[guijacquemet]

First of four stacked pull requests that redesign the notebook app after the owner's test of v0.5.0 on real data (10 November 2026): #269 (this), #270, #271, #272. Design note: docs/restart/designs/notebook-redesign-2026-11-10.md on MorphoFeature branch claude/restart-doc. This one holds the three small fixes, items (g), (i) and (j); nothing about when an answer is accepted or refused changes. Main (#267, #268) is merged in.

Summary

• notebook_style (g): path_list returns plain lines in a box (scrolling above about ten lines) instead of a nested list, and status joins a list of paths to the line before it. A list of files in an error card drew an empty bullet before it, and bullets inside "Show N more". Same paths (still bold), same fold.
• notebook_description (i): choosing what a condition "is compared against" rebuilt every dropdown of the card, twice, so the control vanished while the answer was saved. The rows are built once per set of conditions and updated in place (_build_references, _new_reference_rows, reference_notes). The answer, its refusal (a loop) and the saved draft are unchanged.
• notebook_render.blocked_line (j) and its uses: every disabled action has a visible line beside it saying what blocks it. Confirm axis roles names the axis error (different channel counts), the preview not drawn, or the preview not ticked; also Yes, the preview looks right, Compute histograms, Yes, save the experiment description, Make named-axis copies and its roles question, and the stains confirmation.
• Roles that are not a mapping at all (two X) now disable Confirm axis roles at once; before, the button kept its earlier state and the click did nothing.

Decisions

• The owner's data (3 and 4 channels) stays refused (different_channel_counts); the refusal is now named beside the button instead of only in a tooltip.
• The save button's line is its own widget (save_blocker): save_note keeps meaning "the run will stop" only.

Notes for review

• Measured on 28 images of 2048 x 2048 (3 channels): one "compared against" choice took 0.58 s and takes 0.35 s. The dropdown no longer leaves the page; the remaining time is the refresh of the other pages (mostly Check and save), not touched here.
• The owner reported "dozens of empty bullets" under "Show 70 more". In headless Chromium the old list drew the empty first bullet but did draw the 70 paths, so that part was not reproduced; the new list has no bullets at all.
• Widgets were tested headless and looked at in headless Chromium, not in Google Colab.

Tests

python tools/check.py on b5461fd (main merged): Ruff and the code rules pass; 5182 passed, 4 skipped (the four environment skips: test_cli_log.py:166, test_folder_rules.py:199 and :230 for a file system that tells names apart by case; test_discovery.py:665 for running with permissions that ignore file modes). New: tests/test_notebook_owner_fixes.py (19 tests: the list for 1 to 90 paths, the dropdown object surviving a choice and a refusal, each blocker line).

🤖 Generated with Claude Code

https://claude.ai/code/session_01G3qp9AxjpGPw9xfWAUJNqc

[guijacquemet]

Correction to the description: the stack is #269 (this), #270, #272 and #273 (not #271, #272).

🤖 Generated with Claude Code

https://claude.ai/code/session_01G3qp9AxjpGPw9xfWAUJNqc

---

Generated by Claude Code