A 2-minute look

What if a test question could tell you how a student thinks?

The problem

One right answer collapses understanding to a single bit.

Multiple-choice grades what a learner picked — never the reasoning, the hesitation, or the near-misses along the way.

The idea

Select-All-That-Apply, generated and understood by AI.

Generate rich SATA items on demand — then read the behavior behind every answer.

The central abstraction

15 formats, four families.

Core Selection · 4 Enhanced Response · 3 Adaptive & Progressive · 4 Advanced · 4

Classic, Weighted, Matrix, Temporal, Meta-Cognitive and more — each targeting a different cognitive skill.

How it works

Generate → interact → understand.

Generate a schema-valid item→ Render the format template→ Capture the interaction→ Analyze the behavior

Selections, hovers, revisions and pauses stream as xAPI events into Google Cloud Storage.

Why trust it

Structure enforced at the boundary.

One LLM gateway. Every generated question validated against a JSON Schema with Ajv before it ever reaches a learner.

Coverage

15

SATA formats

9

LLM providers

3

User roles

0

Build steps

Every number verified against the repository — nothing invented.

Capabilities

Provider-agnostic, embeddable, self-hosted.

OpenAI · Anthropic · Gemini Azure · Mistral · Cohere DeepSeek · Groq · Ollama

Bring your own key and model. Embed any assessment via iframe. No bundler, no CDN.

The behavior signal

The trace is the insight.

Selection sequence, change counts, hover time and pauses become a diagnostic read — not just a score.

Who it's for

Instructors, learners, platforms, researchers.

Author varied banks in minutes · practice with 15 live formats · integrate via xAPI · study reasoning across formats.

xiangenSATA

Assessment that measures the thinking.