The digitisation of biodiversity data has reshaped ecological research. Global initiatives such as the Global Biodiversity Information Facility (GBIF; Heberling et al. 2021), the Atlas of Living Australia (Belbin et al. 2021), and the citizen-science platform iNaturalist (Mason et al. 2025) have aggregated billions of occurrence records into increasingly open repositories, underpinning advances in biogeography, conservation planning, and species distribution modelling. These infrastructures are horizontal by design: they serve the entire tree of life, harmonising heterogeneous sources through shared standards such as Darwin Core (Wieczorek et al. 2012) and exposing a rich array of derived products, maps, media, taxonomic treatments, predicted ranges, and analytical tools across all organism groups.

Despite this growth, a persistent gap remains between data accessibility and knowledge accessibility. Occurrence databases record where species have been observed, but extracting ecological meaning from them still demands technical effort. A researcher typically downloads large datasets, assesses their quality and cleans them, performs spatial analyses, computes distribution metrics, and interprets the results before usable insight emerges – a process that itself requires considerable time and expertise. For many users, including conservation practitioners, environmental managers, and students, this is a real barrier: increasingly rich in data, yet still thirsty for synthesis (Hui et al. 2024).

There are sound reasons why most large infrastructures stop at data provision and leave analysis to the user. User requirements evolve quickly and are difficult for a general portal to anticipate; embedding analytical functions risks obscuring methodological choices, making results harder to reproduce and audit; and a tool built for all taxa cannot privilege the specific conventions of any one. The deliberate separation of storage from analysis is therefore often the responsible default for a horizontal aggregator. It does, however, leave room for complementary, vertical resources that accept a narrower scope in exchange for delivering interpretation directly.

Meeting the FAIR principles of Findability, Accessibility, Interoperability, and Reusability (Wilkinson et al. 2016) is a necessary foundation for such resources, and good interface design does not substitute for it. Presentation is not itself a FAIR criterion, and a well-designed front end cannot excuse a system from exposing standardised, machine-retrievable, well-described data. The two operate at different levels: FAIR governs whether data and their derived products are discoverable and reusable by humans and machines, while interface design governs whether the communities who need those products can locate, interpret, and act on them in practice (Gadelha et al. 2021). How the platform meets each of the four principles and where it deliberately departs from the classic aggregator model is set out in the section on technical implementation below. The argument of this paper is that, once FAIR foundations are in place, the interface becomes a substantive and often neglected layer of biodiversity knowledge production.

Freshwater crayfish are a well-suited model for exploring this argument. Following the current global classification (Crandall and De Grave 2017), the group comprises the superfamilies Astacoidea and Parastacoidea, five families, 38 genera, and 669 species, or 692 counting distinct subspecies. This circumscription excludes the marine clawed lobsters that share the wider infraorder Astacidea, and it is the circumscription the platform applies: a query restricted to the infraorder rather than to the two crayfish superfamilies returns a considerably larger figure that is not comparable. Crayfish act as keystone and ecosystem-engineering organisms in fresh waters across the continents they inhabit. Invasive crayfish rank among the most damaging agents of freshwater biodiversity loss (Oficialdegui et al. 2020; O'Hea Miller et al. 2024), and a substantial fraction of species are assessed as threatened (Richman et al. 2015). Accurate IUCN Red List assessments for many European species remain hampered by fragmented distribution data and unresolved taxonomy (Pârvulescu 2025), with major lineage boundaries still under active revision (Pârvulescu et al. 2025). Knowledge of crayfish distributions is dispersed across disparate databases, grey literature, and peer-reviewed sources in many languages, and integrating it into a coherent, community-facing knowledge system is a genuine challenge.

World of Crayfish® (WoC®; Ion et al. 2024) was built as a response to this challenge. Rather than functioning solely as a repository, it is designed as a web environment that turns curated occurrence data into accessible ecological knowledge for the crayfish community. This article presents the platform's design principles, technical implementation, and core functionality. The analytical engines that generate its underlying spatial products are described in separate methodological contributions (Livadariu et al. 2026; Bâcu et al. 2026); here, the focus is the web architecture that translates those products into usable knowledge, what is generic about that approach, and what is specific to crayfish.

Because this article describes a live system rather than a completed artefact, it is worth stating unambiguously at the outset how much of it is functional. The platform is operational in all the components described in this article. Every module named below was verified against the live deployment before submission: the species profiles and their tabs, the distribution map and its layers, search and export, the tiered access model, the check-IN ingestion pipeline, the editorial correction channel, the account dashboard and administrative tooling, the Biodiversity Explorer, the check-Engine and check-Risk services, the species ecology panel and similarity network, the versioned Species Exposure Bundles and the public API, and the public engagement surface. None is a mock-up, a static page, or a design proposal. What is not built is confined to the three engines named in the following paragraph.

Three further engines named on the platform’s tools page (checkLINEAGE, checkMORPH, and checkTHREATS) are conceptual placeholders at varying stages of development, with no functional web component and no data products; a fourth name, chECkOLOGY, denotes an ecological data surface that is live (the species ecology panel and similarity network described below) but is not itself a distinct engine. These names signal the intended direction and are not presented here as peers of the operational components so that a reader does not infer from that page that five comparable engines are available. Only cheCkOVER and check-IN, together with the derived services described below, are operational.

Two limitations are likewise stated plainly rather than left to inference. Records enter through curatorial routes only, with no automated harvest from external aggregators, and no mechanism reconciles a record with subsequent changes in its originating source. Coverage is curated rather than exhaustive: the database holds fewer occurrences for many species than aggregators that ingest all available observations, a deliberate trade-off of volume for validated, taxon-specific depth.

WoC® draws on a curated global occurrence database compiled from museum collections, published literature, structured monitoring programmes, and registered-user contributions. The platform is live and continuously updated: at the time of writing, it held 124,318 validated occurrence records for 676 accepted crayfish taxa, including subspecies, in 34 genera and 99 countries, contributed by 110 registered contributors and spanning 8,565 occupied hexagonal cells. Of these taxa, 674 carry a classifiable native range structure across endemic, regional, and cosmopolitan categories (Fig. 1C); every record carries an explicit population-origin status, with indigenous and non-indigenous populations documented in parallel (Fig. 1D); and network-aware environmental-trait profiles are available for 275 taxa (Fig. 1E). Unlike a general aggregator, the database is not intended to be exhaustive: it favours records that have passed expert validation, and it will therefore hold fewer occurrences for many species than horizontal repositories that ingest all available observations. This is a deliberate trade-off of volume for curation and taxon-specific depth.

Records enter the database through two curatorial routes, and only these two. In the first, a contributor uploads records against a structured template; entries are parsed, validated against a controlled vocabulary, held in a pending queue, and released into the production database only after per-record administrative approval, at which point each receives a permanent accession identifier. In the second, records are extracted from published literature through check-IN, the ingestion engine, which combines AI-assisted extraction with geocoding, taxonomic checking, and bibliographic and sequence enrichment; check-IN produces drafts that a human must complete, and those drafts pass through the same approval gate. AI assistance is therefore never autonomous. Museum, monitoring, and citizen-science material enters by whichever of these routes fits, rather than through a separate channel, and receives the same validation.

Because literature ingestion carries most of the risk of error, check-IN deserves a brief note. A publication is entered via its DOI or as a PDF, and the engine reads its full text and supplementary tables to propose candidate records, grading each by the quality of its locality information as described below. Two disciplines govern the process. Coordinates may never be inferred: any coordinate proposed must be accompanied by the printed coordinate copied verbatim from the source, which the server checks before accepting it; where no coordinate is printed, the locality is resolved through a documented cascade of gazetteer and assisted methods whose output is always graded as low-accuracy and flagged for review. Claims of local extinction are held to stricter rules still: hedged or range-wide wording is never accepted, and a claim must name an exact site. Names are validated against WoRMS, accessions against GenBank, and bibliographic records are enriched through OpenAlex. The guarantee stated above is unchanged: check-IN produces drafts, which a human completes, and every record passes the same approval gate as a manual submission. The full extraction schema and the geocoding cascade are documented in Suppl. material 1.

The schema also carries associated organisms as first-class fields rather than as free text, reflecting the platform’s origin as a resource for crayfish and their pathogens (Ion et al. 2024): the scientific name of the symbiont or pathogen, its sequence accession, genotype group and haplotype, the year of the pathogen record, and a functional classification distinguishing ectosymbionts, viruses, bacteria, and fungi. One recording rule is applied deliberately: an associated organism is attached only where the source reports it as detected, and negative screening results are preserved as remarks rather than as records, since the absence of detection is not an occurrence. The distribution module accordingly offers a symbiont view that inverts the host-centred question, making the symbiont the subject with its own basin-level distribution across all hosts and a per-host breakdown so that the geography of an agent such as the causative organism of crayfish plague can be examined in its own right.

Two consequences follow and are stated explicitly. First, there is no automated harvest from GBIF, from an Integrated Publishing Toolkit installation, or from any other aggregator; all ingestion is curatorial. Second, records are one-way snapshots of their sources at the time of ingestion, and no mechanism detects or reconciles a subsequent correction in an originating museum record or publication; such changes must be found and applied by hand through the editorial channel described below. This is the platform’s principal interoperability gap, and closing it through source-identifier retention and periodic reconciliation is a development priority rather than an implemented feature.

Raw occurrence records are processed through the cheCkOVER framework (Livadariu et al. 2026), a generalisable pipeline that transforms occurrence records into standardised knowledge products. Distribution data are spatially aggregated to hydrographic basin units derived from HydroSHEDS (Lehner et al. 2008), giving an ecologically meaningful spatial frame. Area of occupancy (AOO) is computed at 2 × 2 km resolution following IUCN Red List guidelines (IUCN Standards and Petitions Committee 2024) and extent of occurrence (EOO) as a minimum convex polygon over verified localities. Environmental-trait profiles are obtained through the GeoFRESH platform (Domisch et al. 2024), which itself aggregates rather than originates its layers: each occurrence is snapped to a stream segment of the Hydrography90m global river network (Amatulli et al. 2022) and inherits climate (CHELSA), soil (SoilGrids), land-cover (ESA Land Cover CCI), and topographic variables at both local-segment and upstream-catchment scales. All spatial artefacts, bibliographic data, and summary statistics are downloadable in standardised, Darwin Core-aligned formats (GeoJSON, CSV, and BibTeX) from each species profile.

To protect vulnerable populations, precise coordinates are not displayed publicly and are replaced by aggregated spatial representations in all outputs, in line with recognised conservation data-sharing practice. Access is structured through three confidentiality levels rather than a simple public/private split: Level 0 records are fully open to any visitor; Level 1 records are available to registered users; and Level 2 records are confidential, released only through a formal request-and-approval workflow that involves the original contributor. This tiered model explains why an unregistered query for a heavily studied species may return few openly downloadable records even when many exist: the records are present and counted, but their release is governed by contributor consent. The model is described to the user at the point of query so that restricted counts are interpretable rather than surprising.

Rather than presenting its coverage as a static claim to be taken on trust, the platform exposes that coverage for direct inspection. A Biodiversity Explorer renders the entire dataset as a single, filterable global view in which every occupied spatial cell can be recoloured by taxonomic richness, by dominant population origin, or by sampling density and filtered by continent, country, sub-national unit, genus, species, and year. Any user can, therefore, audit the platform’s taxonomic and geographic coverage and its gaps for themselves, globally or for a custom-selected region, and contrast indigenous against non-indigenous representation instead of relying on a headline figure. This live auditability is itself part of the answer to the question of how comprehensive a curated, single-taxon resource can be: the honest answer is that comprehensiveness varies by species and region, and the platform makes that variation visible rather than concealing it behind a single number.

A distinctive feature of WoC® is the deliberate minimalism of its technical stack. The platform runs on vanilla HTML, CSS, and JavaScript on the front end, with PHP 7.3 and SQLite on the back end, deployed on standard shared hosting with no build step, no container orchestration, and no separate database server. The entire structured dataset – occurrence records, pre-computed hexagonal spatial cells, user accounts, pending submissions, and audit trails – lives in a single SQLite file (~430 MB) that can be backed up by downloading one file and inspected with any SQLite client. The species pages were originally built as a React application exceeding 1 gigabyte in static assets; they were rebuilt from scratch as a framework-free set of files of roughly 50 kilobytes, which improved performance, made mobile responsiveness native, and lowered the maintenance burden to the point where any collaborator with a text editor and file access can contribute. This architectural simplicity is not a limitation but a design choice that maximises portability and longevity, qualities that matter for infrastructure intended to outlast the project and the people who built it.

The platform communicates with a small set of free external services for specific tasks: molecular data (NCBI/GenBank), bibliographic discovery (OpenAlex), taxonomic validation (WoRMS), cartography (OpenStreetMap via Leaflet), geocoding (Nominatim), and open-access retrieval (Unpaywall), but all analytical products are served from its own file system. A dedicated in-house biogeographic API, implemented as PHP endpoints under /processed/api/, exposes pre-computed basin delineations, AOO, EOO, type-locality data, and species-level knowledge packages. These packages, termed Species Exposure Bundles – a concept developed and validated by Bâcu et al. (2026) as the smallest publishable, citable, and machine-retrievable unit of biogeographic evidence – are versioned, citable, machine-readable artefacts retrievable through a deterministic URL convention without authentication, so that any downstream tool, collaborator, or language model can fetch a species' narrative, metrics, and bibliography as a single, bounded unit. The platform's overall structure can be read as five operational layers (the contributor community, a curated data substrate, versioned processing engines, the human-facing knowledge surface, and a parallel machine layer), schematised in Suppl. material 2. The full endpoint catalogue (~48 endpoints, of which roughly 12 are deliberately public-facing) is documented in Suppl. material 1; the platform is therefore a user-interface-first system that exposes a curated machine layer, rather than an API with a front end attached.

The four FAIR principles (Wilkinson et al. 2016) are met as follows, and the qualifications are stated as plainly as the claims. Findability: every processed taxon carries a versioned Species Exposure Bundle addressed by a deterministic, persistent URL, and the public contract that lists those addresses is itself machine-discoverable from the site root through standard service-description link relations so that neither a human nor an agent needs to be told where the evidence lives. Accessibility: the knowledge tier is retrievable over HTTP without authentication, credentials, or negotiation, and the record tier, which is not open, is governed by a documented consent model with an explicit request pathway rather than by silent refusal so that what is withheld and why is itself discoverable. Interoperability: delivery uses standard formats (GeoJSON, KML, JSON, CSV, BibTeX, and CFF) and an OpenAPI 3.1 contract; taxonomy resolves against an external authority; and the occurrence schema is partially, not fully, aligned with Darwin Core, as detailed below. This is the principle the platform satisfies least completely at the record tier, and deliberately so: the platform is designed to deliver knowledge products rather than record streams, and it is at the knowledge tier that interoperability is complete. Reusability: all published knowledge artefacts carry a CC-BY-4.0 licence, provenance is retained per record, and every artefact is versioned so that a result can be reproduced against the exact state of the evidence from which it was derived.

One qualification belongs here rather than in a footnote. FAIR is a framework for data and metadata, and it is against data and metadata that the platform makes its claim; research software is addressed by a separate and later set of principles, and the two should not be conflated. The analytical engine that generates the platform’s knowledge products is published independently, with its artefacts openly licensed, as is conventional for a data-processing framework. The web application code that renders those products is application infrastructure rather than a research output and is not distributed, which means the platform should not be described as open-source software; it does not follow that the data are not FAIR, but the distinction is worth making explicitly rather than leaving a reader to infer a broader claim than is intended.

What is interoperable today operates at the knowledge tier. Each of the 676 processed taxa ships a versioned Species Exposure Bundle: aggregated spatial layers (basins, AOO, and EOO) as GeoJSON and KML, derived metrics split by population origin, narratives in three formats, and citations in four, all released under a CC-BY-4.0 licence and retrievable without authentication or credentials. The platform’s public, read-only API is documented as an OpenAPI 3.1 specification, permanently available at https://world.crayfish.ro/openapi.yaml, with interactive documentation at https://world.crayfish.ro/api.html. That contract covers 24 endpoints spanning platform statistics, the species catalogue, aggregated spatial coverage, biodiversity and risk services, per-species ecology, and the Species Exposure Bundles and is discoverable automatically: the site root declares it through standard service-description link relations, so an OpenAPI-aware client can locate the contract without being given its address. The specification is hand-maintained and is updated together with any change to the public endpoints. Any third party can integrate this layer today.

What is not open is the raw locality layer, and this is a policy rather than an unbuilt feature. Several crayfish species are legally protected, and published point coordinates pose a real exploitation risk, so record-level data move through a governed tier with contributor consent and are outside the public contract; no occurrence points appear in any published bundle. The accurate formulation is therefore that the platform is interoperable at the knowledge tier and governed at the record tier.

Interoperability at the record tier is partial and bounded by design. The occurrence schema is aligned with Darwin Core (Wieczorek et al. 2012) in vocabulary rather than in full: 15 of the 24 exported columns are Darwin Core terms used verbatim, nine are local, and several core terms are not yet represented, including basisOfRecord, eventDate (temporal information is currently stored at the resolution of year), occurrenceStatus, and coordinateUncertaintyInMeters; the establishmentMeans field also carries a platform-specific controlled vocabulary rather than the Darwin Core vocabulary. Record-level exchange with general infrastructures such as GBIF would require completing that alignment and publishing an explicit crosswalk. This lies outside the platform's purpose: WoC® delivers knowledge products rather than record streams, and the knowledge tier is openly and fully interoperable by design. Record-level export is therefore a deliberate limitation of scope rather than an unfinished feature.

Taxonomic names are validated against the World Register of Marine Species (WoRMS Editorial Board 2026), chosen for its specialist, actively curated decapod coverage, with the accepted name resolved for each record. Where a published name differs from the currently accepted name, the name as given in the source is retained in a dedicated field so that both remain recoverable and nomenclatural provenance is preserved at the level of the individual record; this field is functionally equivalent to the Darwin Core verbatimIdentification term under a local name and is a crosswalk candidate. The platform maintains its own WoRMS-anchored checklist, which is not currently reconciled with the GBIF Backbone Taxonomy.

Two conventions govern which taxa the platform counts. First, the checklist is anchored on the two crayfish superfamilies as circumscribed above, so that marine relatives within the wider infraorder are excluded by construction. Second, the platform counts taxa rather than species: subspecies are retained as separate entries because several are under active revision and may be raised to species rank, and collapsing them would discard the distinction the source literature makes. The figure reported by the platform, and in this article, is therefore a count of accepted taxa, and it is not directly comparable with counts of species alone, whether from the classification cited above or from other aggregated checklists that apply different circumscriptions or different treatments of subspecies.

The platform is organised around a simple principle: biodiversity information should be reachable through an intuitive interface that does not require specialised analytical tools (Bâcu et al. 2026). Information is organised into modular components, here meaning the organisation of the user interface into self-contained thematic panels rather than a modular software architecture, so that users can navigate complex material without being overwhelmed. On entering the platform, users encounter a streamlined interface of primary modules: species profiles, distribution mapping, search, and contribution, each addressing a different dimension of species knowledge.

The core of the platform is the species-profile interface. Species are selected through a two-step genus-then-species mechanism that reduces visual clutter in taxonomically rich groups. Once a species is chosen, a dynamic profile page integrates taxonomic information, spatial distribution, and contextual ecological knowledge, with classification displayed at the top and linked, where possible, to external authorities. Several of these interaction patterns – a hierarchical selector and a dynamically generated profile page – are shared with other biodiversity portals and are not in themselves novel; their value here lies not in any single widget but in what the profile integrates and synthesises for one taxonomic community and in the design principles that govern that integration, set out below.

The platform’s structure was driven by the concrete tasks its community performs rather than by an abstract feature list. These tasks were articulated as a small set of user pathways, each an explicit “I want to…” thread that begins with a user need and ends in a completed action on the live platform, and the interface was organised so that each pathway is short and self-evident. Eight pathways are currently supported and are documented in full, with their constituent steps, in the platform's interactive site map (https://world.crayfish.ro/sitemap.html).

They span the principal audiences the platform serves: understanding what the platform is (newcomers, communicators, and decision-makers); obtaining data for a specific species (researchers running an analysis); contributing new data (researchers, museum curators, and monitoring programmes); assembling material for a manuscript (authors and students, who can obtain a narrative, a publication-ready map, a formatted bibliography, and ready-to-paste Methods and Results text in a single session); working on invasive species (invasion biologists and managers); digitising published literature (literature curators); programmatic access (AI agents and automated workflows); and exploring crayfish through play (classrooms and casual visitors). Treating these pathways as the design unit, rather than the modules in isolation, is what keeps the interface legible as functionality grows, and it is the mechanism by which differing user requirements were identified and met. The pathways overlap and share steps, and the platform surfaces those overlaps so that a user pursuing one goal discovers adjacent capabilities.

It should be stated explicitly that these pathways are design hypotheses rather than the output of a formal needs assessment. They were authored from the developer’s own domain experience as a practising astacologist and from informal exchanges within the crayfish research community; no survey, structured interview, persona study, or usability test underlies them, and the platform carries no analytics, so no usage data exist against which to test which pathways are actually followed. The pathways remain the design unit for that reason, but their validation against measured user behaviour remains outstanding and is an obvious direction for future work.

Spatial visualisation is the central element of the species profile. Each page carries an interactive distribution map that occupies most of the screen and serves as the primary entry point for exploring biogeography. Occurrences are aggregated into hydrographic basin units (Lehner et al. 2008), which reduces visual noise relative to point displays while preserving ecological context; to protect sensitive data, precise coordinates are withheld and spatial information is represented through AOO grid cells at 2 × 2 km resolution.

A distinctive feature is the treatment of local extinction as explicit, positive information rather than as the mere absence of recent records. Documented extirpation events are encoded as first-class data points carrying a year, a locality reference, a buffer radius, and a provenance trail (Pârvulescu et al. 2026). This allows users to explore not only current ranges but also documented contraction and historical loss and gives the platform a temporal awareness that propagates into derived metrics. Because every product is versioned, the effect of such a claim is directly observable: the deposited snapshot cited in the Data availability statement contains both the baseline release and the first release in which documented extinction claims were applied, so the resulting change in derived metrics is reproducible from the archive rather than only visible in the live interface.

Users customise the map through selectable layers – indigenous and non-indigenous distributions, type localities, AOO, and EOO (Fig. 1A) – and can export the current view as a publication-ready, fully editable scalable vector graphic (SVG). Because these layers are independently toggleable and a user’s selection persists as they move between species, the map shows only the layers a user has currently enabled; the platform applies sensible per-species defaults so that the relevant layers are displayed on first view, and the legend adjusts dynamically to show only layers that contain data for the current species. As with any live system, the displayed picture may also be incomplete where validation of a given layer for a given species is still in progress, and users are reminded that the information shown is the current best-supported representation rather than a closed account. Species profiles are further contextualised by the GeoFRESH-derived environmental descriptors described above (Fig. 1B).

Population origin is carried throughout as a first-class attribute, with indigenous and non-indigenous records routed through parallel branches that never silently pool. It should be recognised, however, that this formally binary variable describes a biologically fuzzy reality: the indigenous status of a population may be contested and may change as evidence accumulates, as the reassessment of Austropotamobius fluviatilis in Spain illustrates. In the platform, origin is a declarative attribute: the value stored is the one the contributing researcher asserts, recorded against a strict controlled vocabulary and validated at ingestion. Where an incoming record falls within a species’ established indigenous extent, the ingestion interface offers the corresponding assignment as a default, which the contributor confirms or overrides; the suggestion is a convenience presented to a human and never a value written to the record on the platform’s own authority. Populations whose origin cannot be resolved are recorded as cryptogenic, and any assignment can be revised per record through the editorial channel described below, with the change logged and attributed. What the current model does not represent is disagreement itself: each record carries a single value, without a confidence score or competing assertions, so a contested population is represented by whichever assessment the curatorial process has accepted. Extending the schema to carry competing, attributed assertions is a recognised direction for development. The consequence is that a user may encounter an origin assignment they would contest. Populations of Austropotamobius pallipes in Britain and Ireland are a case in point: the platform records them as non-indigenous, following genetic evidence that Irish populations descend from a single western French haplotype absent from English populations and are best explained by human translocation (Gouin et al. 2003) and a wider review of introductions across the species’ western and southern range (Machino and Holdich 2006). Other authors treat the species as native to both islands. The platform does not adjudicate that disagreement; it records the assignment its contributing researchers made, exposes the provenance behind it, and provides a route to revise it. This is precisely the limitation described above, and it is why the correction channel is reachable from every map and species view rather than buried in an administrative interface. One consequence of the basin-level frame deserves emphasis because it is counter-intuitive for anyone accustomed to terrestrial organisms. Crayfish are confined to fresh water, so the unit of natural range is the drainage basin and not the landmass. Two basins may be immediately adjacent on a map and separated by nothing more than a watershed divide, yet remain entirely disconnected as aquatic systems: for an obligate freshwater animal, that divide is an absolute barrier, and there is no natural route across it. A population occupying a basin next to a native one is therefore not evidence of continuous natural range, and where molecular or historical evidence indicates arrival by human agency, it is recorded as non-indigenous. Adjacency on land carries no implication of connectivity in water.

Spatial visualisation gives an intuitive overview, but interpretation often requires context. The platform, therefore, embeds automated biogeographic narrative summaries in each profile (Bâcu et al. 2026): concise textual overviews of occurrence counts, geographic coverage, and representation across ecological and administrative regions, translating complex datasets into readable summaries that a user can lift directly into their own work. The narrative-generation process is grounded in the cheCkOVER framework (Livadariu et al. 2026), and each narrative remains linked to the dataset and the database version from which it was derived. Because every product is versioned, derived summaries are regenerated as the underlying data change, keeping them synchronised with the curated source rather than drifting from it; a version trend panel makes this history navigable. Bibliographic information is presented in multiple export formats (JSON, CSV, BibTeX, and CFF), with references ranked by frequency within the underlying dataset, so that the most influential sources for a species are immediately visible (Fig. 1B).

Beyond its research-facing modules, the platform repurposes the same curated infrastructure for outreach and education through the WoC® Arcade (world.crayfish.ro/craygame/), a suite of nine browser games whose content is drawn from the same validated knowledge base and generative engines that feed the serious modules rather than being invented for play. Four of them turn platform data into questions of recognition and recall: CraySlots is a biodiversity slot machine whose reels carry the platform's five ecological categories (native, endemic, threatened, invasive, and undiscovered) so that relative rarity is learned through repetition; CrayGuess asks the player to name a species from its real distribution map; CrayRank plays higher-or-lower on real platform metrics, such as which of two species holds more records, occupies more countries, or has suffered more documented extinctions; and CrayChallenge runs an escalating ladder of timed questions drawn from live taxa. Two reward exploration rather than wagering: CrayCollector is a field-collection game in which players capture species cards tied to real distribution hexagons and progressively reconstruct a species' mapped range, with the rarest captures flagged as potential discoveries on the player's profile; and CrayRaid is a reflex-based run down a crayfish river in which the chosen species' rarity class scales the score. CrayRescue casts the player as the custodian of a real threatened species, managing it as a living metapopulation network in which its actual river-basin populations are the nodes and making targeted per-population investments against pressures computed from that species' own atlas attributes; invasive pressure is derived from real range overlap and amplified where the invader is an ecological niche twin according to the platform's trait-similarity network, and recovery is rewarded in proportion to how endangered the species was. PinchyArt extends the same logic to cultural heritage: players identify crayfish depicted in works of art and, in a curator mode, submit their own finds for adjudication so that play contributes to a genuine cultural-heritage database rather than only consuming one. The flagship, World of Crayfish®: theGame, ties the arcade together: a player founds a research institute at a real location on the globe and grows it from the biodiversity actually recorded around it, with facilities that perform real operations on live data (surveying nearby occurrences, storing and analysing specimens, and synthesising regional state) and a landmark tower whose height tracks the player's real bibliometric record so that standing cannot be purchased with in-game currency. Throughout, a free virtual arcade currency draws against, but never spends, the contributor tokens that govern real data access, and no real money is involved; the allowance a player can draw is set by their contributor tier, which rises with the validated records they actually add, so the route to more play is more data. The games contain no invented content layer: their material is the same cheCkOVER-derived products and the same live occurrence database that the research modules serve, consumed in read-only mode. An illustrated in-arcade guide documents each game's rules and how they map onto the underlying data so that the educational claim can be inspected rather than merely asserted. Because every reel, question, comparison, map, card, artwork, and management decision is a window onto the same evidence a researcher would download, the Arcade demonstrates that a curated knowledge layer, once built, can be re-presented for audiences well beyond its original analytical scope.

Because the platform's authority rests on the trustworthiness of its records, data quality is treated as a first-order concern with an explicit two-pillar model. The first pillar is professional validation at the point of entry. Contribution is open to any registered user, but every submission, whether submitted manually, by batch upload, or extracted from the literature, passes through a gated review pipeline before reaching the production database. Registration itself is gated and requires institutional affiliation, an ORCID identifier, and a statement of purpose. Uploads pass through a sanitisation layer that normalises identifiers and terminology and flags probable taxonomic typographic errors using edit-distance matching; an administrator then makes per-record decisions to approve, revise, or reject, and each approved record receives a permanent accession identifier. Georeferencing quality is recorded explicitly: records are graded by the precision of their locality information, from explicit coordinates through geocodable water-body names to country-level-only data, and this grade travels with the record. WoC® thus favours expert-validated records over unfiltered crowdsourcing; citizen-science contributions are welcome but pass through the same validation as any other source rather than entering the database automatically.

The second pillar is community-driven correction after ingestion. Records age, taxonomies are revised, and coordinates drift, and a static database silently decays. The platform recognises an editor role between ordinary user and administrator: an approved editor can propose corrections to specific fields on selected records, but no correction is applied unilaterally. Each proposal follows a three-assent principle: the editor proposes, the original record owner accepts (or implicitly accepts through a defined response window), and an administrator applies the change against the master database, with every transition written to an audit trail and every applied correction credited to the editor. Platform-wide curation – including taxonomic sanitisation against WoRMS, normalisation of categorical fields, sub-national administrative attribution, and de-duplication – is performed by a custodial superadministrator role. Together, these mechanisms make curation an ongoing, auditable process rather than a one-off event, and they keep contributor sovereignty intact: ownership and consent are enforceable rather than honorary.

It is important to state precisely what WoC® does and does not claim relative to established infrastructures. General-purpose aggregators are not merely raw-data warehouses. A query to GBIF for a crayfish species returns far more than a table of points: record densities, images, type-specimen information, descriptions of native ranges and habitats, IUCN status, vernacular names, checklists, taxonomic treatments, observation trends, printable maps, and links to further sources, alongside an extensive suite of analytical tools, including species distribution modelling, available within the same environment. iNaturalist mobilises very large volumes of crayfish observations through citizen science, and the Atlas of Living Australia has long demonstrated how thoughtful web design and an open, reusable codebase can deliver national-scale biodiversity knowledge (Belbin et al. 2021). These platforms serve digested knowledge, not only data, and they do so for the general case. They cannot be the benchmark against which a single-taxon resource declares an advantage because they answer a different and broader question.

The appropriate baseline for what WoC® adds is therefore not a horizontal aggregator but the alternative a crayfish researcher would otherwise face: assembling occurrence records from disparate sources and from the general-purpose data repositories in which such records are typically deposited alongside a publication, and then performing the cleaning, harmonisation, spatial analysis, and synthesis themselves. Measured against that baseline, the platform's contribution is concrete. It integrates verified, taxon-specific distributional data with automated synthesis and interactive visualisation in a purpose-built environment, delivering outputs that need no post-processing and are immediately interpretable, and it does so using the conventions, the population-origin distinctions, the lineage sensitivities, and the conservation framings that the crayfish community actually uses and that a general platform cannot privilege without doing so for every other taxon as well. WoC® is therefore best understood as a complementary, vertical layer that consumes the same kinds of standardised evidence as the aggregators and returns community-tailored knowledge above them, not as an improvement upon them within their own domain.

Four explicit design principles distinguish this layer from a generic repository and recur across the platform’s modules. First, population origin is carried as a first-class attribute: indigenous and non-indigenous records are routed through parallel branches and never silently pooled, so that range metrics under IUCN criteria are restricted to the native range while the introduced range is preserved as a separate object for invasion biology. Second, loss is treated as information through the first-class extinction records described above. Third, environmental traits are network-aware, inheriting variables along the dendritic structure of river systems rather than through terrestrial proxies. Fourth, outputs are evidence-manifesting: structurally bounded, version-aware, and citable, designed to be retrieved as semantically complete units by humans and machines alike. These principles, rather than any individual interface element, are what the platform offers as a reusable contribution.

Although the platform currently serves freshwater crayfish, much of the approach is generic. The data model, the layered architecture, the curation workflow with its three-assent editorial channel, the tiered confidentiality and consent model, the narrative-generation and versioning logic, and the deterministic, citable machine layer are taxon-agnostic and could be applied to other organism groups with little modification. What is specific to crayfish is the content and the conventions: the WoRMS-based checklist and decapod taxonomy, the basin-level spatial frame suited to obligate freshwater organisms, the indigenous/non-indigenous and invasion-status vocabulary, and the particular conservation framings of the group. The platform is, in this sense, a pilot whose generic skeleton is reusable and whose taxon-specific flesh would be re-supplied for a new group.

Two boundaries should be stated plainly. First, the platform is under active development. Of the family of analytical engines presented on its tools page, cheCkOVER, the biogeographic pipeline underlying the products described here, and check-IN, the literature ingestion engine, are operational; the companion engines for ecology, phylogenomics, morphology, and conservation-threat assessment are planned and are designed to integrate when ready, but they are not functional at present and are not the subject of this paper. Second, on reuse: the platform’s data products and cheCkOVER artefacts are released under a CC-BY-4.0 licence, and the architecture is documented in sufficient detail (Suppl. material 1) to be reproduced; the platform source code is available from the author on reasonable request, and a generalised, openly licensed reference implementation is part of the development roadmap. Operational continuity, moreover, does not rest on the author alone: server and repository credentials are held in sealed deposit by the board of the International Association of Astacology, the international scientific society for the group, to be opened should the author become unable to administer the platform. Custodianship of the platform is therefore an institutional arrangement rather than a personal one. These are honest limits on the “pilot for other taxa” claim rather than caveats that undermine it: the design is transferable, and the path to a reusable codebase is explicit.

As biodiversity datasets continue to expand and the urgency of freshwater conservation grows (Richman et al. 2015), the challenge of turning data into usable knowledge becomes more pressing. General-purpose aggregators have largely solved the mobilisation problem for the whole tree of life; the complementary problem – delivering interpretable, community-tailored knowledge for specific taxonomic groups – remains comparatively under-served. WoC® is one worked example of how that second problem can be met, for a single group, with modest means: an expert-curated database, a deliberately simple and portable technical stack, a curation model that keeps the data alive, and an interface designed around what its community actually needs to do. Biodiversity informatics is entering a phase in which the binding constraint is no longer only data mobilisation but knowledge accessibility (Groom et al. 2023; Feng et al. 2025; Islam et al. 2026). Purpose-built, taxon-tailored knowledge interfaces, sitting above and interoperating with the general infrastructures rather than competing with them, may become a standard and necessary component of that future, and the design principles set out here are offered as a transferable starting point for building them.