Skip to content

Native reopen interprets absent acquisition spacing through numeric micrometers defaults #907

Closed

Description

@trissim

Native reopen fills absent acquisition spacing from numeric source defaults

Named source/receiving owner: Singer (existing #541 follow-up). This is not a
new implementation assignment or a blanket hold on graph reopen/science.
Related #404 is superseded; #134 is closed. Preserve all historical refs/failures.

Determining merged #541 receipt:
docs/validation/neurite-public-pixel-acceptance-20261005.rst
at mergeb37403854776acd110228b1c17ddb4a0212a74d9.
Ordinary saved graph ZIP inventory/reopen succeeds as three feature-bearing
Shapes paths/69coordinates with channel2/site1/Z1/time1/A01, personally opened
matched raw/graph/combined and exact typed closure. Saved graph features/CSVs
declare pixels; acquisition SourceVoxelSpacing values are empty. Native scale1
summary nevertheless reports micrometers. This is not physical calibration.

Current-mainb1e720f5ed8201a59bfe53488581f1f6704ededf source:
ViewerStreamingSource.calibrated_metadata fills an existing metadata object's
absent spacing using plate_image_metadata. That calls the acquisition metadata
handler's source_voxel_spacing; the retained receipt traces its get_pixel_size
numeric default projected as MICROMETERS. No newer live physical proof is claimed.
Merged893 separately verifies its declared calibrated synthetic source, not a
graph-specific proof or fulfillment of the historical calibrated Site001 case.

Acceptance: through existing source-spacing/metadata owners, distinguish absent
acquisition calibration from a truly declared physical source; do not fabricate
micrometers from a numeric fallback. Retain strict source admission and pixel
graph metrics/features. Exercise the original public saved graph/native path
with absent versus explicitly calibrated source contracts; preserve geometry,
source identities and exact typed closure. Historical Site001 remains unverified
at that exact scope, not silently substituted by the synthetic pixel receipt.
No new runtime/client, patch, scientific adoption or replay was made for this
disposition. Coordinate shared source hunks with Singer before any implementation.

Activity

  1. trissim commented on Oct 6, 2026

    @trissim
    CollaboratorAuthor

    Working source repair published on fix/acquisition-spacing-907-20261006 at 1cd64e4 (production 3ae77c3). OpenHCS and BioFormats now project existing typed source spacing rather than numeric compatibility defaults; the legacy numeric-to-physical fallback is deleted. Unknown/relative/mixed sources cannot provide a physical scalar; explicit physical/anisotropic coordinates remain intact. Original NRA parser/consumer evidence covers 1056 production/dependency modules. Affected existing tests: 27 PASS, source03 exit0; initial two bootstrap failures retained. No dependency gitlinks, compilation/catalog code, installed package or scientific runtime changed. Checkpoint: docs/validation/acquisition-spacing-907/CHECKPOINT.rst. Planck whole-builder/Dewey recorded-lane receiving requested on this exact candidate: reuse saved engineering541 raw/GraphROI artifacts (no detector rerun), public inventory/stream/native summaries and matched captures for empty acquisition spacing, plus explicitly calibrated control, then exact closure. I retain implementation and receiving ownership; installed acceptance is not yet claimed. No new routine PR or CI hold.

  2. trissim commented on Oct 6, 2026

    @trissim
    CollaboratorAuthor

    Reviewed and normally merged the complete source repair into main ed1bbed. The source-spacing owner and both metadata consumers carry the relation; the legacy numeric-to-physical fallback is deleted, no second unit store or viewer codec was added. All27 affected checks are retained; current main changes since their qualification are paper/audit updates, not another production variation. This is merged and source-qualified, NOT installed/live saved-reopen acceptance. Singer is assigned the original saved raw/GraphROI and explicit-physical-control receiving journey without detector rerun or new worktree/environment; no extra PR or CI wait. Keeping this issue open until that actual path confirms pixel units for unknown acquisition spacing and unchanged physical units for declared calibration.

  3. trissim commented on Oct 6, 2026

    @trissim
    CollaboratorAuthor

    Installed/public saved-reopen acceptance completed on whole merged-main candidate99cb4880d. Published receipt-only checkpoint2c53f50c7 on fix/acquisition-spacing-907-20261006: docs/validation/acquisition-spacing-907/LIVE-ACCEPTANCE01.rst and original live01.tar.gz. One original recorded client,22 successful public replies, terminal0. Saved engineering541 unknown-spacing raw16384/16384 pixels and three GraphROI paths/all69 coordinates exactly preserved; graph features coordinate_unit=pixels, native XY identity scale, acquisition spacing remains EMPTY (not invented1um). Empty typed spacing still serializes its default micrometers enum, explicitly qualified as absent values, not physical calibration. Paired retained engineering440441 raw+ROI reopening retains explicit[.5,.5]micrometers and native XY[.5,.5]; all12 raw values match. Original raw/result/combined and physical-combined PNGs personally opened. Both viewer closes ACK+process_exited+succeeded; exact PIDs/listeners/MCP/client absent afterward. Five original input/metadata hashes unchanged. No detector rerun, new WT/env, global install, SCI hotpatch or catalog launch. One posthoc verifier ZIP-metadata decoding mistake retained separately; corrected verifier terminal0, no MCP replay. Only own finished source/build+tmp/cache removed; wheel/target/source/native logs and all original evidence retained. No remaining907 saved-reopen acceptance gap; no biological or leak claim. Parent can integrate this receipt-only checkpoint and close907 at this actual scope.

  4. trissim commented on Oct 6, 2026

    @trissim
    CollaboratorAuthor

    Fixed on main by 3ae77c3 (integrated through ed1bbed). The original saved-reopen case now has actual installed-candidate MCP/native acceptance, not only source tests: 22 public replies without errors; all 16,384 raw pixels and all 69 coordinates from the three saved ROI paths match; absent acquisition spacing remains absent with unit native XY scale, while the explicitly calibrated fixture preserves 0.5 µm scale for both image and ROI. Original fixture hashes are unchanged and both viewers closed successfully. Published evidence: https://github.com/OpenHCSDev/openhcs/blob/2c53f50c7edeac1feae33e7ee0e4084d9c51fa67/docs/validation/acquisition-spacing-907/LIVE-ACCEPTANCE01.rst . This qualifies the real isolated installed candidate; it does not claim the unrelated default installation was replaced.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions