Evidence gain
Acquired images, localized structures, morphology, specimen context, and reviewable derivatives move forward together.
Named architecture · bounded disclosure
Named architecture. Conservative claims.
SEPS V10 is a specimen-linked evidence architecture for keeping acquired images, localized morphology, limitations, and expert-reviewed language in the same record. The module names are visible. The production logic is not.
TEGUMENTHookFuseFUZZARIA
Design thesis
The system has two jobs: gain evidence and restrain claims.
Support should become more visible, measurable, and biologically coherent without allowing missing scale, conflicting context, or weak review to disappear behind a confidence score.
Claim restraint
Limitations remain active. The system may advance, pause, route to review, or remain unknown.
2× design move
Support and limitation travel together.
The public architecture exposes two counterbalanced directions without publishing the internal scoring recipe.
Public roles
Internal names, deliberately narrow public meanings.
The names create a memorable architecture; the descriptions below state only the role each component plays in the evidence chain.
TEGUMENT V10
The specimen-linked evidence spine: source imagery, context, review state, and versioned derivatives remain connected.
FAME
Registers interpretable support contributed by an acquired image without allowing visibility alone to become a taxonomic claim.
FLAME
Tracks whether visible support is localized to a defensible structure or region.
HookFuse
Brings morphology, selected regions, life stage, and specimen context into one biological review surface.
VOID
Retains missing scale, damaged views, conflicts, and other reasons a claim should stop or remain provisional.
FUZZARIA
Constrains how far reviewed evidence is allowed to travel in scientific language.
Morpho-Barcode
Carries the specimen-linked image, review state, derived measurements, limitations, and provenance forward as a portable record.
Evidence glass
Four views. One specimen-linked chain.
The visual layer distinguishes what was acquired, what was annotated, and what was derived.
Second design move
Claims are routes through an evidence state—not a single number.
Reviewed evidence can advance, pause, branch to molecular work, return for another image, or remain unknown. This topology is more conservative than allowing confidence alone to decide the language.
AdvanceEvidence supports the next bounded statement.
PauseMore scale, orientation, or review is required.
BranchVoucher or molecular confirmation becomes the next step.
UnknownThe record remains useful without being forced into a label.
Disclosure boundary
Serious enough to explain. Disciplined enough not to publish the recipe.
The public site names the functional architecture and the evidence standards. It does not expose production thresholds, weights, transformations, model internals, proprietary rule structure, sensitive specimen information, or unpublished taxonomic logic.
WithheldProduction thresholdsInternal weightsTransformation detailsModel internalsUnpublished taxonomic logic
Development audit
A locked evaluation matters more than an impressive demo.
The current reference audit includes 779 microscope fields and 1,620 linked annotations. Field-level partitions keep annotations from the same microscope field together, and 117 fields remain untouched for final evaluation.
These figures describe a recall-first localization component—not species identification, segmentation accuracy, or autonomous diagnosis.
779microscope fields
1,620linked annotations
117locked final-test fields
0final-test fields used for selection
Hard caps
Missing evidence limits the claim before confidence is considered.
No object supportNo parasite claim.
No usable mask or outlineNo quantitative morphometrics.
No verified scaleNo absolute-size-dependent rule.
No diagnostic viewNo species-level language.
Novel or high-impact recordExpert review and voucher linkage remain mandatory.
Collaborate at the evidence boundary
Bring the specimen, the comparison, the validation site, or the hard negative.
SEPS welcomes research, curation, microscopy, molecular, museum, agency, and external-validation partnerships.
Discuss collaborationView active projects
No diagnostic service is offered. Partnerships are research, curation, validation, and education focused.