diff --git a/CLAUDE.md b/CLAUDE.md
index 23ed4c5..21fae77 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -52,12 +52,27 @@ persistence) → GUI (Qt widgets that read the manager and connect to its signal
   hex `persistent_id` (folder containment via `parent_persistent_id`).
 
 - **`lintunes/storage/json_storage.py`** — the library is **multiple files** in
-  the data dir: `library.json` (all tracks), `library_metadata.json`, and one
-  `playlists/<persistent_id>.json` per playlist. Writes are atomic (`*.json.tmp`
+  the data dir: `library.json` (all tracks), `library_metadata.json`, one
+  `playlists/<persistent_id>.json` per playlist, and one
+  `plays/<machine-id>.json` per machine. Writes are atomic (`*.json.tmp`
   → rename). **`storage/conflict_resolver.py`** merges Syncthing
   `*.sync-conflict-*` files on startup: play counts take the max, edited fields
   take the newest, playlist membership takes the union.
 
+- **`lintunes/storage/play_journal.py`** — why play counts can't conflict. Since
+  Round 38 `library.json` holds only a **base** count and each machine owns
+  `plays/<machine-id>.json` with *its own* per-track totals; effective count =
+  base + the sum of every journal. Only the owner ever writes its journal, so
+  two machines never touch the same file — and finishing a track no longer
+  rewrites 15 MB, which is what handed Syncthing a conflict once per song.
+  Totals, not an append log, so there is no compaction step to double-count in.
+  `PlayJournal.load()` must be handed a library whose tracks still carry **base**
+  counts (the fresh load at startup, the `disk` copy inside `reload_from_disk`) —
+  folding an already-folded library promotes the effective count to base.
+  Mirror-image rule in `json_storage.save_tracks()`: it writes
+  `journal.base_fields(tid)`, never the `Track`'s effective count. Those two
+  places are the whole hazard; `tests/test_round38.py` pins both.
+
 - **`lintunes/library_manager.py`** — `LibraryManager(QObject)` owns the
   `Library`, is the single funnel for all mutations, and persists them
   **debounced** (3 s) with per-area dirty tracking (a play-count bump rewrites
diff --git a/TASKS.md b/TASKS.md
index 16a3936..5d551ed 100644
--- a/TASKS.md
+++ b/TASKS.md
@@ -14,10 +14,13 @@ When a round closes, move its finished items to `tasks-done.md`.
 Folder + web-mix export, audio.js/jQuery dropped for a dependency-free
 player, lossless-only conversion. Round 36 below is still open.
 
-## Round 36 — the merge rework (planned, Round 35 covered the speed half)
+## Round 38 (2026-08-20) — merge windows + play journals: done, see tasks-done.md
+
+## Round 36 — the merge rework (the playlist half is still open)
 
 The three symptoms in Round 35 shared a root cause; that round fixed the
-performance half. This is the correctness half.
+performance half. Round 38 took the play-count and dialog items; the two
+playlist items below are what's left.
 
 - [ ] **Playlist merges lose position.** `_merge_playlist` is a 2-way union with
       no common ancestor: it takes one side's order wholesale and *appends* the
@@ -31,13 +34,10 @@ performance half. This is the correctness half.
       So "most recently edited" often means "most recently resized". Give
       `Playlist` a `date_modified` bumped only in `_set_track_ids`, and merge on
       that.
-- [ ] **`max()` play counts discard concurrent plays.** Base 100, one machine
-      plays 5 (105), the other plays 3 (103) → merge keeps 105 and those 3 are
-      gone. Per-machine `plays/<machine-id>.json` journal (machine id kept
-      outside the synced dir): only its owner ever writes it, so play data can
-      never conflict, effective count = base + sum of journals, and `library.json`
-      stops being rewritten every 3 s during playback — which is what generates
-      the conflicts in the first place.
+- [x] **`max()` play counts discard concurrent plays.** Done as described —
+      per-machine `plays/<machine-id>.json` journals, machine id in the config
+      dir. Three plays on the real library now write 83 bytes instead of 15 MB,
+      and `library.json` isn't touched at all.
 - [x] **Quiet the merge dialog.** One dialog per session: each merge is folded
       in as a timestamped entry (newest first) instead of opening another
       window. Fifteen had stacked up. Kept the window for every merge rather
diff --git a/lintunes/__init__.py b/lintunes/__init__.py
index 87127ca..517314e 100644
--- a/lintunes/__init__.py
+++ b/lintunes/__init__.py
@@ -1,3 +1,3 @@
 """LinTunes — iTunes-style music library manager and player for Linux."""
 
