claude-code-sessions 0.16.5__tar.gz → 0.17.0__tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (41) hide show
  1. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/LICENSE +21 -21
  2. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/PKG-INFO +185 -12
  3. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/README.md +184 -11
  4. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/claude_code_sessions.egg-info/PKG-INFO +185 -12
  5. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/claude_code_sessions.egg-info/SOURCES.txt +3 -0
  6. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/claude_code_sessions.py +3125 -22
  7. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/claude_code_sessions_gui.py +257 -85
  8. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/pyproject.toml +50 -50
  9. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/tests/test_alignment.py +144 -144
  10. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/tests/test_cli.py +172 -172
  11. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/tests/test_containment.py +71 -71
  12. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/tests/test_discovery.py +48 -48
  13. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/tests/test_encoding.py +41 -41
  14. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/tests/test_engine.py +274 -274
  15. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/tests/test_golden.py +18 -18
  16. claude_code_sessions-0.17.0/tests/test_gui_dpi.py +171 -0
  17. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/tests/test_gui_models.py +153 -0
  18. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/tests/test_helpers.py +39 -39
  19. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/tests/test_journal.py +87 -87
  20. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/tests/test_journal_reroot.py +8 -0
  21. claude_code_sessions-0.17.0/tests/test_realign.py +1826 -0
  22. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/tests/test_rows.py +73 -73
  23. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/tests/test_scaffold.py +18 -18
  24. claude_code_sessions-0.17.0/tests/test_split_rows.py +654 -0
  25. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/tests/test_transcripts.py +24 -24
  26. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/claude_code_sessions.egg-info/dependency_links.txt +0 -0
  27. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/claude_code_sessions.egg-info/entry_points.txt +0 -0
  28. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/claude_code_sessions.egg-info/top_level.txt +0 -0
  29. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/setup.cfg +0 -0
  30. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/tests/test_anonymize.py +0 -0
  31. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/tests/test_commands.py +0 -0
  32. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/tests/test_converge.py +0 -0
  33. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/tests/test_data_root.py +0 -0
  34. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/tests/test_duplicate_transcripts.py +0 -0
  35. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/tests/test_gui_destination.py +0 -0
  36. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/tests/test_gui_env_rebuild.py +0 -0
  37. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/tests/test_plan_move.py +0 -0
  38. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/tests/test_recover.py +0 -0
  39. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/tests/test_retitle.py +0 -0
  40. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/tests/test_sync.py +0 -0
  41. {claude_code_sessions-0.16.5 → claude_code_sessions-0.17.0}/tests/test_undo.py +0 -0
@@ -1,21 +1,21 @@
1
- MIT License
2
-
3
- Copyright (c) 2026 Craig Stoller
4
-
5
- Permission is hereby granted, free of charge, to any person obtaining a copy
6
- of this software and associated documentation files (the "Software"), to deal
7
- in the Software without restriction, including without limitation the rights
8
- to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
9
- copies of the Software, and to permit persons to whom the Software is
10
- furnished to do so, subject to the following conditions:
11
-
12
- The above copyright notice and this permission notice shall be included in all
13
- copies or substantial portions of the Software.
14
-
15
- THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16
- IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17
- FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
18
- AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19
- LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20
- OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21
- SOFTWARE.
1
+ MIT License
2
+
3
+ Copyright (c) 2026 Craig Stoller
4
+
5
+ Permission is hereby granted, free of charge, to any person obtaining a copy
6
+ of this software and associated documentation files (the "Software"), to deal
7
+ in the Software without restriction, including without limitation the rights
8
+ to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
9
+ copies of the Software, and to permit persons to whom the Software is
10
+ furnished to do so, subject to the following conditions:
11
+
12
+ The above copyright notice and this permission notice shall be included in all
13
+ copies or substantial portions of the Software.
14
+
15
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16
+ IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17
+ FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
18
+ AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19
+ LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20
+ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21
+ SOFTWARE.
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: claude-code-sessions
3
- Version: 0.16.5
3
+ Version: 0.17.0
4
4
  Summary: Unofficial tool for Claude Code sessions on disk: make a session visible under your other Claude account on the same machine, or move one between project folders. Command line or GUI. Journaled, reversible, fails closed.
5
5
  Author: Craig Stoller
6
6
  License: MIT
@@ -157,7 +157,7 @@ read. See `docs/internals.md`, RULING 7.
157
157
 
158
158
  ## Before any move: close the Claude app.
159
159
 
160
- `move`, `undo`, `recover`, `repoint --apply`, `new-row --apply`, and every mutating `sync`
160
+ `move`, `undo`, `recover`, `repoint --apply`, `new-row --apply`, `realign --apply`, and every mutating `sync`
161
161
  route (`--apply`, `undo` of a completed sync, `recover --back` on a stuck one) all refuse
162
162
  while they can see the Claude desktop app running - whoever the identity files currently say
163
163
  is signed in. That includes `undo` of a completed `repoint` or `new-row`, and `recover` on an
@@ -183,7 +183,7 @@ itself, and the guard counts it - if a refusal names that process, fully exit th
183
183
 
184
184
  ## Usage
185
185
 
186
- Twelve commands. All mutating commands default to a dry run; add `--apply` to execute.
186
+ Thirteen commands. All mutating commands default to a dry run; add `--apply` to execute.
187
187
  (`trust-signed-helper` is the opt-in described above; it changes a setting, not your
188
188
  sessions.)
189
189
 
@@ -204,8 +204,9 @@ claude-code-sessions list gate
204
204
  ```
205
205
 
206
206
  **`doctor`** - read-only health report (stale locks, unresolved operations, orphaned rows,
207
- encoding-scheme ambiguity, a synthesized row that has since vanished, and **conversations no
208
- account's sidebar points at**):
207
+ encoding-scheme ambiguity, a synthesized row that has since vanished, **conversations no
208
+ account's sidebar points at**, and **conversations whose accounts open different copies of one
209
+ transcript**):
209
210
 
210
211
  ```
211
212
  claude-code-sessions doctor
@@ -219,6 +220,30 @@ Resuming a session while signed into a different account repoints its row, which
219
220
  conversation you were in an hour ago reachable from no sidebar at all. That is what this
220
221
  ranking is for, and `repoint` is how you put it back.
221
222
 
