`/`transcribe`/`detect-ads`/`cut` are thin CLI wrappers around
+this package. An earlier pass at the merge fully copied this source into
+`src/hark/`, which was the wrong shape and got reverted — see hark's
+CHANGELOG 0.4.0.
 
 ## AI use disclosure
 

docs/PLAN.md

@@ -79,16 +79,25 @@ Milestones. Each one ships something usable and gets a CHANGELOG version.
   commands. `$ADSCRUB_BASE_URL` must be set to wherever the podcast player can actually
   reach the container — `serve` prints a warning if left at the `localhost` default.
 
-## M5 — hark module decision
+## M5 — hark module decision (resolved 2026-07-11)
 
-- Now that M4 works end-to-end, the actual decision point: fold this into `flan/hark`
-  as a module (shared feed-ingest code, one deployed service) or keep it standalone?
-  **Owner call, not decided here** — don't pre-build shared infrastructure before this
-  is actually decided.
+- Decision: stays a separate product. `flan/hark` depends on this repo as a library
+  (`uv` path dependency, editable), not a source merge — hark's schema was shaped to
+  match this one so adscrub's schema-coupled functions work unchanged against hark's
+  database. `hark chapters`/`transcribe`/`detect-ads`/`cut` are thin CLI wrappers that
+  call straight into this package.
+- An earlier pass in the same session fully copied this source into `src/hark/`
+  (wrong shape — two products' worth of code entangled in one), pushed to hark's
+  main, then reverted via `git revert -m 1` once caught. See hark's CHANGELOG 0.4.0
+  for the full story.
+- Practical effect on this repo: still developed and versioned independently, own
+  CHANGELOG/SemVer, own test suite. hark's Docker build doesn't yet resolve the path
+  dependency (build context only has hark's own files) — a packaging gap noted in
+  hark's own docs/PLAN.md, not this repo's problem to solve.
 
 ## Open questions (owner input needed, don't block on these)
 
-- M5's hark-merge decision (see above).
+- ~~M5's hark-merge decision~~ Resolved 2026-07-11 (see above).
 - M3 currently defaults to `claude-opus-4-8`; revisit cost vs. accuracy on ad-span
   boundaries once it's run against real transcripts (a cheaper model may be plenty
   for a fairly mechanical "find the sponsor read" task).

tests/test_cut.py

@@ -172,7 +172,12 @@ def test_cut_pending_isolates_per_episode_failures(conn, tmp_path, monkeypatch):
     conn.commit()
 
     def fake_probe_duration(path):
-        if "2" in str(path):
+        # Match the episode-2 audio file by its deterministic name
+        # (data_dir/audio/<episode_id>.mp3), not a substring of the full
+        # path — tmp_path itself is pytest's auto-numbered temp dir
+        # ("pytest-26", "pytest-102", ...) and can coincidentally contain
+        # "2", which made this test flaky depending on run order.
+        if path.stem == str(ep2["id"]):
             raise RuntimeError("ffprobe failed")
         return 100.0