G2: kernel modules spell no microscopy axis member

[trissim]

Surface G2 (receipt): kernel modules no longer spell a microscopy axis member.

Census. At 66a0b30a0, kernel modules had 134 spellings of a microscopy axis name, collection key or multi-letter label in 22 files, as string literals, attributes or class fields. After this PR there is one: the Fiji slot's ImageJ "channel", which saved configs persist as FijiDimensionMode.CHANNEL. The guard pins it exactly.

What changed

• The plate-file filter well is now partition (kernel query, MCP DTOs, services, dev client --partition). Synthetic-plate and compile-request wells are now partition_values (--partition-value). The runtime tree's partition node is PartitionProgressTreeNode (node_kind="partition").
• NGFF axis types now come from the axis name in PolyStore 0.5.0. This is a lockstep change: Derive the NGFF axis type from the axis name (0.5.0) PolyStore#36, submodule pointer bumped, pin polystore>=0.5.0,<0.6. The Zarr projection leaves are named by role.
• Fiji hyperstack coordinates are now ImageJ's c/z/t, iterated through one axes() helper. Progress declarations use progress_channel. Other renames: ColorToGray fixed_channels and numbered_channels, InvertForPrinting rgb_channels, ImagePlaneSource.colour_sample, SaveImages TIME/SLICE, PlatemapData.well_name. CellProfiler's Plate/Well tags are declared once in interop.
• The MetaXpress experimental-analysis format modules move from openhcs/formats/ into processing/backends/experimental_analysis/. The domain allowlist loses one entry.

Guard. tests/unit/test_axis_family_guards.py::test_kernel_modules_spell_no_microscopy_member scans every module outside the domain allowlist and interop/ (the CellProfiler domain). The set of found spellings must equal the single pinned Fiji declaration. test_domain_allowlist_names_existing_domain_modules checks that every allowlist entry exists.

Witness. A new test_zarr_layout_follows_roles_and_ngff_types checks that the remote-sensing family's Zarr layout follows its roles. The end-to-end subprocess witness (no domain module loaded) still passes.

Minimal edits outside the kernel (coordination with U1). In openhcs/agent, only dto/plate.py, services/plate_inspection_service.py, services/plate_streaming_service.py and services/synthetic_plate_service.py changed: the field renames that PlateFileInventoryQuery.partition and SyntheticPlateGenerationParameters.partition_values force. Nothing in openhcs/pyqt_gui changed.

Durable state. Saved pipelines that set ImagePlaneSource(channel=…) or a non-default SaveImagesSeriesAxis no longer load. Both parameters were never read. Decision G2-Q1 in the index defaults to no cutover tool.

Handoffs (recorded in the receipt):

• G7: well_filter/WellFilterConfig/owned_wells/total_wells identifiers.
• U1: the plate view widget's coord_to_well, and component_filters.
• P/P1: moving the consolidation, experimental-analysis and CellProfiler backend packages into domains/.

Verified locally: guards 10/10 (axis-family + dataset-source), witness 11/11, official30 parity 29/30 (baseline; cp_tutorial_3d_monolayer .pkl), test_main disk+zarr direct 6/6. Touched tests: 724 passed; the 18 failures also fail on origin/main in the same environment, apart from one flaky progress-tree test that passed on 3 reruns.

G2 · PR #1185 · guards 5/5 · production −294 +264 · tests −68 +198

🤖 Generated with Claude Code

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.