[trissim]

Validated paired input-preparation checkpoint

Parent integration owner. Existing PR206 normally includes current main74eac059
and typed namespace worker328; paired PolyStore16 production91fd7e854 is adopted.
Metadata ownership stays on existing frozen declarations and backend state, with
MI metadata detection and declaration-derived registry behavior. No singleton
mutation, duplicate reader/store, fallback, assertion weakening or guard waiver.

Current29 source checks pass in both import orders; five namespace controls also
pass with independent custom filenames. Original packaged structural guards pass
with zero positive deltas for application, benchmark and scripts. Earlier125
quality controls and ROI work-count/profile are historical evidence at their
recorded source, not additional current tests or full NRA/ZIP acceptance.

Actual isolated paired wheels import from their installed environment. Persistent
installed MCP health/reopen/pixel samples succeed with generic dependency filename
independent of OpenHCS metadata. Raw ImageXpress generation at zero geometry then
real initialization/compile and persisted virtual sampling succeeds. Empty-step
compile tests initialization only, not a processing or biological pass.

Raw singleton Opera generation succeeds but initialization fails grid inference;
Averroes owns its focused follow-up under issue172. Original persisted Opera also
has A01/R01C01 well disagreement. Keep172 open; real Java/CZI/native execution and
full NRA/R1/ZIP acceptance remain follow-ups. No hosted CI waiting gate or public
registry/global-live readiness claim. Original failures and aggregate CLI exit2
remain preserved; final same-process health passed and all owned PIDs exited.

Canonical receipt: docs/validation/input_preparation_installed_20261001.rst.
Refs #327 and issue172, not automatic
closure of unfinished acceptance. Parent merges the useful validated checkpoint
and continues the remaining scope.

Fixes #327.

[trissim]

Architectural review: finish the owner boundaries for discovery filtering and reader lifecycle

Reviewed 006361ff28528c0159bed0999149c71b8befda94, pinned PolyStore f94bbbe8631a4c78f7ceaec72fb99671b58c9a18, and paired PolyStore #16 (8c886e7).

Two production ownership seams still need attention:

1. Physical binding selection is authored in two consumers. New SourceBindingsConfig.discovery_path_matches (source_bindings.py:1854–1872) reconstructs explicit_source.resolved(root).uri equality and loops over binding.selector.filters. SourceBindingWorkspaceProjector.candidate_matches_binding (source_binding_workspace.py:973–989) independently implements those same decisions after decode. Keeping predecode possible membership distinct from final metadata/component membership is correct; duplicating their shared physical-path decision is not. A change to explicit-source identity or selector path policy requires synchronized edits in both.

   Put physical-path admission on the existing NamedSourceBinding/SourceSelector contract, reusing the original SourceFilterMatcher and SourceFilterTargetResolver families. Both discovery and final projection should invoke that owner; final projection additionally checks decoded metadata/components. Preserve companion-path provenance and unrestricted-binding union behavior rather than forcing predecode to guess metadata. A new path-selector behavior should be one matcher/owner extension, not another pair of consumer edits (IMPL-12 / BOUND-2).

2. The new Java probe reconstructs a dependency-owned lifecycle. BioFormatsPlaneStoreAdapter._is_single_file, lines 827–841 calls ensure_initialized, constructs context.ImageReader(), invokes Java and closes the reader itself. The pinned polystore.bioformats_java.BioFormatsJavaContext already owns reader construction, initialization, failure cleanup and the related declares_path probe. Add the single-file capability there and consume its public operation from OpenHCS. The source-selection policy remains OpenHCS-owned; Java reader lifetime does not. Moving this body to another OpenHCS helper would preserve the duplicated mechanism (IMPL-13 / BOUND-2).

The paired ROI change is a good example of the intended direction: TwoDimensionalLabeledMaskROIExtractor uses scikit-image's already-owned RegionProperties.image/bbox instead of rescanning/reconstructing them. FijiROIConverter.rois_to_imagej_members still legitimately projects each parent's shapes to individual external ZIP members; those are not additional biological identities. Keep the recorded gitlink/paired-validation gate explicit so the OpenHCS PR does not claim the upstream optimization before adopting that validated revision.

These requests concern production owners, not fixture/test style. The existing 16-container, companion-source and raster/contour regressions are useful behavioral controls to retain through the factoring.

[trissim]

Both production findings are addressed in the existing paired drafts, not just fixture/lint changes:

1. IMPL-12 / BOUND-2, OpenHCS commit32a2b27e2: NamedSourceBinding.physical_path_matches owns explicit-source resolution/equality; SourceSelector.path_filters_match delegates clauses to the original matcher/target-resolver families. Both discovery and final projection call this binding operation. Removed duplicated physical procedures/consumer filter loops. Metadata/components still require decoding; config retains global filters and alias union. Companion filters retain store provenance but do not replace exact entrypoint identity.
2. IMPL-13 / BOUND-2, PolyStore commit0aa057b in draft16: BioFormatsJavaContext.is_single_file and existing declares_path share context-owned _probe_reader initialization/construction/unconditional close. Deleted OpenHCS's helper; its adapter invokes the public capability. No metadata/setId open in the probe, new context/helper, companion roster or cache. Existing in-place ROI optimization is unchanged.

Executed 28 lightweight passes in0.50s/72.8 MiB peak RSS:17 parent physical-admission cases/guard;11 child Java-context cases/guard. Exact relative/absolute/file-URI identity, directory-regex/OR groups, metadata/component deferral, decoded companions, unrestricted alias union, true/false/failed probe cleanup/retry, existing context cleanup. Controlled external responses, actual owner operations. AST guards prohibit the reviewed bypasses; seven touched Python files parse; new tests/touched child source pass Ruff, both diffs pass whitespace checks.

This is not a resolution of the retained141 passed/26 failed combined-suite result. Separately bootstrapped application preservation, broader sixteen-container/AUTO/TIFF rerun, bounded after-profile (not run), real Java/CZI and installed acceptance remain open. Recorded gitlink remainsf94bbbe; parent now requires the published paired capability before coordinator adoption. Explicit source imports verified; no MCP/JVM/GUI, shared lock takeover, installation/skill/merge or blind runtime/data changes. Guard11.4 GiB available, final11.2 GiB; tiny owned92 KiB scratch removed.

Full updated evidence/boundaries are in both PR bodies and durable receipts: parent6ffea8d61 / childa14a4fc. Lovelace retains ownership of the remaining source/preservation/profile checks; runtime depends on the next finite coordinator handoff.

[trissim]

Authorized critical cache/progress integration checkpoint is published as OpenHCS228 at bf9aae4620585e98d10e469037ef98237884b4f8, paired with PolyStore17 at c16b8fc99a07f17e92e5fbf46f312d73cc9c46d7. Both remote heads verified; exact child gitlink adopted by228. Base current main213439a04 includes merged218. References224,206 and16; original branches/source/receipts preserved, no force/rebase/reset.

This extracts the already-owned cache/progress implementation, not another runtime/materializer. Source-selection/Java probe/ROI optimization stays in206/16. Their 141-pass/26-failure receipt remains unresolved; after-ROI profile did not run. No claim of broad preservation or installed acceptance from this checkpoint.

Before application/MCP startup set POLYSTORE_IMAGEJ_CACHE_ROOT=/home/ts/.cache/polystore/imagej (bundle parent, not hashed directory) and POLYSTORE_IMAGEJ_ALLOW_DOWNLOAD=false. Both settings belong to existing PolyStore distribution/download-policy owners, projected by existing OpenHCS launch owner. Independent XDG logs/caches remain independent. Existing progress/worker binding stays canonical.

Current-main pair: 36 passed5.14s, peakRSS277712KiB, availableRAM14.4→13.96GiB. Two fresh native processes with production download-forbidden policy, separate XDG roots, actual same existing Fiji/Java paths, zero new cache entries/no downloads/no JVM. Reproducer and durable receipt are in228. No override/copy/archive in native proof, no installed/frozen/scientific change or replay. Parent owns review/merge/install and resource-coordinated real installed stdio inspection + concurrent discovery/heartbeat. No Closes/readiness claim until that live gate.

[trissim]

Parent is taking a narrow newly reproduced acquisition-calibration publication defect through issue302/PR304: a real synthetic HTD raw route uses0.65um but ordinary Gaussian compiled output loses spacing and defaults to1.0. Independent read-only trace identifies save_virtual_workspace_metadata publishing calibration separately from per-source payload metadata. PR304 will preserve native source spacing and publish acquisition-owned calibration through existing typed owners; no edits to this worktree or physical-selection/BioFormats/ROI changes here. Integrate304 normally when continuing206. Original failures and source claims retained; this is not a duplicate input-selection implementation.

[Copilot]

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