Plate IIIMorphology → Identity
A morphological search engine that bridges human observational expertise with automated classification speed — guiding users from specimen to ranked species candidate through structured, auditable faceted filters.
⟢ Overview
Foram identification has historically required either years of specialist training or access to expert colleagues. Testaria restructures that bottleneck: it encodes the taxonomic key as a faceted filter over a character matrix, so a user with a specimen under the microscope can work systematically from morphological observations to a ranked candidate list — without needing to hold the entire literature in memory.
The natural-language bridge lowers the interface barrier further: rather than learning filter syntax, a user describes what they see and the system proposes the corresponding character states. Human judgment remains in the loop at every step — the user confirms the proposed filters before the query runs, and the final identification is always a human decision, not an automated assignment.
⟢ Identification Workflow
Describe the test in plain language or select character states directly from the faceted interface. A natural-language bridge converts typed or spoken descriptions into structured filter parameters — you review and correct before any query runs.
Filters span a minimum set of ~5–8 diagnostic characters — chamber arrangement, aperture type, wall structure, coiling direction, and related features — each drawn from a controlled state vocabulary.
Testaria returns a ranked list of species candidates, ordered by how closely their character profiles match the observed specimen. Each candidate is AphiaID-keyed and cross-referenced to WoRMS.
Accepted identifications export as a CSV identification sheet and a reliability report. The complete decision path — characters applied, weights, source references — is included for peer review.
⟢ Character Model
Testaria is built around a minimum set of approximately five to eight diagnostic morphological characters — the fewest needed to discriminate among common taxa at the genus and species level. Each character has a finite, governed set of states drawn from the controlled vocabulary. Adding more characters is possible; adding free-text state values is not.
Character examples include: chamber arrangement (planispiral / trochospiral / uniserial / biserial), aperture position and shape, wall composition, coiling direction, and sutural characteristics. The exact minimum set is under active design alongside the morphological character model.
Chamber arrangement
Aperture type
Wall structure
Coiling direction
⟢ System Capabilities
Each morphological character is a discrete facet. Filters narrow the candidate pool at each step; the result set updates immediately.
Describe a specimen in text (or by voice). The NL bridge proposes corresponding facet states; you verify them before the query runs. The LLM never acts autonomously.
Results are scored by character-profile similarity, not alphabetical order. Ties and near-matches are surfaced explicitly, not silently resolved.
Every identification is reproducible: the same character inputs produce the same ranked output. The decision path is exportable for protocol documentation.
All candidates are keyed by AphiaID and synced to the World Foraminifera Register. Testaria never forks taxonomy or introduces local name variants.
Accepted results export as CSV — including the specimen description, applied characters, ranked candidates, and a reliability indicator.
"A systematic filter that guides users through taxonomic keys while blending expert intuition with structured AI logic — reproducible, auditable, and compatible with existing lab protocols."
⟢ Try It Now
The current MVP runs on mock data. The filtering workflow, natural-language interface, and candidate ranking are fully functional — ready for exploration and feedback.