SEPSSociety for Evolutionary
Parasite Systematics
Home / Methods
Technical overview

A morphology-first evidence system—not an autonomous taxonomist.

SEPS uses computation to organize, localize, measure, compare, and audit biological evidence while keeping specimen context and expert authority attached to every claim.

SEPS V10 suite

Advanced enough to integrate evidence. Conservative enough to stop.

TEGUMENT V10 is the public name for a specimen-linked, fuzzy-first evidence architecture that separates what an image supports, where that support is located, whether the biology is coherent, what limits the claim, and how the result may be described.

This is a public architecture view. Internal thresholds, weights, production settings, transformation details, and unpublished decision logic are intentionally not disclosed.

High-level public architecture of the SEPS V10 suite
Functional roles only. The system may advance, pause, route to review, or retain an unknown outcome.
Architecture

Three layers, one evidence chain.

TEGUMENT

Image and measurement layer

Preserves raw files, links biological and technical metadata, supports candidate localization, masks, calibration, landmarks, outlines, texture, and versioned derivative outputs.

HookFuse

Biological reasoning layer

Asks whether visible structures, life stage, host, infection site, scale, specimen condition, and comparison material make biological sense together.

FUZZARIA

Claim-control layer

Determines what the record is allowed to say: archival only, broad-group support, review required, confirmation ready, or unknown.

Evidence surfaces

One specimen can produce several kinds of evidence. They do not become interchangeable.

Acquired images, expert annotations, and derived geometry are displayed as different objects and linked back to the same specimen record.

Acquired light-microscopy specimen view
Acquired imageWhole-context microscopyField, orientation, and acquisition context remain part of the record.
SEM view of attachment structures
Acquired imageDiagnostic SEM structureFine morphology is retained without detaching it from the parent specimen.
Expert annotated nematode image
Expert annotationAnatomical interpretationHuman review remains distinguishable from the raw microscopy beneath it.
Derived ellipse and landmark geometry over a selected image region
Derived overlayReviewable geometryMeasurements and overlays are versioned derivatives—not replacements for the image.
Signature modules

Internal names, deliberately narrow public roles.

Each name identifies a function in the evidence chain rather than a claim of autonomous diagnosis. Their separation is intentional: support can rise while the allowable conclusion remains limited.

Public-safe by design. The website describes responsibilities and review boundaries, not proprietary scoring formulas or unresolved taxonomic logic.
00

TEGUMENT V10

Evidence spine

Links source imagery, specimen context, calibrated derivatives, review state, and provenance across the workflow.

01

FAME

Earned image evidence

Summarizes how much interpretable support the acquired image contributes to the selected morphology lane.

02

FLAME

Localized support

Tracks whether the supporting cues are attached to a defensible region, structure, or review target.

03

HookFuse

Biological fusion

Reconciles regional image evidence with morphology, life stage, host context, and specimen condition.

04

VOID

Claim-limiting evidence

Preserves blur, obstruction, missing scale, uncertain orientation, source ambiguity, and other reasons to abstain.

05

FUZZARIA

Reporting restraint

Maps the reviewed evidence state to controlled scientific language and appropriate escalation.

06

Morpho-Barcode

Portable record

Carries the specimen-linked images, derivatives, measurements, limitations, decisions, and provenance forward.

Hard caps

Missing evidence limits the claim before confidence is considered.

No object supportNo parasite claim.
No usable maskNo quantitative morphometrics or outline analysis.
No verified scaleNo absolute-size diagnostic rule.
No diagnostic viewNo species-level claim.
High out-of-distribution riskPreserve, route to expert review, and retain an unknown outcome.
High-impact or novel recordExpert review and voucher linkage remain mandatory even when visual support is strong.
Data integrity

The hierarchy prevents images from becoming detached from biology.

Records are organized as collection event → host → parasite specimen → imaging session → raw image → versioned derivatives. That structure supports traceability, limits pseudoreplication, and makes specimen-level partitioning possible.

Raw evidence
Original files are preserved and never overwritten. Crops, masks, overlays, measurements, and figures are derived products with their own identifiers and processing notes.
Scale and calibration
Calibration status is stored explicitly. Images without verified scale remain usable for some normalized shape analyses but are blocked from absolute measurement claims.
Review and provenance
Reviewer status, corrections, reasons for routing, software versions, configuration, and voucher or sequence links remain attached to the record.
Field-level split with locked final test
The current development audit keeps all annotations from a microscope field in the same partition and leaves 117 fields locked for final evaluation.
Recall-first localization

Find plausible objects first; refine the boundary later.

The current detector is a proposal engine. It scans whole microscope fields, ranks candidate regions, and gives later modules a manageable set of places to review. False positives are acceptable at this stage when they reduce the risk of missing a true object.

Centroid capture is therefore the primary developmental target. Boundary overlap, taxonomy, diagnostic interpretation, and reporting language belong to later, separately validated stages.

Exact production settings, unpublished training details, sensitive specimen information, and unreviewed taxonomic conclusions are intentionally omitted from the public site.

Contact sheet of microscope fields with candidate boxes
Developmental full-field candidate localization on reference images.
Validation sequence

Every module must earn the next claim.

Module tests

Localization recall, mask quality, scale validation, measurement repeatability, shape residuals, calibration, and failure maps.

Expert comparison

Specimen-level splits, expert masks, inter-reviewer agreement, difficult negatives, and audit of final reporting language.

External challenge

New specimens, microscopes, operators, preservation states, laboratories, and out-of-distribution material.