223
+ The copies check covers a case every other number misses. A conversation's transcript can end
224
+ up in two project folders, usually after a session moves into a second worktree. Each account's
225
+ row names a folder through its `cwd`, and the app opens the copy in that folder, even a stale
226
+ one. So one sidebar can show a conversation as it was weeks ago while another shows it today,
227
+ and `alignment` still counts it complete. `doctor` lists each such conversation, which
228
+ copy each account's row opens, and how each pair of copies relates: one behind the other (one
229
+ history, one copy stopped growing), the same conversation, byte-identical, or diverged (each
230
+ holds turns the other lacks). It compares the bytes first and then reads what lies past the
231
+ bytes two copies share, because the app writes bookkeeping records (`tools-used` when it opens
232
+ a copy, `cost-state` when it quits) to whichever copy it has open, the stale one included. A
233
+ copy ahead only by those records holds the same conversation; a stale copy the app has opened
234
+ and quit is still behind, although its bytes no longer match. Any other record type counts as
235
+ conversation. `doctor` exits 1 when an unarchived row opens an older or a diverged copy.
236
+ Identical copies and copies that differ only in bookkeeping are listed and do not change the
237
+ exit code. Neither does a row whose folder holds no copy at all. A row whose `cwd` names a main repository while the transcript sits in
238
+ one of its worktree folders is the app's normal placement, not a fault.
239
+
240
+ Deleting the stale copy is a catch-up, not a fix. On one conversation, measured on Desktop
241
+ 5.11492.5, opening a row whose folder held no copy made the app copy the transcript back into
242
+ that folder, so the split returns once the conversation is worked from one account and opened
243
+ from another. For the same reason `doctor` also lists, without counting them, conversations with
244
+ a single copy and a row whose folder holds none: opening that row starts a split. `realign`
245
+ (below) is the repair for both.
246
+
222
247
  **`alignment`** - read-only scoreboard: how close your accounts are to holding one shared
223
248
  history. `doctor` answers "is anything broken"; this answers "how far apart are my sidebars,
224
249
  and which way is it moving":
@@ -237,8 +262,10 @@ Two of those actively trade against each other: making a conversation reachable
237
262
  both halves of a disagreeing pair in every sidebar under one name, which is a direct hit to
238
263
  *distinguishable*. A single "aligned: yes/no" would hide exactly the decision you have to make.
239
264
 
240
- It reads row files and transcript **filenames** only - no transcript content - so it stays fast
241
- on a store with gigabytes behind it. **It always exits 0** when it could produce the report. It
265
+ It reads row files and transcript **filenames and sizes** only - no transcript content - so it
266
+ stays fast on a store with gigabytes behind it. That is why its line on conversations whose
267
+ accounts open different copies only counts them, and how many differ in length; telling a stale
268
+ copy from a diverged one means reading both, which `doctor` does. **It always exits 0** when it could produce the report. It
242
269
  is a scoreboard, not a check: these numbers stay non-zero for as long as the work takes, and a
243
270
  command that exits 1 for months trains you to stop reading it.
244
271
 
@@ -579,6 +606,90 @@ stopped being redundant is presence it leaves alone. `doctor`'s vanished-row che
579
606
  converge-created rows the same way it watches `new-row`'s, with the same journal-retention
580
607
  window.
581
608
 
