Developers
The coding scheme is a file, not a prompt
Tagaroo's agent does not have a hidden prompt that decides how your constructs are coded. It reads a skill file: a plain Markdown document that states the definition, the boundary cases, the reasoning required before a verdict, and the severity anchors. That file is yours, it is reviewable by the person who owns the construct, and the format it follows is open.
The skill-file format
Write a construct down once — definition, boundary cases, reasoning scaffold, severity anchors — and any conforming implementation can apply it. Two conformance levels, so a definition-only file is still valid.
Project interchange
REFI-QDA (.qdpx) in and out: text sources become annotatable assets, codes become phenomena, existing codings arrive as annotations. Annotations also export as CSV. Migration works in both directions, which is the only honest way to ask anyone to switch tools.
Reproducible evaluation
The validation suites are downloadable, every published result records the model and prompt version it was measured under, and the evaluation code is in the repository. You can disagree with our numbers using our own data.
Licensing, plainly
- The specification is CC BY 4.0 and the reference parser is MIT. Implement it in anything, including a competing product.
- Tagaroo's own curated skill files are not open. The Scale Library's operationalisations are what a subscription pays for; they run server-side and are never sent to a client. What is public is everything you need to write your own.
- Instruments are a separate matter. Many rating scales are copyrighted, and some permit free use only for individual clinical or supervisory practice — an operationalised version is still a derivative work. Write your own operational wording rather than reproducing a manual, and record the instrument's terms in the file so whoever uses it next inherits the constraint.
The format is described in Gutiérrez E, Zhang Y, Navarro J-B, Barajas A, Translating rating-scale ontologies into LLM agents for ambient clinical phenotyping (npj Digital Medicine, submitted). The research implementation is archived at doi:10.5281/zenodo.20435439 and developed at github.com/Kikolo3000/clinical-skill-architecture.
Frequently asked questions
What is a skill file?
A Markdown document that operationalises one construct from a rating scale or codebook: its definition, positive and negative examples, contrastive pairs against confusable neighbours, the questions a coder must answer before deciding, the cases that disqualify it, and the severity anchors. A model applies it; a human reads and corrects it. The format is open and documented.
Is the format tied to Tagaroo?
No. The specification is CC BY 4.0 and the reference parser is MIT, so you can implement it in anything — including something that competes with us. Tagaroo is the reference implementation, and the research implementation from the accompanying paper is a second one.
Does Tagaroo have a public API?
Not yet. The interoperability that exists today is at the data level: REFI-QDA (.qdpx) import and export, CSV annotation export, and the open skill-file format. A documented HTTP API is not published, so treat the format and the exchange files as the integration surface for now.
Can I export my data?
Yes, in open formats. A study serializes to REFI-QDA .qdpx — the exchange format NVivo, ATLAS.ti and MAXQDA all read — and annotations export as CSV. The same path works inbound, so a codebook does not have to be retyped to try Tagaroo, and nothing is trapped here if you leave.
Are Tagaroo's own curated skill files open?
The format is; those particular files are not. The Scale Library's operationalisations are what a subscription pays for, so they run server-side and are never sent to a client. Everything needed to write your own — the specification, a reference parser, worked examples and a conformance corpus — is public.