Skip to content

Measure representative raw-feature envelopes before scale selection #525

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
trissim merged 2 commits into from
Oct 3, 2026
Merged

Measure representative raw-feature envelopes before scale selection #525

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
05927d8
Measure representative feature envelopes before scale selection
trissim Oct 3, 2026
4ced6eb
Merge remote-tracking branch 'openhcsdev/main' into docs/raw-feature-…
trissim Oct 3, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
73 changes: 73 additions & 0 deletions docs/validation/raw-feature-envelope-guide-20261003.rst
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,73 @@
Representative feature-envelope guide: source-only continuation
==============================================================

Canonical owner and scope
-------------------------

The original knowledge manifest maps ``openhcs_measurement_interpretation`` to
``packaging/codex/openhcs/skills/use-openhcs/references/measurement-interpretation.md``.
The original plugin skill bundle packages that same source; no new manifest,
generated guide, copied parameter catalogue or runtime policy is introduced.
Canonical reStructuredText entry/reference documents remain unchanged.

Existing guidance already requires distributed raw measurements, earliest-stage
diagnosis, one semantic change, regression controls and bounded development.
Its smoothing/spot/ridge evidence row nevertheless named only the narrowest
supported width. The edit replaces that incomplete sampling description with
an observed width/size/intensity envelope and explicit positive/faint controls.
It requires reflection of a compatible callable before a multiscale comparison,
including actual units/arguments/combination semantics and response-memory cost.
It promises no available filter, parameter name, scale setting or biological gain.
Claim-level conclusions link to the existing strategy owner rather than copy it.

Complete use-openhcs and Diataxis skills plus the relevant measurement and
installation references were read. Diataxis keeps this a working how-to
correction inside the current procedure. No scientific context, report, images,
pipeline settings, reference answers or object identities were consulted.

Ownership and preserved source
------------------------------

Current remote main is f7de9efad393bce83a3756a271f0cd4383783ea9. Existing isolated
``openhcs-input-preparation-20260929`` was reused non-recursively on a docs branch;
no new worktree, environment or package was created. Planck confirmed no borrower
of this checkout. Foreign dirty submodules remain untouched and excluded.
Published catalog511/b1d0b6c3e and its immutable qualified target are preserved.

Singer owns the separate viewer-retirement guide seam. Current guide520 already
distinguishes hiding from buffer release and preserves saved/current evidence,
but does not name the actual producer-incarnation request or measured memory
comparison. The target's real request takes ``expected_producers`` from complete
``producer_identities`` (including invocation key), not guessed route names.
Singer was asked to publish that correction through a new source-only guide PR;
this change does not edit his guide or runtime family.

Actual target04 check, not a new qualification
--------------------------------------------

Readonly target: ``engineering494/target04`` under the existing batch root.
Authoritative originals: ``READY-MAIN13.json``, ``PACKAGE09-QUALIFICATION.json``
and ``package08.log``. The ordinary wheel's SHA256 is
``d3cb03f47c2831a6287a6f510de931efcc861bd7c8c43a8076704eae074b3639``.
Original build source b26334ace99512970c81809b29d5ad19c7e07c4c; reviewed source
24927b864e3c7bc7e413f9929fbf917a5973ea5e. Planck confirms original full source,
knowledge and13-skill equality; no whole-wheel verification was rerun here.

Direct comparisons show installed strategy/custom-authoring/viewer-QA bytes equal
released511 b1d. Installed hashes:

* strategy: d9fef83e84a1d1cea928c9207971f8516ced350b5219b5c7f770dcb458e2710a;
* custom: 794b3ebab69e0256eeffeec4a2f683e9d82cc682e78aae43f6bcbe57da05b43d;
* viewer: 416e65e2e1745bd2547506ee97970d123cd8595a55dfc37be4fb00693daa2d90;
* measurement: fb25ac9aa3b777f775daa76242546406642d2a75b587ae8adb81ca74f18cfdd0.

Installed capability/DTO source exposes ``openhcs_retire_viewer_window_layers``
and its typed producer proof; the original guide bytes remain generic520 wording.
The522 validation/receiving notes are published source, not manifest-projected
knowledge resources. No claim is made that they were added to the installed skill.