-__version__ = "0.8.0"
+__version__ = "0.9.0"
diff --git a/lintunes/library_manager.py b/lintunes/library_manager.py
index 7174cfd..ad32e6d 100644
--- a/lintunes/library_manager.py
+++ b/lintunes/library_manager.py
@@ -12,6 +12,7 @@ from lintunes.importers.file_importer import organized_destination, unique_path
 from lintunes.models import Library, Playlist, PlaylistType
 from lintunes.perf import timed
 from lintunes.storage import json_storage
+from lintunes.storage.play_journal import PlayJournal
 from lintunes.undo import Command, UndoStack
 
 
@@ -53,6 +54,13 @@ class LibraryManager(QObject):
         self._dirty_playlists: set[str] = set()
         self._deleted_playlists: set[str] = set()
 
+        # Play counts live in per-machine journals, not library.json — see
+        # storage/play_journal. Folding them in here means the rest of the app
+        # keeps reading one plain track.play_count.
+        self.play_journal = PlayJournal()
+        self.play_journal.load(self.data_dir, self.library)
+        self._dirty_journal = False
+
         self._save_timer = QTimer(self)
         self._save_timer.setSingleShot(True)
         self._save_timer.setInterval(SAVE_DEBOUNCE_MS)
@@ -680,9 +688,14 @@ class LibraryManager(QObject):
         track = self.library.tracks.get(track_id)
         if not track:
             return
+        now = _utc_now_iso()
         track.play_count += 1
-        track.play_date_utc = _utc_now_iso()
-        self._dirty_tracks = True
+        track.play_date_utc = now
+        # Deliberately does NOT mark library.json dirty: a play only moves this
+        # machine's journal, so finishing a track no longer rewrites 15 MB (and
+        # no longer hands Syncthing a conflict once per song).
+        self.play_journal.record_play(track_id, now)
+        self._dirty_journal = True
         self._schedule_save()
         self.track_updated.emit(track_id)
         self._touch_smart({"play_count", "play_date_utc"})
@@ -691,9 +704,11 @@ class LibraryManager(QObject):
         track = self.library.tracks.get(track_id)
         if not track:
             return
+        now = _utc_now_iso()
         track.skip_count += 1
-        track.skip_date = _utc_now_iso()
-        self._dirty_tracks = True
+        track.skip_date = now
+        self.play_journal.record_skip(track_id, now)
+        self._dirty_journal = True
         self._schedule_save()
         self.track_updated.emit(track_id)
         self._touch_smart({"skip_count", "skip_date"})
@@ -801,10 +816,16 @@ class LibraryManager(QObject):
 
     def flush(self):
         self._save_timer.stop()
+        if self._dirty_journal:
+            self.play_journal.save(self.data_dir)
+            self._dirty_journal = False
+            self._record_sig(self.data_dir / "plays"
+                             / f"{self.play_journal.machine}.json")
         if self._dirty_tracks:
             with timed("flush library.json (%d tracks)",
                        len(self.library.tracks)):
-                json_storage.save_tracks(self.library, self.data_dir)
+                json_storage.save_tracks(self.library, self.data_dir,
+                                         self.play_journal)
             self._dirty_tracks = False
             self._record_sig(self.data_dir / "library.json")
         for pid in list(self._dirty_playlists):
@@ -831,6 +852,9 @@ class LibraryManager(QObject):
         playlists_dir = self.data_dir / "playlists"
         if playlists_dir.exists():
             files.extend(sorted(playlists_dir.glob("*.json")))
+        plays_dir = self.data_dir / "plays"
+        if plays_dir.exists():
+            files.extend(sorted(plays_dir.glob("*.json")))
         return files
 
