[guijacquemet]

What

Rebuilt again on 12 November 2026, after #292 merged as 25c561f. #292 covers the owner's v0.5.2 Colab test: the axes confirmation is shown at once and is faster, and the image type moved to Channels. The PR branch was not force-pushed. In order, it holds:

• the first pair: 1244527 (A) and c4173dc (B);
• e5acf8c, which merged main at ac2dc27 (Conversion record version 4: refuse version 3, require source_relative_path where written #291);
• the pair A' 6a151b8 and B' 90f86cc;
• 6723955, which merges main at 25c561f (Axes: show the confirmation at once and make it faster; image type moves to Channels (v0.5.2 owner feedback) #292);
• the new pair below.

Since v0.5.1 (8726cb4), main has merged #278, #280, #282, #288, #289, #284-#287, #291 and #292.

1. cadbed4bb049995b2385e3b8499c8d247d3754bd Prepare v0.5.2 again (commit A'').
   • CHANGELOG ## v0.5.2.
     • The summary and Changed gain Axes: show the confirmation at once and make it faster; image type moves to Channels (v0.5.2 owner feedback) #292:
       • Working… and a 5-part progress bar on Confirm these axes, Confirm axis roles and Change axis roles;
       • an unexpected failure is shown in red instead of nothing;
       • no pictures are read for the folded per-folder image-type rows (93 → 9 plane reads on 28 folders of 3 × 2048²);
       • the image type is proposed at the quick scan (display only) and confirmed first on Channels.
     • Axes: show the confirmation at once and make it faster; image type moves to Channels (v0.5.2 owner feedback) #292's notes stay as their own subsection. Its rules and the published files are unchanged.
   • docs/colab-release-check.md:
     • Images shows the image type, display only ("nothing is recorded here");
     • Axes shows Working… and the bar, and no image type;
     • Channels starts with the image type: Use for all folders, Set per folder… with pictures once opened, R_HETEROGENEOUS_LAYOUT_NOT_ENABLED.
   • docs/user_guide.md: the image type moves from step 2 to step 3, with the scan's display-only line and the Working… state. Axes: show the confirmation at once and make it faster; image type moves to Channels (v0.5.2 owner feedback) #292 had left the guide describing it on Axes.
   • docs/conformance.md: the v0.5.2 record, re-measured on A''s tree (below).
   • RELEASE_REF goes back to v0.5.1's d023ff9, with its test and no v0.5.2 entry in RELEASE_CALLS, as in A and A'.
2. 4db97e9016d773bd97e8d29ab62911455082327e Points the notebook at A'' (commit B'').
   • RELEASE_REF = "cadbed4bb049995b2385e3b8499c8d247d3754bd", with the pair cadbed4… → PreprocessingNotebook() in RELEASE_CALLS and the one test that asserts the ref. RELEASE_TAG stays "v0.5.2".
   • git diff A'' B'' touches only the notebook's RELEASE_REF line, tests/test_notebook_file.py and tests/test_notebook_setup.py.
   • git diff A'' B'' -- src pyproject.toml is empty, and git diff main B'' -- src is empty.
   • claude/release-test-0.5.2 is now at B'' (4db97e9).

Breaking (summary; details in the CHANGELOG)

• Run manifest version 5.
  • Version 4 (every v0.5.1 run) is refused by read_manifest and RunInputs.open with ManifestFormatError.
  • The fix: run preprocessing again into the same output folder.
• Conversion record version 4 (Conversion record version 4: refuse version 3, require source_relative_path where written #291).
  • Version 3 (every v0.5.1 folder of copies) is refused with conversion_record_unreadable, naming the fix: convert the originals again into a new folder.
  • An OME-Zarr copy's entry must name its source in source_relative_path, and no other entry may.
  • A projected entry's two source paths must agree.
• plane_histograms.csv schema version 5 (normalization v0.5.0). Columns are named by channel ID, an absent channel is blank (never zero), and a version 4 table is refused.
• Resolver contract version 2:
  • Capabilities.parent_labels_support;
  • parent_labels_as_area;
  • as objects, A_MASK_NOT_INSTANCES.
• Mask scan: pieces are counted 8-connected, and mask_relationships.json is report version 4.
• Channels by ID in the API.
• Spec pin 69a58b3.

#292 changes no published file: for the same answers, morphobricks.toml, experiment_design.csv and plane_histograms.csv are byte-identical to the files before it. This was checked on one RGB experiment and one fluorescence experiment, both with masks.