609
+ **`realign`** - point every account's row of a split conversation at the folder holding its
610
+ live transcript copy:
611
+
612
+ ```
613
+ claude-code-sessions realign
614
+ claude-code-sessions realign --apply
615
+ ```
616
+
617
+ These are the split rows `doctor` reports. A probe on Desktop 5.11492.5, run on one plain and
618
+ one worktree conversation, found that a row's `cwd` alone decides which copy it opens; that the
619
+ app rewrites the row's `folderPlace` to match `cwd`, and never writes `cwd` back; and that a
620
+ turn typed after `cwd` moved runs in the new folder, in the same transcript. So `realign`
621
+ rewrites `cwd`, and `folderPlace.cwd` and `folderPlace.realPath` where the row has a
622
+ `folderPlace` (it never adds one), and leaves every other key alone, including `originCwd` and
623
+ the title. The rewritten rows then open the live copy. No row opens the stale copy any more, so
624
+ it moves into the operation's journal, with its sidecar folder, where `undo` can put it back
625
+ and `doctor` lists it. Without that step, `move`, `repoint`, `new-row` and `converge` would
626
+ keep refusing the conversation for its duplicated transcript. A copy that no row opened even
627
+ before `realign` stays where it is: that is the duplicated-transcript case, which `realign` did
628
+ not cause.
629
+
630
+ By default it plans conversations with a copy behind the live one, and rows whose folder holds
631
+ no copy. Copies that hold the same conversation (byte-identical, or different only in
632
+ bookkeeping records) are realigned only with `--include-identical`, onto the copy the most
633
+ recently active row opens. `--only` narrows to one conversation, resolved the way `converge
634
+ --only` is. A conversation is **held**, with the reason printed, when:
635
+
636
+ - its copies have diverged (each holds turns the other lacks), so which history to keep is
637
+ your decision;
638
+ - a copy no row opens holds turns the live copy lacks;
639
+ - a row to change has a worktree link (`worktreePath` or `worktreeName`). In the probe, moving
640
+ such a row's path keys either detached it from its worktree or left it unable to start;
641
+ - the live copy is reached only through the worktree placement, sits in a legacy-encoded
642
+ folder, or has no path on record that encodes to its folder;
643
+ - the paths that reach the live copy's folder disagree, or its directory no longer exists;
644
+ - an unfinished operation involves it, or anything about it changed between plan and apply.
645
+
646
+ `--apply` needs the desktop app fully closed, re-plans under its lock and writes only the
647
+ conversations whose rows, copies and target are unchanged, prints any it held at that point,
648
+ and exits 3 when any conversation was held. An interrupted run recovers in both directions:
649
+ `recover --forward` finishes from the journal, and is offered only when every row and copy it
650
+ would touch is as the journal expects; `recover --back` puts each conversation's copies and
651
+ rows back, and keeps realign's placement for a conversation whose copy cannot go back (its
652
+ place is taken, for instance), saying why. `undo` works per conversation. It puts back only
653
+ the keys realign changed, keeps whatever the app wrote to other keys since, and refuses a
654
+ conversation, naming the reason, when its rows no longer hold what realign wrote, now open
655
+ another conversation, were written by a later operation, or when the conversation has gained
656
+ turns since (bookkeeping records do not count). The other conversations go ahead, and a later
657
+ `undo --apply` retries the refused ones.
658
+
659
+ A copy realign sets aside stays in its journal until `undo` puts it back or `realign --release`
660
+ sends it to the Recycle Bin; `doctor` lists every copy held. Once a conversation has gained
661
+ turns, undo refuses it for good, which is what release is for:
662
+
663
+ ```
664
+ claude-code-sessions realign --release
665
+ claude-code-sessions realign --release --apply
666
+ ```
667
+
668
+ A copy is released only when every row of its conversation now opens one copy, that copy holds
669
+ the set-aside copy's whole conversation (the bookkeeping records `tools-used` and `cost-state`
670
+ aside, which carry no conversation), and its sidecar folder holds every file the set-aside
671
+ copy's does, with the same content: a copy that holds anything else the live one lacks stays in
672
+ the journal, with the reason printed. It goes to the Windows Recycle Bin, and the step is built
673
+ so that a permanent delete cannot happen silently. A shell that cannot recycle a file destroys
674
+ it instead. So release refuses a bin turned off (by you or by policy), one whose capacity the
675
+ registry does not record, one without room for the whole run without pushing out your older
676
+ deleted files (asked again just before each move), and a path too long for the shell, and it
677
+ asks Windows to warn rather than delete in any case it cannot see. After each move it checks
678
+ that a new bin record names the file, with its size and its contents present. It needs the
679
+ desktop app closed, since its checks read the rows and transcripts the app rewrites. A copy
680
+ whose release is interrupted is finished by the next run, and only after the same checks; once
681
+ its transcript is in the bin, only while the live copy still starts with everything it held
682
+ when the release began. One gap is stated rather than closed: a case the checks cannot see
683
+ rests on Windows' warning, whose behaviour beside the silent flags release uses is untested. Once
684
+ every copy of an operation is released, the journal ages the operation out like any other;
685
+ `undo` refuses a conversation whose copy was released or is part-way through release, and
686
+ `--json` gives each copy's outcome, a failed run included.
687
+
688
+ A released copy can be restored from the Recycle Bin by hand until the bin is emptied or
689
+ Windows clears it to make room. It comes back to its journal folder, not to its project folder;
690
+ the operation's record names where each part went. Windows only, like every other store mutation
691
+ here.
692
+
582
693
  ## Safety design
583
694
 
584
695
  - **Dry-run by default.** Every mutating command prints its plan and does nothing until you
@@ -620,13 +731,20 @@ window.
620
731
  | Platform | Status | Mutations |
621
732
  |---|---|---|
622
733
  | Windows 11 + Claude Desktop | verified 2026-07-31; sync end-to-end incl. live continuation 2026-08-03; sync-undo drift refusal live 2026-08-04 | read-only + mutations |
623
- | Windows 11 + Claude Desktop 2.42304.2 (user data moved to `%LOCALAPPDATA%\Claude-Data`) | from 0.16.3: read-only paths probed 2026-10-19 against the moved store. 0.16.2 and earlier find no store | read-only + mutations, `--apply` not yet verified on this build |
734
+ | Windows 11 + Claude Desktop 2.42304.2 (user data moved to `%LOCALAPPDATA%\Claude-Data`) | from 0.16.3: read-only paths probed 2026-10-19 against the moved store. 0.16.2 and earlier find no store where the app moved the folder, and read the old copy where it copied the folder and left the original in place | read-only + mutations, `--apply` never run on this build (the app updated to 3.x first) |
735
+ | Windows 11 + Claude Desktop 3.6556.2 (user data still in `%LOCALAPPDATA%\Claude-Data`) | on 0.16.5: read-only paths and `converge --apply` verified 2026-10-23, the apply with the app closed and the browser helper running. Needs 0.16.3 or later to find the store | read-only + mutations |
736
+ | Windows 11 + Claude Desktop 4.4212.0 (user data still in `%LOCALAPPDATA%\Claude-Data`) | on 0.16.5: `doctor`, `alignment` and `converge --apply` verified 2026-11-01, the apply with the app closed and the browser helper running. Needs 0.16.3 or later to find the store | read-only + mutations |
737
+ | Windows 11 + Claude Desktop 5.1231.0 (user data still in `%LOCALAPPDATA%\Claude-Data`) | on 0.16.5: `doctor` and `alignment` verified read-only 2026-11-10. Needs 0.16.3 or later to find the store | read-only + mutations, `--apply` not yet run on this build |
738
+ | Windows 11 + Claude Desktop 6.6395.0 (user data still in `%LOCALAPPDATA%\Claude-Data`) | from 0.17.0: `doctor` and `realign --apply` verified 2026-11-17, the apply with the app closed; after the app reopened, `doctor` and a fresh `realign` plan found every rewritten row opening the live copy and no set-aside copy back in its old folder. Needs 0.16.3 or later to find the store | read-only + mutations |
624
739
  | macOS / Linux desktop | unverified | **read-only only** - desktop-store mutations refuse, with no override |
625
740
  | CLI-only sessions (any OS) | transcript layout verified | mutations allowed via `--transcript-only` |
626
741
 