     @staticmethod
@@ -877,6 +901,11 @@ class LibraryManager(QObject):
                      or self._dirty_metadata)
         dirty_pids = set(self._dirty_playlists)
         disk = json_storage.load_library(self.data_dir)
+        # Fold the journals onto the disk copy *before* reconciling. The tracks
+        # in memory already carry effective counts, so without this the merge
+        # would compare effective against base and the max() would win by an
+        # amount equal to this machine's journal.
+        self.play_journal.load(self.data_dir, disk)
 
         for tid, disk_track in disk.tracks.items():
             mem_track = self.library.tracks.get(tid)
diff --git a/lintunes/storage/conflict_resolver.py b/lintunes/storage/conflict_resolver.py
index a468867..481785d 100644
--- a/lintunes/storage/conflict_resolver.py
+++ b/lintunes/storage/conflict_resolver.py
@@ -66,6 +66,8 @@ def resolve_conflicts(data_dir: Path) -> list[ConflictSummary]:
                 summary = _merge_metadata(original_path, conflict_path)
             elif original_path.parent.name == "playlists":
                 summary = _merge_playlist(original_path, conflict_path)
+            elif original_path.parent.name == "plays":
+                summary = _merge_play_journal(original_path, conflict_path)
             else:
                 summary = _keep_newer(original_path, conflict_path)
         except Exception as e:  # never let a bad file abort startup/live reload
@@ -264,6 +266,37 @@ def _merge_playlist(original_path: Path, conflict_path: Path) -> ConflictSummary
     return ConflictSummary(name, "playlist", lines)
 
 
+def _merge_play_journal(original_path: Path, conflict_path: Path) -> ConflictSummary:
+    """Insurance only. A journal is owned by one machine, so it should never
+    conflict — but if a filesystem or clock oddity produces one, falling through
+    to _keep_newer would throw away a machine's entire play history. Take the
+    higher total per track instead."""
+    original = read_json(original_path)
+    incoming = read_json(conflict_path)
+    merged = dict(original)
+    changed = 0
+    for tid, entry in incoming.items():
+        mine = merged.get(tid)
+        if mine is None:
+            merged[tid] = entry
+            changed += 1
+            continue
+        combined = dict(mine)
+        for count in ("plays", "skips"):
+            combined[count] = max(mine.get(count, 0), entry.get(count, 0))
+        for stamp in ("last_played", "last_skipped"):
+            a, b = mine.get(stamp), entry.get(stamp)
+            combined[stamp] = a if b is None else (b if a is None else max(a, b))
+        if combined != mine:
+            changed += 1
+        merged[tid] = combined
+    write_json(original_path, merged)
+    return ConflictSummary(
+        f"plays/{original_path.name}", "plays",
+        [f"{changed} track(s) of play history reconciled (highest total kept)."]
+        if changed else ["No differences needed reconciling."])
+
+
 def _merge_metadata(original_path: Path, conflict_path: Path) -> ConflictSummary:
     summary = _keep_newer(original_path, conflict_path)
     summary.file, summary.kind = "library_metadata.json", "metadata"
diff --git a/lintunes/storage/json_storage.py b/lintunes/storage/json_storage.py
index 4540552..c5bf9a6 100644
--- a/lintunes/storage/json_storage.py
+++ b/lintunes/storage/json_storage.py
@@ -22,7 +22,7 @@ def save_library(library: Library, data_dir: Path):
             playlist_file.unlink()
 
 
-def save_tracks(library: Library, data_dir: Path):
+def save_tracks(library: Library, data_dir: Path, journal=None):
     data_dir.mkdir(parents=True, exist_ok=True)
     tracks_dict = {}
     for tid, track in library.tracks.items():
@@ -31,10 +31,26 @@ def save_tracks(library: Library, data_dir: Path):
             # Store paths relative to the data dir so the library is portable
             # across machines that sync the folder to different mount points.
             data["location"] = to_relative(data["location"], data_dir)
+        if journal is not None:
+            # A Track carries the *effective* play count (base plus every
+            # machine's journal). library.json must get the base back — writing
+            # the effective value here would fold this machine's plays into the
+            # base and count them twice on the next load. See play_journal.
+            _write_base_play_fields(data, journal.base_fields(tid))
         tracks_dict[str(tid)] = data
     write_json(data_dir / "library.json", tracks_dict)
 
