Prototyping · Shape
Prototyping · ShapeExplanation Types
A progressive-disclosure pattern for explanations: surface (one line), detail (structured reasons), full (provenance + confidence), matched to the stakes of each decision.
Why it matters
Explanations calibrated to trust needs: terse when stakes are low, richer when the user is deciding whether to comply.
The canvas
The worksheet you fill in, as a template. Use it with your team here, or fill it in online in the workspace.
Project Team Date
Which AI decision does the user need to evaluate or override?
Default explanation type
- Data use: what data is collected and how it is used (always)
- Partial · generic: high-level system logic (low stakes)
- Partial · specific: why this decision was made (medium stakes)
- Full description: complete logic and decision factors (high stakes, experts, regulatory)
- Categorical: outcome categories (binary or categorical decisions)
- N-best: top alternatives with confidence (several valid options)
- Numeric: exact probability or confidence (technical users, financial or risk)
- Data viz: confidence as charts (complex data)
- Experimentation: what-if exploration (users need cause and effect)
- No explanation: the system acts without explaining (trivial, routine)
High-stakes explanation type
- Data use: what data is collected and how it is used (always)
- Partial · generic: high-level system logic (low stakes)
- Partial · specific: why this decision was made (medium stakes)
- Full description: complete logic and decision factors (high stakes, experts, regulatory)
- Categorical: outcome categories (binary or categorical decisions)
- N-best: top alternatives with confidence (several valid options)
- Numeric: exact probability or confidence (technical users, financial or risk)
- Data viz: confidence as charts (complex data)
- Experimentation: what-if exploration (users need cause and effect)
- No explanation: the system acts without explaining (trivial, routine)
Surface
The one-line explanation
Detail
The structured reasons
Full
Provenance and confidence
Where could you over-explain?
Full transparency on trivial decisions destroys trust as fast as opacity on serious ones.
app.jcerejo.com · The Anticipatory Design Framework
Fill it in online: start freeFree: one framework project. No card.