627
- The dates are measurements against specific builds - most recently Claude Desktop
628
- 2.42304.2.0 (Microsoft Store MSIX install) for the read-only paths, and 2.42304.0.0 with Claude
629
- Code CLI 2.1.302 for the last `--apply` runs. On 2026-10-14 the
742
+ The dates are measurements against specific builds of the MSIX that Anthropic signs and
743
+ distributes itself from claude.ai - most recently Claude Desktop 6.6395.0.0 with Claude Code
744
+ CLI 2.1.326 for `doctor` and the last `--apply` run, a `realign`; `converge --apply` last ran
745
+ on 4.4212.0.0 with CLI 2.1.309. The Microsoft Store build is a separate package, `AnthropicPBC.Claude_<id>`.
746
+ This tool does not look in that package's folder and has never been run against it.
747
+ On 2026-10-14 the
630
748
  read-only paths were verified against a real Desktop 2.x store, build 2.37337.1.0 (`list`,
631
749
  `doctor`, `alignment`, `recover` without `--apply`, and a `converge` dry run): the row format
632
750
  had only gained keys, and every row this tool had written under 1.x was still present, some
@@ -647,6 +765,30 @@ the move. 0.16.4 reads those paths under the moved store: `doctor` checks
647
765
  the rows where they now are, and `undo` and `recover` treat such an operation as they did
648
766
  before the move. A row in a store the tool cannot read at all is listed by `doctor` apart, not
649
767
  counted in its exit code, and `undo` refuses it.
768
+ On 2026-10-23, under 3.6556.2.0, the data folder had not moved again, and rows written by the
769
+ 3.x builds had only gained two keys. A `converge --apply` on 0.16.5 ran with the app closed while the browser
770
+ helper kept running, and the running-app guard excused the helper under the
771
+ `trust-signed-helper` opt-in. After the app restarted, every row the apply had written was
772
+ present and byte-identical to what it wrote, and `doctor` reported none vanished.
773
+ On 2026-11-01, under 4.4212.0.0, the data folder had not moved again. A `converge --apply` on
774
+ 0.16.5 ran with the app closed and the browser helper running, and the guard excused the
775
+ helper as it had under 3.x. After the app restarted, every row the apply had written was
776
+ present and `doctor` reported none vanished. Within a few minutes the app had re-saved six of
777
+ those rows, and each had only gained keys. By 2026-11-02 the 4.x app had created 36 rows for
778
+ itself. None lacked a key that every older row carries, and four had gained one new key,
779
+ `isStarred`.
780
+ On 2026-11-10, under 5.1231.0.0, the data folder had not moved again, and `doctor` and
781
+ `alignment` on 0.16.5 found the store where 4.x left it. That build's release notes say
782
+ Windows 10 21H2 or later and Windows Server 2022 now keep the app's data in
783
+ `%LOCALAPPDATA%\Claude-Data` too, as the Windows 11 machine these measurements come from has
784
+ since 2.42304.2, so on those systems the move happens at 5.1231.0. If the folder layout there
785
+ matches that machine's, the same versions apply: 0.16.3 or later to find the store, 0.16.4 or
786
+ later for operations from before the move. The notes also say the Microsoft Store build keeps
787
+ its data where it was, and they mention moves that copied the folder instead of moving it.
788
+ Where such a copy leaves the old store in place, run this tool from an ordinary terminal: from
789
+ a shell the desktop app started, `%APPDATA%\Claude` is the app's view of its old folder, and
790
+ 0.16.5 can read the old copy through it. Not measured here: a Windows 10 machine, a Microsoft
791
+ Store install, and an `--apply` under 5.1231.0.
650
792
  The on-disk format has changed once already during this tool's own development; treat a
651
793
  much newer build as unverified territory.
652
794
 