 
+def _write_base_play_fields(data: dict, base: dict):
+    """Put the base play fields back into a to_dict() result, keeping to_dict's
+    rule that a default-valued field is omitted rather than stored."""
+    for field, value in base.items():
+        if value in (0, None, ""):
+            data.pop(field, None)
+        else:
+            data[field] = value
+
+
 def save_metadata(library: Library, data_dir: Path):
     data_dir.mkdir(parents=True, exist_ok=True)
     metadata = {
diff --git a/lintunes/storage/play_journal.py b/lintunes/storage/play_journal.py
new file mode 100644
index 0000000..8df72f5
--- /dev/null
+++ b/lintunes/storage/play_journal.py
@@ -0,0 +1,191 @@
+"""Per-machine play journals — the reason play counts can no longer conflict.
+
+``library.json`` used to carry the live play count, so finishing a track
+rewrote all 15 MB of it. Both machines did that, so Syncthing saw two edits to
+one big file between syncs and produced a conflict file roughly once per song.
+The merge then took ``max()`` of the two counts, silently discarding whichever
+side had played less.
+
+Now ``library.json`` holds only a **base** count, and each machine additionally
+owns exactly one file ``<data_dir>/plays/<machine-id>.json`` recording *its own*
+per-track totals. Effective count = base + the sum of every journal.
+
+Three properties fall out of that:
+
+* **No conflict is possible.** A journal file is only ever written by the
+  machine that owns it, so two machines never touch the same file.
+* **No compaction step.** The journal stores running totals, not an append log,
+  so it is bounded by "tracks this machine has played" and never needs folding
+  back into the base — which is what would risk double-counting.
+* **Version skew is safe.** A machine still on older code keeps bumping its own
+  base in ``library.json``; we never write base, so the sum stays correct. When
+  it updates, its already-counted plays stay frozen in base and it starts its
+  own journal.
+
+The machine id lives in the config dir, deliberately *outside* the synced data
+dir — a synced id would make two machines share one journal.
+"""
+import uuid
+from pathlib import Path
+
+from lintunes.config import config_path
+from lintunes.storage.json_storage import read_json, write_json
+
+
+JOURNAL_DIR = "plays"
+
+# The four Track fields a journal contributes to. Everything else about a track
+# still lives in library.json.
+PLAY_FIELDS = ("play_count", "skip_count", "play_date_utc", "skip_date")
+
+
+def machine_id() -> str:
+    """This machine's stable id, created on first use. Kept next to config.json
+    (not in the data dir) so Syncthing never copies it to the other machine."""
+    path = config_path().parent / "machine_id"
+    try:
+        text = path.read_text(encoding="utf-8").strip()
+        if text:
+            return text
+    except OSError:
+        pass
+    new_id = uuid.uuid4().hex[:12]
+    path.parent.mkdir(parents=True, exist_ok=True)
+    path.write_text(new_id + "\n", encoding="utf-8")
+    return new_id
+
+
+def _newer(a, b):
+    """The later of two ISO-8601 stamps, either of which may be None."""
+    if a is None:
+        return b
+    if b is None:
+        return a
+    return a if a >= b else b
+
+
+class PlayJournal:
+    """The base counts from library.json plus every machine's journal.
+
+    ``load()`` snapshots the base off the freshly loaded Tracks and then folds
+    the journals in, so the rest of the app keeps reading one plain
+    ``track.play_count``. Only the save path needs to know the difference —
+    see ``base_fields``.
+    """
+
+    def __init__(self, machine=None):
+        self.machine = machine or machine_id()
+        self._base = {}     # track_id -> {field: value} as stored in library.json
+        self._own = {}      # track_id -> {"plays", "skips", "last_played", "last_skipped"}
+        self._others = {}   # machine id -> same shape as _own
+        self.dirty = False
+
+    # ---- loading ----
+
+    def load(self, data_dir: Path, library):
+        """Read every journal in the data dir and fold them into ``library``.
+
+        ``library`` must be one whose tracks still hold the **base** counts —
+        the freshly loaded library at startup, or the disk copy during a reload.
+        Passing an already-folded library would promote the effective count to
+        base and double-count it.
+        """
+        self._base = {
+            tid: {f: getattr(track, f) for f in PLAY_FIELDS}
+            for tid, track in library.tracks.items()
+        }
+        # Our own journal is only ever written by us, so unsaved bumps in memory
+        # are always more current than the file — never re-read over them.
+        keep_own = self._own if self.dirty else None
+        self._own = {}
+        self._others = {}
+        journal_dir = Path(data_dir) / JOURNAL_DIR
+        if journal_dir.is_dir():
+            for path in sorted(journal_dir.glob("*.json")):
+                if path.name.endswith(".tmp"):
+                    continue
+                try:
+                    entries = read_json(path)
+                except (ValueError, OSError):
+                    continue  # a half-synced journal must never abort startup
+                entries = {int(k): v for k, v in entries.items()
+                           if isinstance(v, dict)}
+                if path.stem == self.machine:
+                    self._own = entries
+                else:
+                    self._others[path.stem] = entries
+        if keep_own is not None:
+            self._own = keep_own
+        else:
+            self.dirty = False
+        self.apply(library)
+
+    def apply(self, library):
+        """Set every track's play fields to base + the sum of all journals."""
+        for tid, track in library.tracks.items():
+            added = sum(e[tid].get("plays", 0)
+                        for e in self._all_journals() if tid in e)
+            skipped_n = sum(e[tid].get("skips", 0)
+                            for e in self._all_journals() if tid in e)
+            base = self._base.get(tid)
+            if base is None:
+                # A track added after the snapshot. Its current count already
+                # includes anything recorded into the journal since, so back
+                # that out rather than baking it into the base.
+                base = {f: getattr(track, f) for f in PLAY_FIELDS}
+                base["play_count"] = (base["play_count"] or 0) - added
+                base["skip_count"] = (base["skip_count"] or 0) - skipped_n
+                self._base[tid] = base
+            plays = (base["play_count"] or 0) + added
+            skips = (base["skip_count"] or 0) + skipped_n
+            played = base["play_date_utc"]
+            skipped = base["skip_date"]
+            for entries in self._all_journals():
+                e = entries.get(tid)
+                if not e:
+                    continue
+                played = _newer(played, e.get("last_played"))
+                skipped = _newer(skipped, e.get("last_skipped"))
+            track.play_count = plays
+            track.skip_count = skips
+            track.play_date_utc = played
+            track.skip_date = skipped
+
+    def _all_journals(self):
+        yield self._own
+        yield from self._others.values()
+
+    # ---- recording ----
+
+    def record_play(self, track_id: int, when: str):
+        entry = self._own.setdefault(track_id, {})
+        entry["plays"] = entry.get("plays", 0) + 1
+        entry["last_played"] = when
+        self.dirty = True
+
+    def record_skip(self, track_id: int, when: str):
+        entry = self._own.setdefault(track_id, {})
+        entry["skips"] = entry.get("skips", 0) + 1
+        entry["last_skipped"] = when
+        self.dirty = True
+
+    # ---- saving ----
+
+    def base_fields(self, track_id: int) -> dict:
+        """What library.json should hold for this track. Writing the *effective*
+        count here instead would double-count it on the next load — this is the
+        one place in the change where that can go wrong."""
+        base = self._base.get(track_id)
+        if base is None:  # never seen at load time — nothing to preserve
+            return {"play_count": 0, "skip_count": 0,
+                    "play_date_utc": None, "skip_date": None}
+        return dict(base)
+
+    def save(self, data_dir: Path):
+        """Write this machine's journal only. Nothing else ever writes this file,
+        so it cannot conflict."""
+        journal_dir = Path(data_dir) / JOURNAL_DIR
+        journal_dir.mkdir(parents=True, exist_ok=True)
+        write_json(journal_dir / f"{self.machine}.json",
+                   {str(k): v for k, v in sorted(self._own.items())})
+        self.dirty = False
diff --git a/lintunes/sync_watcher.py b/lintunes/sync_watcher.py
index 559de92..87a0b4e 100644
--- a/lintunes/sync_watcher.py
+++ b/lintunes/sync_watcher.py
@@ -34,11 +34,15 @@ class LibrarySyncWatcher(QObject):
         the old inode, so we re-add paths after every event."""
         wanted = [self._data_dir,
                   self._data_dir / "playlists",
+                  self._data_dir / "plays",
                   self._data_dir / "library.json",
                   self._data_dir / "library_metadata.json"]
         playlists_dir = self._data_dir / "playlists"
         if playlists_dir.exists():
             wanted.extend(playlists_dir.glob("*.json"))
+        plays_dir = self._data_dir / "plays"
+        if plays_dir.exists():
+            wanted.extend(plays_dir.glob("*.json"))
         already = set(self._watcher.files()) | set(self._watcher.directories())
         to_add = [str(p) for p in wanted if p.exists() and str(p) not in already]
         if to_add:
diff --git a/tasks-done.md b/tasks-done.md
index a69985d..10a4136 100644
--- a/tasks-done.md
+++ b/tasks-done.md
@@ -1,5 +1,45 @@
 ## Done
 