This new measurement correction is NOT in target04. It needs normal source review
and future package qualification; no hot sync, import, test, build, install,
client/native/UI/provider call or scientific contact occurred. Frozen evaluations
and previous failure dispositions remain immutable. Source diff/link review is
not public installed retrieval or biological acceptance.
22 changes: 17 additions & 5 deletions ...aging/codex/openhcs/skills/use-openhcs/references/measurement-interpretation.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,10 +16,13 @@ measurements, not ground truth or an automatically validated parameter choice.
canvas pixels are not source pixels. World coordinates must be converted
through the layer transform before becoming index-space sample coordinates.
2. Choose clear isolated objects, a genuine close pair and faint/small examples
across preselected bright/dim, sparse/dense and centre/edge regions. Retain
extremes and ambiguity rather than measuring only objects the current
detector finds. For 3-D, inspect multiple Z planes and orthogonal views when
exposed; a projected width does not establish Z extent.
across preselected bright/dim, sparse/dense and centre/edge regions, at context
and native feature scales. Measure an envelope of supported widths, sizes
and local signal/background values, including narrow and broad structures,
clear positives and faint controls. One early feature is not the assay's
scale range. Retain extremes and ambiguity rather than measuring only objects
the current detector finds. For 3-D, inspect multiple Z planes and orthogonal
views when exposed; a projected width does not establish Z extent.
3. Use exposed native measurement capabilities where available and retain their
receipts. For bounded raw evidence, `openhcs_sample_viewer_window_image`
takes `route_key`, route-local `axis_indices`, native `y`, `x`, `height`,
Expand All @@ -40,7 +43,7 @@ measurements, not ground truth or an automatically validated parameter choice.
| --- | --- | --- |
| Object-size range | Long/short raw boundary spans across isolated and touching examples; per-axis extent for 3-D | Calling a current mask's size independent evidence, or confusing radius with diameter |
| Seed separation | Centre-to-centre distance of genuine neighbours and multiple maxima within one textured object | Using diameter as minimum separation and suppressing real close pairs |
| Smoothing/spot/ridge scale | Narrowest supported feature width, noise texture and nearby close-pair/path control | Equating diameter with Gaussian sigma or erasing a faint neurite |
| Smoothing/spot/ridge scale | Supported narrow-to-broad width envelope, noise texture, positive/faint-path and close-pair controls | Equating diameter with Gaussian sigma or assuming one scale preserves every supported width |
| Background-removal scale | Target width plus extent and variation of nearby background in multiple regions | A universal kernel radius or subtracting cell signal as background |
| Threshold/prominence | Raw object-versus-local-background values, weak positives, noise and saturation on the consumed channel | Deriving analytical thresholds from contrast limits, gamma or label colours |
| Roundness/shape prior | Isolated raw contours, elongated/lobed examples and an unsupported-shape control | Forcing every cell to be round or treating a round-looking mask as validation |
Expand All @@ -61,12 +64,21 @@ measurements, not ground truth or an automatically validated parameter choice.
differ between algorithms. Keep unsupported precision as an interval or
limitation. Label mask-derived estimates provisional and check against raw,
including missed objects; do not tune a detector solely from its own output.
A single-scale filter can favour one width class while suppressing another.
Inspect its response across the measured envelope before changing downstream
thresholds. If discovery returns a compatible multiscale callable, describe
its actual scale units, supported arguments and response-combination contract;
do not invent a scale-list parameter or assume a single-scale argument accepts
one. Estimate intermediate/response memory before a bounded comparison.
6. Compile one bounded candidate, inspect its earliest changed intermediate,
then compare matched raw/result/combined at the measured failures and
regression controls. Revisit distributed regions after every change; a
local repair can fail elsewhere under uneven illumination. Freeze measurement
receipts and rationale with the complete candidate before held-out access.
Expected counts or reference masks must not choose measurements in a blind run.
Link the measured envelope and remaining exclusions to the affected phenotype
claims through [the analysis strategy](analysis-strategy.md), which owns
claim-scoped conclusions rather than blanket abstention.

### Native ruler, profile and independently specified region operations

Expand Down
Loading