@@ -1007,7 +1149,9 @@ window into `journaled`/`completed`/`rolled_back`.
1007
1149
  README, the window's checkbox and the command's help, says Chrome or Edge, since either can
1008
1150
  launch the helper. Three constants left unused since the August anti-fabrication fix are gone.
1009
1151
  - **0.16.3 finds the session store where Desktop 2.42304.2 moved it.** That build moved the
1010
- app's whole user-data folder out of the Microsoft Store package, from
1152
+ app's whole user-data folder out of the Microsoft Store package (corrected 2026-11-10: out of
1153
+ the MSIX package Anthropic distributes itself, not the Microsoft Store build, whose data the
1154
+ 5.1231.0 release notes say does not move), from
1011
1155
  `%LOCALAPPDATA%\Packages\Claude_<pkg>\LocalCache\Roaming\Claude` to
1012
1156
  `%LOCALAPPDATA%\Claude-Data\claude_<pkg>`, and left a 0-byte file in the old folder naming
1013
1157
  the new one. 0.16.2 looked only in the old places, so `doctor` and `alignment` reported no
@@ -1058,6 +1202,35 @@ window into `journaled`/`completed`/`rolled_back`.
1058
1202
  left every route refusing until a restart, or, when the app copied its data instead of moving
1059
1203
  it, left the window writing into the old copy. A press whose store layout changed since its
1060
1204
  plan now stops and replans before its next write.
1205
+ - **0.17.0 shows which copy of a transcript each account's row opens, and repairs it.** A
1206
+ conversation's transcript can end up in two project folders, and each account's row opens
1207
+ the copy in the folder its `cwd` names, so one sidebar can show a conversation as it was
1208
+ weeks ago while another shows it today. `doctor` now lists each such conversation, which copy
1209
+ each row opens, and how the copies relate, reading past the bookkeeping records the app
1210
+ writes to whichever copy it has open; it exits 1 when an unarchived row opens an older or a
1211
+ diverged copy. It also lists, without counting them, conversations with one copy and a row
1212
+ whose folder holds none, because opening such a row made the app copy the transcript into
1213
+ that folder (measured on Desktop 5.11492.5). The new `realign` command points those rows at
1214
+ the folder holding the live copy, by rewriting `cwd` and the `folderPlace` keys the app keeps
1215
+ in step with it, and moves the stale copy into its journal, where `undo` can put it back. It
1216
+ holds anything it cannot decide safely, with the reason: diverged copies, a row with a
1217
+ worktree link, a live copy whose folder is gone, and identical copies unless
1218
+ `--include-identical` is given. On the machine it was built on, the first apply realigned 43
1219
+ conversations (84 rows) and held 52; `doctor` then listed 26 split conversations where it
1220
+ had listed 30, and 26 waiting ones where it had listed 65. `realign --release` sends a
1221
+ set-aside copy to the Recycle Bin once every row of its conversation opens one copy and that
1222
+ copy holds everything the set-aside one did, sidecar files included. It refuses each case it
1223
+ can see in which Windows would delete the copy outright, and asks Windows to warn in any
1224
+ other; how that warning behaves beside the silent flags release passes is untested, and no
1225
+ `--release --apply` had run when this shipped. The window's health report counts the split
1226
+ conversations, the waiting ones and the copies realign holds. `realign` itself is CLI-only,
1227
+ and while one is the latest operation the window offers no undo.
1228
+ The window also declares system DPI awareness before it opens, on Windows with Tk 9.0, so
1229
+ at 250 % it draws its text and widgets at the display's resolution and at the size it had
1230
+ before, where it used to be a stretched bitmap. Every pixel constant goes through one factor,
1231
+ the process's DPI over 96. Verified at 250 % and in a faithful 100 % run, and every other
1232
+ GUI harness passes in an aware process. 150 %, a common laptop setting, is untested; Tk 8.6
1233
+ is unchanged.
1061
1234
  - Platform rows above move from "unverified" to "verified" as contributors confirm the store
1062
1235
  paths and behavior on their own machines.
1063
1236
 
@@ -127,7 +127,7 @@ read. See `docs/internals.md`, RULING 7.
127
127
 
128
128
  ## Before any move: close the Claude app.
129
129
 
130
- `move`, `undo`, `recover`, `repoint --apply`, `new-row --apply`, and every mutating `sync`
130
+ `move`, `undo`, `recover`, `repoint --apply`, `new-row --apply`, `realign --apply`, and every mutating `sync`
131
131
  route (`--apply`, `undo` of a completed sync, `recover --back` on a stuck one) all refuse
132
132
  while they can see the Claude desktop app running - whoever the identity files currently say
133
133
  is signed in. That includes `undo` of a completed `repoint` or `new-row`, and `recover` on an
@@ -153,7 +153,7 @@ itself, and the guard counts it - if a refusal names that process, fully exit th
153
153
 
154
154
  ## Usage
155
155
 
156
- Twelve commands. All mutating commands default to a dry run; add `--apply` to execute.
156
+ Thirteen commands. All mutating commands default to a dry run; add `--apply` to execute.
157
157
  (`trust-signed-helper` is the opt-in described above; it changes a setting, not your
158
158
  sessions.)
159
159
 
@@ -174,8 +174,9 @@ claude-code-sessions list gate
174
174
  ```
175
175
 
176
176
  **`doctor`** - read-only health report (stale locks, unresolved operations, orphaned rows,
177
- encoding-scheme ambiguity, a synthesized row that has since vanished, and **conversations no
178
- account's sidebar points at**):
177
+ encoding-scheme ambiguity, a synthesized row that has since vanished, **conversations no
178
+ account's sidebar points at**, and **conversations whose accounts open different copies of one
179
+ transcript**):
179
180
 
180
181
  ```
181
182
  claude-code-sessions doctor
@@ -189,6 +190,30 @@ Resuming a session while signed into a different account repoints its row, which
189
190
  conversation you were in an hour ago reachable from no sidebar at all. That is what this
190
191
  ranking is for, and `repoint` is how you put it back.
191
192
 
193
+ The copies check covers a case every other number misses. A conversation's transcript can end
194
+ up in two project folders, usually after a session moves into a second worktree. Each account's
195
+ row names a folder through its `cwd`, and the app opens the copy in that folder, even a stale
196
+ one. So one sidebar can show a conversation as it was weeks ago while another shows it today,
197
+ and `alignment` still counts it complete. `doctor` lists each such conversation, which
198
+ copy each account's row opens, and how each pair of copies relates: one behind the other (one
199
+ history, one copy stopped growing), the same conversation, byte-identical, or diverged (each
200
+ holds turns the other lacks). It compares the bytes first and then reads what lies past the
201
+ bytes two copies share, because the app writes bookkeeping records (`tools-used` when it opens
202
+ a copy, `cost-state` when it quits) to whichever copy it has open, the stale one included. A
203
+ copy ahead only by those records holds the same conversation; a stale copy the app has opened
204
+ and quit is still behind, although its bytes no longer match. Any other record type counts as
205
+ conversation. `doctor` exits 1 when an unarchived row opens an older or a diverged copy.
206
+ Identical copies and copies that differ only in bookkeeping are listed and do not change the
207
+ exit code. Neither does a row whose folder holds no copy at all. A row whose `cwd` names a main repository while the transcript sits in
208
+ one of its worktree folders is the app's normal placement, not a fault.
209
+
210
+ Deleting the stale copy is a catch-up, not a fix. On one conversation, measured on Desktop
211
+ 5.11492.5, opening a row whose folder held no copy made the app copy the transcript back into
212
+ that folder, so the split returns once the conversation is worked from one account and opened
213
+ from another. For the same reason `doctor` also lists, without counting them, conversations with
214
+ a single copy and a row whose folder holds none: opening that row starts a split. `realign`
215
+ (below) is the repair for both.
216
+
192
217
  **`alignment`** - read-only scoreboard: how close your accounts are to holding one shared
193
218
  history. `doctor` answers "is anything broken"; this answers "how far apart are my sidebars,
194
219
  and which way is it moving":
@@ -207,8 +232,10 @@ Two of those actively trade against each other: making a conversation reachable
207
232
  both halves of a disagreeing pair in every sidebar under one name, which is a direct hit to
208
233
  *distinguishable*. A single "aligned: yes/no" would hide exactly the decision you have to make.
209
234
 