Gate

	Result
python tools/check.py on A''s tree (before the record)	Ruff clean, 459 files formatted, pytest 5487 passed, 4 skipped, 0 failed, code rules passed, exit 0
python tools/check.py on B'' 4db97e9	Ruff clean, 459 files formatted, pytest 5487 passed, 4 skipped (the same four), 0 failed, code rules passed, exit 0
Spec modules pytest -rsxX --junitxml tests/test_spec_*.py	1799 passed, 0 skipped, 0 xfailed, 0 failed, all 25 EXPECTED_MODULES. The JUnit file has no skipped, failure or error case
Install from GitHub by B''s SHA, fresh venv	pip install --no-cache-dir "morphobricks-preprocessing[notebook] @ git+…@4db97e9016d773bd97e8d29ab62911455082327e" installs 0.5.2 (direct_url.json commit_id 4db97e9…), and PreprocessingNotebook() with .steps() builds. Not run: the shipped notebook's cell 1 in a real kernel
CI on B'' (run 43211911872)	lint, and test on ubuntu, macos and windows with Python 3.11 and 3.13: all 7 jobs succeeded

Not run (environment skips, outside the Spec modules):

• test_cli_log.py:166, test_folder_rules.py:199, test_folder_rules.py:230: the file system is case-sensitive;
• test_discovery.py:665: the tests run as root.

Spec gates: unchanged. This is a G1-style record. G3 is not declared, because these are not shown: the v05 per-repeat coverage, HV3's per-channel failure, and the bricks' readers. The decision on a gate stays the owner's.

For the owner

• Test the notebook with docs/colab-release-check.md. The Colab link is pinned to B'', which installs RELEASE_REF A'':
  • Colab: https://colab.research.google.com/github/CellMigrationLab/morphobricks-preprocessing/blob/4db97e9016d773bd97e8d29ab62911455082327e/notebooks/MorphoBricks_Preprocessing.ipynb
  • Raw download for JupyterLab: https://raw.githubusercontent.com/CellMigrationLab/morphobricks-preprocessing/4db97e9016d773bd97e8d29ab62911455082327e/notebooks/MorphoBricks_Preprocessing.ipynb
  • The same commit is on claude/release-test-0.5.2.
• Merge with a merge commit, not squash or rebase, so that A'' stays reachable from main.
• This PR creates no tag. After the merge: https://github.com/CellMigrationLab/morphobricks-preprocessing/releases/new?tag=v0.5.2&target=<merge sha>&title=v0.5.2

Open questions

1. Stains left after switching to colour. If the stains are named and then every folder is switched to colour (RGB) on Channels, the stain names stay declared. The card says "stay saved, unused", and they are written as [channels.*] in morphobricks.toml (W_UNUSED_CHANNEL). Switching back to separate channels clears every tag, and no stale R/G/B remain. This behaviour is unchanged since Notebook: an all-RGB experiment should not need a channel label on "Your channels" #240. Recommendation: keep it for v0.5.2. If you would rather drop unused stains when every folder becomes colour, that is a small follow-up, but it changes the published file on that path.
2. G3 and heterogeneous processing. Recommendation: release as is. Every brick refuses manifest v5 until its own S3.3 pin moves. Add a v05 coverage test before G3 is declared.
3. Mixed RGB and channel units (Stage 3 S3.3: per-unit channel lists, plane_histograms v5 #288). Recommendation: keep them refused for v0.5.2.
4. The Spec's docs/pins.md still lists preprocessing at 1bed024. Recommendation: update it in the Spec after the tag.

🤖 Generated with Claude Code

https://claude.ai/code/session_01G3qp9AxjpGPw9xfWAUJNqc

[guijacquemet]

Measured on B c4173dcd4e713135f06b8a838f796624ea635af9

All runs: Linux, Python 3.13.18, a fresh venv with .[dev,notebook,zarr] (metadata 0.5.2, normalization 0.5.0).

	Result
python tools/check.py	Ruff clean, 457 files formatted, pytest 5460 passed, 4 skipped, 0 failed, code rules passed, exit 0. Same counts as on A's tree
Spec modules pytest -rsxX tests/test_spec_*.py	1799 passed, 0 skipped
Install from GitHub at B	In a fresh venv, pip install "morphobricks-preprocessing[notebook] @ git+…@c4173dcd4e713135f06b8a838f796624ea635af9" gives 0.5.2, with normalization 0.5.0 and ipywidgets 8.1.9. The installed package equals src (diff -r). PreprocessingNotebook() builds welcome(), steps() and results_step() (three VBoxes). RESOLVER_VERSION is 2 and MANIFEST_VERSION is "5"
Shipped notebook, unmodified, real kernel (nbclient), fresh venv with only ipykernel/nbclient	1. Cell 3 run first prints one line, CellStop "… run cell 1 (Install MorphoBricks preprocessing) first …", with traceback == [evalue]. 2. Cell 1 pip-installs RELEASE_REF A from GitHub (direct_url.json commit 1244527…) and prints Ready: preprocessing 0.5.2; normalization 0.5.0; ipywidgets 8.1.9. and "pinned v0.5.2 candidate (1244527)". The installed files equal src. 3. Cell 2 prints its one-line note. 4. Cell 3 displays the app (a widget view)

Skips (not run):

• test_cli_log.py:166, test_folder_rules.py:199 and test_folder_rules.py:230: the file system is case-sensitive.
• test_discovery.py:665: the tests run as root.

---

Generated by Claude Code