+### Round 38 (2026-08-20) — One merge window, and play counts that can't conflict (v0.9.0)
+
+Fifteen "Synced changes merged" windows were stacked on the desktop. Two
+independent bugs, and the `.resolved/` backups said which mattered: every one of
+the last nine `library.json` merges was ~98% play counts (43-47 of ~45 changed
+tracks) plus exactly one real edit — the same single rating each time. The same
+~45 tracks disagreed in *every* merge and the count only crept down (48 → 44 over
+a day), so `max()` had been quietly discarding plays for at least that long.
+
+- [x] **One window, not one per merge.** `show_conflict_summary` built a fresh
+      `ConflictSummaryDialog` each time and only reassigned the attribute — the
+      old dialog stayed a live child of the window. Since
+      `check_for_external_changes` runs `resolve_conflicts` on every Syncthing
+      watch tick, that was one window per merge for the life of the process. The
+      dialog is now a session log: `add_event()` folds each merge in as its own
+      timestamped entry (newest first), the restore button names the merge it
+      would undo, and `finished` clears MainWindow's reference so a close lets
+      the next merge open a fresh one.
+- [x] **Per-machine play journals.** `library.json` now holds only a base count;
+      each machine owns `plays/<machine-id>.json` with its own per-track totals,
+      and the effective count is base + the sum of all journals. Only the owner
+      writes its journal, so play data can't conflict; totals rather than an
+      append log, so there's no compaction step to double-count in; and a machine
+      still on older code keeps bumping its own base, which stays additive with
+      our journal, so the version-skew window is safe too. On the real 21k
+      library: three plays wrote **83 bytes** and left `library.json` untouched,
+      where before each one rewrote 15 MB.
+- [x] **The two hazards, pinned by tests.** `save_tracks` must write the base and
+      `PlayJournal.load` must be handed a base-valued library — the second is why
+      `reload_from_disk` folds the journals onto the `disk` copy *before*
+      reconciling, or the max() against the effective in-memory count would win
+      by exactly this machine's journal.
+- [x] Journal conflicts merge by highest total rather than newest-wins. They
+      shouldn't be possible; if a filesystem oddity makes one, falling through to
+      `_keep_newer` would discard a machine's whole history.
+
+Deliberately not done: lossless merges still open the window rather than
+demoting to a status-bar line. With journals landing, a merge stops being
+routine — the ones left are real edits, and worth seeing.
+
 ### Round 37 (2026-08-19) — Export a playlist: a folder, or a whole website (v0.8.0)
 