210
- It reads row files and transcript **filenames** only - no transcript content - so it stays fast
211
- on a store with gigabytes behind it. **It always exits 0** when it could produce the report. It
235
+ It reads row files and transcript **filenames and sizes** only - no transcript content - so it
236
+ stays fast on a store with gigabytes behind it. That is why its line on conversations whose
237
+ accounts open different copies only counts them, and how many differ in length; telling a stale
238
+ copy from a diverged one means reading both, which `doctor` does. **It always exits 0** when it could produce the report. It
212
239
  is a scoreboard, not a check: these numbers stay non-zero for as long as the work takes, and a
213
240
  command that exits 1 for months trains you to stop reading it.
214
241
 
@@ -549,6 +576,90 @@ stopped being redundant is presence it leaves alone. `doctor`'s vanished-row che
549
576
  converge-created rows the same way it watches `new-row`'s, with the same journal-retention
550
577
  window.
551
578
 
579
+ **`realign`** - point every account's row of a split conversation at the folder holding its
580
+ live transcript copy:
581
+
582
+ ```
583
+ claude-code-sessions realign
584
+ claude-code-sessions realign --apply
585
+ ```
586
+
587
+ These are the split rows `doctor` reports. A probe on Desktop 5.11492.5, run on one plain and
588
+ one worktree conversation, found that a row's `cwd` alone decides which copy it opens; that the
589
+ app rewrites the row's `folderPlace` to match `cwd`, and never writes `cwd` back; and that a
590
+ turn typed after `cwd` moved runs in the new folder, in the same transcript. So `realign`
591
+ rewrites `cwd`, and `folderPlace.cwd` and `folderPlace.realPath` where the row has a
592
+ `folderPlace` (it never adds one), and leaves every other key alone, including `originCwd` and
593
+ the title. The rewritten rows then open the live copy. No row opens the stale copy any more, so
594
+ it moves into the operation's journal, with its sidecar folder, where `undo` can put it back
595
+ and `doctor` lists it. Without that step, `move`, `repoint`, `new-row` and `converge` would
596
+ keep refusing the conversation for its duplicated transcript. A copy that no row opened even
597
+ before `realign` stays where it is: that is the duplicated-transcript case, which `realign` did
598
+ not cause.
599
+
600
+ By default it plans conversations with a copy behind the live one, and rows whose folder holds
601
+ no copy. Copies that hold the same conversation (byte-identical, or different only in
602
+ bookkeeping records) are realigned only with `--include-identical`, onto the copy the most
603
+ recently active row opens. `--only` narrows to one conversation, resolved the way `converge
604
+ --only` is. A conversation is **held**, with the reason printed, when:
605
+
606
+ - its copies have diverged (each holds turns the other lacks), so which history to keep is
607
+ your decision;
608
+ - a copy no row opens holds turns the live copy lacks;
609
+ - a row to change has a worktree link (`worktreePath` or `worktreeName`). In the probe, moving
610
+ such a row's path keys either detached it from its worktree or left it unable to start;
611
+ - the live copy is reached only through the worktree placement, sits in a legacy-encoded
612
+ folder, or has no path on record that encodes to its folder;
613
+ - the paths that reach the live copy's folder disagree, or its directory no longer exists;
614
+ - an unfinished operation involves it, or anything about it changed between plan and apply.
615
+
616
+ `--apply` needs the desktop app fully closed, re-plans under its lock and writes only the
617
+ conversations whose rows, copies and target are unchanged, prints any it held at that point,
618
+ and exits 3 when any conversation was held. An interrupted run recovers in both directions:
619
+ `recover --forward` finishes from the journal, and is offered only when every row and copy it
620
+ would touch is as the journal expects; `recover --back` puts each conversation's copies and
621
+ rows back, and keeps realign's placement for a conversation whose copy cannot go back (its
622
+ place is taken, for instance), saying why. `undo` works per conversation. It puts back only
623
+ the keys realign changed, keeps whatever the app wrote to other keys since, and refuses a
624
+ conversation, naming the reason, when its rows no longer hold what realign wrote, now open
625
+ another conversation, were written by a later operation, or when the conversation has gained
626
+ turns since (bookkeeping records do not count). The other conversations go ahead, and a later
627
+ `undo --apply` retries the refused ones.
628
+
629
+ A copy realign sets aside stays in its journal until `undo` puts it back or `realign --release`
630
+ sends it to the Recycle Bin; `doctor` lists every copy held. Once a conversation has gained
631
+ turns, undo refuses it for good, which is what release is for:
632
+
633
+ ```
634
+ claude-code-sessions realign --release
635
+ claude-code-sessions realign --release --apply
636
+ ```
637
+
638
+ A copy is released only when every row of its conversation now opens one copy, that copy holds
639
+ the set-aside copy's whole conversation (the bookkeeping records `tools-used` and `cost-state`
640
+ aside, which carry no conversation), and its sidecar folder holds every file the set-aside
641
+ copy's does, with the same content: a copy that holds anything else the live one lacks stays in
642
+ the journal, with the reason printed. It goes to the Windows Recycle Bin, and the step is built
643
+ so that a permanent delete cannot happen silently. A shell that cannot recycle a file destroys
644
+ it instead. So release refuses a bin turned off (by you or by policy), one whose capacity the
645
+ registry does not record, one without room for the whole run without pushing out your older
646
+ deleted files (asked again just before each move), and a path too long for the shell, and it
647
+ asks Windows to warn rather than delete in any case it cannot see. After each move it checks
648
+ that a new bin record names the file, with its size and its contents present. It needs the
649
+ desktop app closed, since its checks read the rows and transcripts the app rewrites. A copy
650
+ whose release is interrupted is finished by the next run, and only after the same checks; once
651
+ its transcript is in the bin, only while the live copy still starts with everything it held
652
+ when the release began. One gap is stated rather than closed: a case the checks cannot see
653
+ rests on Windows' warning, whose behaviour beside the silent flags release uses is untested. Once
654
+ every copy of an operation is released, the journal ages the operation out like any other;
655
+ `undo` refuses a conversation whose copy was released or is part-way through release, and
656
+ `--json` gives each copy's outcome, a failed run included.
657
+
658
+ A released copy can be restored from the Recycle Bin by hand until the bin is emptied or
659
+ Windows clears it to make room. It comes back to its journal folder, not to its project folder;
660
+ the operation's record names where each part went. Windows only, like every other store mutation
661
+ here.
662
+
552
663
  ## Safety design
553
664
 
