# Blind test request: AISA 11.1 against BCIO-annotated cases

[oleksiikartashovde-glitch]

BCIO tool

Query or bug description

What I am asking for

I would like to request a blind test.

If the BCIO team has annotated cases, benchmark tasks, or any structured examples where the correct ontology mapping is already known — I am ready to run them through AISA 11.1 and return the output for your evaluation.

The test would work as follows:

1. You provide a case or task in any format — plain language, intervention description, study excerpt, or a structured prompt
2. I run it through AISA 11.1 without knowing your expected output
3. I return the full bot response
4. You evaluate the result against your known correct mapping

I do not participate in the evaluation. You assess it on your own terms.

Why this may be useful for the BCIO team

AISA 11.1 is an AI system built entirely on the BCIO ontology. It uses BCIO IDs — not free labels — as the core output layer. A blind test would show how a real AI runtime handles your ontology under natural-language input conditions: where the mappings are accurate, where they drift, and where the ontology itself has gaps that a runtime system needs to work around.

This is external validation of BCIO in an applied AI context, which — to my knowledge — has not been done systematically before.

What AISA 11.1 can currently do

The system works across all 13 BCIO modules. It performs:

• natural-language normalisation into ontology IDs
• module-level routing using full ID suffix chains (.bc, .be, .ma, .d, .r, .e, .f, .sh etc.)
• neighbourhood checking (parent / children / siblings) before confirming a match
• multi-step linked analysis: explicit problems → patterns → ontology IDs → target behaviour → gap analysis → COM-B → TDF → APEASE → 13-module solution breakdown
• cross-domain application: healthcare, workplace, retail, policy, digital products

For the five Coming Soon modules (Dose, Reach, Engagement, Fidelity, Schedule), the system uses an author-built interim codification called DREFS — 260 entries, 224 unique IDs, all grounded in verified ontology identifiers from BCIO, MeSH, GO, APA, and iSci. This is documented separately and treated explicitly as a runtime layer, not official BCIO doctrine.

Resources

• Bot: https://chatgpt.com/g/g-69a6ab4b5adc81919a5be72eecbd2f81-aisa-11-1
• Article (BCIO use vs AISA 11.1 comparison): https://ssrn.com/abstract=6341738
• Zenodo release: https://doi.org/10.5281/zenodo.19388359
• GitHub: https://github.com/oleksiikartashovde-glitch/AISA-11.1-BCIO
• User guide: https://github.com/oleksiikartashovde-glitch/AISA-11.1-BCIO/blob/main/docs/REFERENCE_UserGuide.md

What I am not asking for

I am not asking for endorsement, co-authorship, or any form of validation of AISA 11.1 as an official BCIO product. I am asking for a task. You run the evaluation. The results go wherever you find them useful.

If no annotated cases exist

If the BCIO team does not currently have annotated benchmark tasks, I would be glad to discuss whether creating a small shared benchmark set would be of interest — even a set of 3 to 5 cases with agreed correct mappings. This could be useful for the broader community working with BCIO in computational contexts.

---

Oleksii Kartashov
oleksii.kartashov.de@gmail.com
ORCID: 0009-0003-2898-6422