 `File → Export Playlist…`, and the same item on a playlist's right-click menu.
diff --git a/tests/test_round38.py b/tests/test_round38.py
index 57da872..3f5369c 100644
--- a/tests/test_round38.py
+++ b/tests/test_round38.py
@@ -99,3 +99,213 @@ class TestWindowReuse:
         host.show_conflict_summary([_summary()])
         assert host._conflict_dialog is not None
         assert host._conflict_dialog is not first
+
+
+# --------------------------------------------------------------------------
+# Per-machine play journals
+# --------------------------------------------------------------------------
+
+from lintunes.models import Library, Track  # noqa: E402
+from lintunes.storage import json_storage  # noqa: E402
+from lintunes.storage.play_journal import PlayJournal  # noqa: E402
+
+
+@pytest.fixture
+def data_dir(tmp_path, monkeypatch):
+    """A data dir with a two-track library, and a machine id kept out of the
+    real ~/.config."""
+    monkeypatch.setenv("XDG_CONFIG_HOME", str(tmp_path / "config"))
+    d = tmp_path / "data"
+    library = Library(tracks={
+        1: Track(track_id=1, name="Thunder Peel", play_count=102),
+        2: Track(track_id=2, name="Black Blood", play_count=42),
+    })
+    json_storage.save_library(library, d)
+    return d
+
+
+def _base_on_disk(data_dir, tid=1, field="play_count"):
+    return json_storage.read_json(data_dir / "library.json")[str(tid)].get(field, 0)
+
+
+def _write_journal(data_dir, machine, entries):
+    (data_dir / "plays").mkdir(parents=True, exist_ok=True)
+    json_storage.write_json(data_dir / "plays" / f"{machine}.json", entries)
+
+
+class TestPlayJournal:
+    def test_machine_id_is_stable_and_outside_the_data_dir(self, data_dir, tmp_path):
+        from lintunes.storage.play_journal import machine_id
+        first = machine_id()
+        assert first and machine_id() == first
+        # A synced id would make both machines share one journal.
+        assert not (data_dir / "machine_id").exists()
+        assert (tmp_path / "config" / "lintunes" / "machine_id").exists()
+
+    def test_effective_count_is_base_plus_every_journal(self, data_dir):
+        _write_journal(data_dir, "machine-a", {"1": {"plays": 5, "last_played": "2026-08-20T10:00:00"}})
+        _write_journal(data_dir, "machine-b", {"1": {"plays": 3, "last_played": "2026-08-20T12:00:00"}})
+
+        library = json_storage.load_library(data_dir)
+        PlayJournal(machine="machine-a").load(data_dir, library)
+
+        # This is the case max() used to throw away: 102 + 5 + 3, not 107.
+        assert library.tracks[1].play_count == 110
+        assert library.tracks[1].play_date_utc == "2026-08-20T12:00:00"
+        assert library.tracks[2].play_count == 42  # untouched
+
+    def test_a_play_moves_the_journal_and_leaves_library_json_alone(self, data_dir):
+        library = json_storage.load_library(data_dir)
+        journal = PlayJournal(machine="machine-a")
+        journal.load(data_dir, library)
+
+        before = (data_dir / "library.json").read_bytes()
+        journal.record_play(1, "2026-08-20T18:42:00")
+        journal.save(data_dir)
+
+        assert (data_dir / "library.json").read_bytes() == before
+        assert json_storage.read_json(data_dir / "plays" / "machine-a.json") == {
+            "1": {"plays": 1, "last_played": "2026-08-20T18:42:00"}}
+
+    def test_save_tracks_writes_the_base_not_the_effective_count(self, data_dir):
+        """The one sharp edge: writing the folded total back into library.json
+        would fold this machine's plays into the base and count them twice."""
+        _write_journal(data_dir, "machine-a", {"1": {"plays": 5}})
+        library = json_storage.load_library(data_dir)
+        journal = PlayJournal(machine="machine-a")
+        journal.load(data_dir, library)
+        assert library.tracks[1].play_count == 107
+
+        json_storage.save_tracks(library, data_dir, journal)
+        assert _base_on_disk(data_dir) == 102
+
+        # And a second round trip must not drift either.
+        again = json_storage.load_library(data_dir)
+        PlayJournal(machine="machine-a").load(data_dir, again)
+        assert again.tracks[1].play_count == 107
+
+    def test_zero_base_is_omitted_from_library_json(self, data_dir):
+        """to_dict() omits default-valued fields; the base rewrite must too, or
+        21k tracks each grow a redundant "play_count": 0."""
+        library = json_storage.load_library(data_dir)
+        library.tracks[3] = Track(track_id=3, name="New Song")
+        journal = PlayJournal(machine="machine-a")
+        journal.load(data_dir, library)
+        journal.record_play(3, "2026-08-20T18:42:00")
+
+        json_storage.save_tracks(library, data_dir, journal)
+        assert "play_count" not in json_storage.read_json(
+            data_dir / "library.json")["3"]
+
+    def test_other_machine_on_old_code_does_not_double_count(self, data_dir):
+        """While the other machine still bumps its base in library.json, our
+        journal stays additive on top — no double count, no loss."""