554
665
  - **Dry-run by default.** Every mutating command prints its plan and does nothing until you
@@ -590,13 +701,20 @@ window.
590
701
  | Platform | Status | Mutations |
591
702
  |---|---|---|
592
703
  | Windows 11 + Claude Desktop | verified 2026-07-31; sync end-to-end incl. live continuation 2026-08-03; sync-undo drift refusal live 2026-08-04 | read-only + mutations |
593
- | Windows 11 + Claude Desktop 2.42304.2 (user data moved to `%LOCALAPPDATA%\Claude-Data`) | from 0.16.3: read-only paths probed 2026-10-19 against the moved store. 0.16.2 and earlier find no store | read-only + mutations, `--apply` not yet verified on this build |
704
+ | Windows 11 + Claude Desktop 2.42304.2 (user data moved to `%LOCALAPPDATA%\Claude-Data`) | from 0.16.3: read-only paths probed 2026-10-19 against the moved store. 0.16.2 and earlier find no store where the app moved the folder, and read the old copy where it copied the folder and left the original in place | read-only + mutations, `--apply` never run on this build (the app updated to 3.x first) |
705
+ | Windows 11 + Claude Desktop 3.6556.2 (user data still in `%LOCALAPPDATA%\Claude-Data`) | on 0.16.5: read-only paths and `converge --apply` verified 2026-10-23, the apply with the app closed and the browser helper running. Needs 0.16.3 or later to find the store | read-only + mutations |
706
+ | Windows 11 + Claude Desktop 4.4212.0 (user data still in `%LOCALAPPDATA%\Claude-Data`) | on 0.16.5: `doctor`, `alignment` and `converge --apply` verified 2026-11-01, the apply with the app closed and the browser helper running. Needs 0.16.3 or later to find the store | read-only + mutations |
707
+ | Windows 11 + Claude Desktop 5.1231.0 (user data still in `%LOCALAPPDATA%\Claude-Data`) | on 0.16.5: `doctor` and `alignment` verified read-only 2026-11-10. Needs 0.16.3 or later to find the store | read-only + mutations, `--apply` not yet run on this build |
708
+ | Windows 11 + Claude Desktop 6.6395.0 (user data still in `%LOCALAPPDATA%\Claude-Data`) | from 0.17.0: `doctor` and `realign --apply` verified 2026-11-17, the apply with the app closed; after the app reopened, `doctor` and a fresh `realign` plan found every rewritten row opening the live copy and no set-aside copy back in its old folder. Needs 0.16.3 or later to find the store | read-only + mutations |
594
709
  | macOS / Linux desktop | unverified | **read-only only** - desktop-store mutations refuse, with no override |
595
710
  | CLI-only sessions (any OS) | transcript layout verified | mutations allowed via `--transcript-only` |
596
711
 
