[guijacquemet]

Redesign iteration 3, part 3 of 4, stacked on #285. Items (aa) and (bb) of "OWNER v0.5.1 NOTEBOOK TEST" (RESTART.md). Refs #176.

Summary

• Z and T in the viewer: the t and z sliders of Notebook: one image viewer for every step, formatted report, X/Y pixel inputs (redesign 3, part 1) #284 are on every step; a time-lapse goes through the steps unchanged (one plane per time point; tested end to end).
• Z-stacks: save the dataset as projections (notebook_projection). A Z-stack is still refused (z_stack_not_supported, the red message stays). The Axes step then shows one card:
  • the Projection (maximum, mean, sum; the list starts on "choose a method": no default) and the slices used (all, unless changed);
  • Show this projection of the image above: the example's image projected from its reduced planes with projection.PlaneAccumulator, for looking only;
  • Write the projected copies and continue with them, available once a method is chosen and the example is ticked (the roles the copies are read with are the user's), with its blocker line otherwise. It calls convert_experiment(projection=Projection(...)): the copies, conversion record version 3 and every refusal of convert --project, nothing new;
  • the copies go to <output>/projected_<method>[_z<start>-<stop>] (an output folder may hold its experiment, contract §15), and the app opens them as the experiment: scanned and checked again, a line says what they are copies of, and the user confirms their axes (one click: the copies name them). Nothing is confirmed for the user.
  • Opening the stacks again with the same output folder offers Continue with the copies written earlier (nothing is rewritten).
• 3-D kept compatible (bb). The original stacks are only read (tested: their SHA-256 is the same before the projection and after the run). They are recorded, with no new field: the record's source_root and, per image, the same relative path, the stack's fingerprint, source_content_sha256, source_z_axis, source_z_size and z_range; the run records the record's SHA-256 and RunInputs.projection hands source_root and the per-image entries to the bricks. A volume request can therefore find the stack of any analysed image later without a migration. Keep the 3-D stacks for 3-D-capable tools (fiberfeatures) is shown greyed out with the line "3-D analysis is not enabled yet". Capabilities.volume_support and A_VOLUME_NOT_SUPPORTED are untouched.

Notes for review

• The projected tree is a new source. In the notebook it shares the user's output folder (nothing is published there for the stacks: they cannot pass the axes). The CLI text still recommends a new output folder; both are consistent with the record.
• OME-Zarr Z-stacks are refused by the projection as before (projection_zarr_not_supported), shown under the button.
• The stacked axis_step other bricks embed does not get the card.

Tests

tests/test_notebook_projection.py (new): Z, T and Z+T fixtures through save and run; the copies equal numpy max/mean bit for bit; the record's source facts; the greyed option and its reason; refusals; reopening. Full gate counts in the last PR of the stack.

---

Generated by Claude Code

[guijacquemet]

Independent review: #286 (Z/T sliders, explicit Z projection step)

Verdict: approve after one fix (pushed). Merged the axes branch forward as 3d43b7a, with no conflicts. Retargeted to main after #285 merged (6940bb6).

Defect found and fixed: aa55464. The projection method (which has no default) and the slice range chosen for one folder stayed chosen after another folder was scanned and checked. My script reproduced it: method max and slices 1 to 3 chosen on folder A were still set on folder B. They now reset on every new folder check, keyed on the check's discovery. A pixel size or masks stated on the same check keep the choice, because recalibrated/remasked keep the same discovery. Regression test: test_a_projection_chosen_for_one_folder_is_not_chosen_for_another. I confirmed it fails without the fix.

Checked with my own scripts:

• The originals' SHA-256 is unchanged after a mean projection of slices 1 to 3.
• Record v3: source_root, and per entry source_relative_path, source_content_sha256 (equal to my own SHA-256 of each original), source_z_axis, source_z_size = 5, z_range = [1, 4].
• After the copies are opened, confirmed_mapping is None: their axes are confirmed again by the user. The example tick does not survive a folder change.
• review_axes on a ZCYX stack still gives z_stack_not_supported, and Confirm axis roles stays disabled with its reason.
• Capabilities.volume_support and A_VOLUME_NOT_SUPPORTED are untouched (no diff to run_inputs*). Keep the 3-D stacks is disabled and shows "3-D analysis is not enabled yet".

Gate on aa55464 (Ruff check, Ruff format, code rules, pytest -n 2): 5447 passed, 4 skipped. The skips are the same four as #284: root-user file modes (one), case-sensitive file system (three). CI is 14/14 green.

---

Generated by Claude Code