+        _write_journal(data_dir, "machine-a", {"1": {"plays": 5}})
+        library = json_storage.load_library(data_dir)
+        PlayJournal(machine="machine-a").load(data_dir, library)
+        assert library.tracks[1].play_count == 107
+
+        # The other machine, still on old code, syncs in a base bumped by 3.
+        raw = json_storage.read_json(data_dir / "library.json")
+        raw["1"]["play_count"] = 105
+        json_storage.write_json(data_dir / "library.json", raw)
+
+        library = json_storage.load_library(data_dir)
+        PlayJournal(machine="machine-a").load(data_dir, library)
+        assert library.tracks[1].play_count == 110
+
+    def test_unsaved_plays_survive_a_journal_reread(self, data_dir):
+        """A reload re-reads every journal; ours is only ever written by us, so
+        in-memory bumps must not be read over."""
+        library = json_storage.load_library(data_dir)
+        journal = PlayJournal(machine="machine-a")
+        journal.load(data_dir, library)
+        journal.record_play(1, "2026-08-20T18:42:00")
+
+        fresh = json_storage.load_library(data_dir)
+        journal.load(data_dir, fresh)  # what reload_from_disk does
+        assert fresh.tracks[1].play_count == 103
+        assert journal.dirty
+
+
+class TestManagerRoundTrip:
+    """End to end through LibraryManager, which is where a base/effective mix-up
+    would actually cost trav play counts."""
+
+    @pytest.fixture
+    def manager(self, data_dir, qapp):
+        from lintunes.library_manager import LibraryManager
+        return LibraryManager(json_storage.load_library(data_dir), data_dir)
+
+    def test_playing_does_not_rewrite_library_json(self, manager, data_dir):
+        before = (data_dir / "library.json").read_bytes()
+        manager.record_play(1)
+        manager.flush()
+
+        assert (data_dir / "library.json").read_bytes() == before
+        assert manager.library.tracks[1].play_count == 103
+
+    def test_a_play_survives_a_restart_exactly_once(self, manager, data_dir):
+        manager.record_play(1)
+        manager.flush()
+
+        from lintunes.library_manager import LibraryManager
+        restarted = LibraryManager(json_storage.load_library(data_dir), data_dir)
+        assert restarted.library.tracks[1].play_count == 103
+        assert _base_on_disk(data_dir) == 102
+
+    def test_an_edit_after_a_play_still_writes_the_base(self, manager, data_dir):
+        """The dangerous sequence: a play inflates the in-memory count, then an
+        unrelated edit flushes the whole library.json."""
+        manager.record_play(1)
+        manager.add_track(Track(track_id=9, name="Sev Beni Beni"))
+        manager.flush()
+
+        assert _base_on_disk(data_dir) == 102
+        assert manager.library.tracks[1].play_count == 103
+
+        from lintunes.library_manager import LibraryManager
+        restarted = LibraryManager(json_storage.load_library(data_dir), data_dir)
+        assert restarted.library.tracks[1].play_count == 103
+        assert 9 in restarted.library.tracks
+
+    def test_reload_does_not_inflate_the_count(self, manager, data_dir):
+        """reload_from_disk reconciles in-memory (effective) against disk (base);
+        folding the journals onto the disk copy first is what keeps max() honest."""
+        manager.record_play(1)
+        manager.flush()
+        manager.reload_from_disk()
+        assert manager.library.tracks[1].play_count == 103
+
+        manager.reload_from_disk()
+        manager.flush()
+        assert manager.library.tracks[1].play_count == 103
+        assert _base_on_disk(data_dir) == 102
+
+    def test_the_other_machines_plays_arrive_on_reload(self, manager, data_dir):
+        _write_journal(data_dir, "machine-other", {"1": {"plays": 7}})
+        manager.reload_from_disk()
+        assert manager.library.tracks[1].play_count == 109
+
+
+class TestJournalConflictInsurance:
+    """A journal has one writer, so it should never conflict. If one ever does,
+    falling through to newest-wins would discard a machine's whole history."""
+
+    def test_conflicting_journals_merge_by_highest_total(self, data_dir):
+        from lintunes.storage import conflict_resolver
+        _write_journal(data_dir, "machine-a", {"1": {"plays": 9, "last_played": "2026-08-20T10:00:00"},
+                                               "2": {"plays": 4}})
+        json_storage.write_json(
+            data_dir / "plays" / "machine-a.sync-conflict-20260820-184200-ABCDEFG.json",
+            {"1": {"plays": 3, "last_played": "2026-08-20T18:00:00"},
+             "3": {"plays": 6}})
+
+        summaries = conflict_resolver.resolve_conflicts(data_dir)
+        assert [s.kind for s in summaries] == ["plays"]
+
+        merged = json_storage.read_json(data_dir / "plays" / "machine-a.json")
+        assert merged["1"]["plays"] == 9                       # not 3
+        assert merged["1"]["last_played"] == "2026-08-20T18:00:00"
+        assert merged["2"]["plays"] == 4                       # kept
+        assert merged["3"]["plays"] == 6                       # gained