597
- The dates are measurements against specific builds - most recently Claude Desktop
598
- 2.42304.2.0 (Microsoft Store MSIX install) for the read-only paths, and 2.42304.0.0 with Claude
599
- Code CLI 2.1.302 for the last `--apply` runs. On 2026-10-14 the
712
+ The dates are measurements against specific builds of the MSIX that Anthropic signs and
713
+ distributes itself from claude.ai - most recently Claude Desktop 6.6395.0.0 with Claude Code
714
+ CLI 2.1.326 for `doctor` and the last `--apply` run, a `realign`; `converge --apply` last ran
715
+ on 4.4212.0.0 with CLI 2.1.309. The Microsoft Store build is a separate package, `AnthropicPBC.Claude_<id>`.
716
+ This tool does not look in that package's folder and has never been run against it.
717
+ On 2026-10-14 the
600
718
  read-only paths were verified against a real Desktop 2.x store, build 2.37337.1.0 (`list`,
601
719
  `doctor`, `alignment`, `recover` without `--apply`, and a `converge` dry run): the row format
602
720
  had only gained keys, and every row this tool had written under 1.x was still present, some
@@ -617,6 +735,30 @@ the move. 0.16.4 reads those paths under the moved store: `doctor` checks
617
735
  the rows where they now are, and `undo` and `recover` treat such an operation as they did
618
736
  before the move. A row in a store the tool cannot read at all is listed by `doctor` apart, not
619
737
  counted in its exit code, and `undo` refuses it.
738
+ On 2026-10-23, under 3.6556.2.0, the data folder had not moved again, and rows written by the
739
+ 3.x builds had only gained two keys. A `converge --apply` on 0.16.5 ran with the app closed while the browser
740
+ helper kept running, and the running-app guard excused the helper under the
741
+ `trust-signed-helper` opt-in. After the app restarted, every row the apply had written was
742
+ present and byte-identical to what it wrote, and `doctor` reported none vanished.
743
+ On 2026-11-01, under 4.4212.0.0, the data folder had not moved again. A `converge --apply` on
744
+ 0.16.5 ran with the app closed and the browser helper running, and the guard excused the
745
+ helper as it had under 3.x. After the app restarted, every row the apply had written was
746
+ present and `doctor` reported none vanished. Within a few minutes the app had re-saved six of
747
+ those rows, and each had only gained keys. By 2026-11-02 the 4.x app had created 36 rows for
748
+ itself. None lacked a key that every older row carries, and four had gained one new key,
749
+ `isStarred`.
750
+ On 2026-11-10, under 5.1231.0.0, the data folder had not moved again, and `doctor` and
751
+ `alignment` on 0.16.5 found the store where 4.x left it. That build's release notes say
752
+ Windows 10 21H2 or later and Windows Server 2022 now keep the app's data in
753
+ `%LOCALAPPDATA%\Claude-Data` too, as the Windows 11 machine these measurements come from has
754
+ since 2.42304.2, so on those systems the move happens at 5.1231.0. If the folder layout there
755
+ matches that machine's, the same versions apply: 0.16.3 or later to find the store, 0.16.4 or
756
+ later for operations from before the move. The notes also say the Microsoft Store build keeps
757
+ its data where it was, and they mention moves that copied the folder instead of moving it.
758
+ Where such a copy leaves the old store in place, run this tool from an ordinary terminal: from
759
+ a shell the desktop app started, `%APPDATA%\Claude` is the app's view of its old folder, and
760
+ 0.16.5 can read the old copy through it. Not measured here: a Windows 10 machine, a Microsoft
761
+ Store install, and an `--apply` under 5.1231.0.
620
762
  The on-disk format has changed once already during this tool's own development; treat a
621
763
  much newer build as unverified territory.
622
764
 
@@ -977,7 +1119,9 @@ window into `journaled`/`completed`/`rolled_back`.
977
1119
  README, the window's checkbox and the command's help, says Chrome or Edge, since either can
978
1120
  launch the helper. Three constants left unused since the August anti-fabrication fix are gone.
979
1121
  - **0.16.3 finds the session store where Desktop 2.42304.2 moved it.** That build moved the
980
- app's whole user-data folder out of the Microsoft Store package, from
1122
+ app's whole user-data folder out of the Microsoft Store package (corrected 2026-11-10: out of
1123
+ the MSIX package Anthropic distributes itself, not the Microsoft Store build, whose data the
1124
+ 5.1231.0 release notes say does not move), from
981
1125
  `%LOCALAPPDATA%\Packages\Claude_<pkg>\LocalCache\Roaming\Claude` to
982
1126
  `%LOCALAPPDATA%\Claude-Data\claude_<pkg>`, and left a 0-byte file in the old folder naming
983
1127
  the new one. 0.16.2 looked only in the old places, so `doctor` and `alignment` reported no
@@ -1028,6 +1172,35 @@ window into `journaled`/`completed`/`rolled_back`.
1028
1172
  left every route refusing until a restart, or, when the app copied its data instead of moving
1029
1173
  it, left the window writing into the old copy. A press whose store layout changed since its
1030
1174
  plan now stops and replans before its next write.
1175
+ - **0.17.0 shows which copy of a transcript each account's row opens, and repairs it.** A
1176
+ conversation's transcript can end up in two project folders, and each account's row opens
1177
+ the copy in the folder its `cwd` names, so one sidebar can show a conversation as it was
1178
+ weeks ago while another shows it today. `doctor` now lists each such conversation, which copy
1179
+ each row opens, and how the copies relate, reading past the bookkeeping records the app
1180
+ writes to whichever copy it has open; it exits 1 when an unarchived row opens an older or a
1181
+ diverged copy. It also lists, without counting them, conversations with one copy and a row
1182
+ whose folder holds none, because opening such a row made the app copy the transcript into
1183
+ that folder (measured on Desktop 5.11492.5). The new `realign` command points those rows at
1184
+ the folder holding the live copy, by rewriting `cwd` and the `folderPlace` keys the app keeps
1185
+ in step with it, and moves the stale copy into its journal, where `undo` can put it back. It
1186
+ holds anything it cannot decide safely, with the reason: diverged copies, a row with a
1187
+ worktree link, a live copy whose folder is gone, and identical copies unless
1188
+ `--include-identical` is given. On the machine it was built on, the first apply realigned 43
1189
+ conversations (84 rows) and held 52; `doctor` then listed 26 split conversations where it
1190
+ had listed 30, and 26 waiting ones where it had listed 65. `realign --release` sends a
1191
+ set-aside copy to the Recycle Bin once every row of its conversation opens one copy and that
1192
+ copy holds everything the set-aside one did, sidecar files included. It refuses each case it
1193
+ can see in which Windows would delete the copy outright, and asks Windows to warn in any
1194
+ other; how that warning behaves beside the silent flags release passes is untested, and no
1195
+ `--release --apply` had run when this shipped. The window's health report counts the split
1196
+ conversations, the waiting ones and the copies realign holds. `realign` itself is CLI-only,
1197
+ and while one is the latest operation the window offers no undo.
1198
+ The window also declares system DPI awareness before it opens, on Windows with Tk 9.0, so
1199
+ at 250 % it draws its text and widgets at the display's resolution and at the size it had
1200
+ before, where it used to be a stretched bitmap. Every pixel constant goes through one factor,
1201
+ the process's DPI over 96. Verified at 250 % and in a faithful 100 % run, and every other
1202
+ GUI harness passes in an aware process. 150 %, a common laptop setting, is untested; Tk 8.6
1203
+ is unchanged.
1031
1204
  - Platform rows above move from "unverified" to "verified" as contributors confirm the store
1032
1205
  paths and behavior on their own machines.
1033
1206