Glossary
Typed decisions
Typed decisions are answers limited to a format you declare up front (a pick from a list, a level on a scale, or yes/no) and returned with probabilities.
Swiss-hosted inference. Nothing you send is ever stored. Swiss data residency
Definition
What Typed decisions means
Typed decisions are model outputs that belong to a type you declared in advance, instead of free text. You ask a choice question and get one of your labels; a score question and get a level on your scale; a noul question and get a probability that a statement is true. Each answer comes with the full distribution and a confidence, so downstream code can branch on it directly.
The typed-decisions benchmark and checkpoint
In the Laya family the term also names a specific benchmark and a fine-tuned checkpoint. The benchmark has 400 cases and 2,000 decisions across four synthetic workflows, each with an exact set of question ids:
| workflow | question ids |
|---|---|
| customer_service | action, category, churn_risk, needs_human, urgency |
| invoice_processing | discrepancy_severity, disposition, duplicate, matches_order, urgency |
| security_incidents | credential_compromise, disposition, severity, true_positive, urgency |
| agent_trace_observability | action, needs_review, outcome, risk, urgency |
laya-typed-decisions (ModernBERT-large, 1,024-token context) is fine-tuned on that benchmark's training split. It scores 0.766 accuracy with a Brier score of 0.062, above the 0.735 teacher self-agreement ceiling and above Jev 1.13.0's published 0.727. Per workflow: invoice processing 0.804, security incidents 0.766, customer service 0.764, agent-trace observability 0.730.
The base checkpoints are near chance on the same benchmark zero-shot (0.362 English, 0.352 multilingual, against 0.318 random and 0.461 majority class). The capability comes from fine-tuning.
When it is used
The Router selects the typed-decisions checkpoint only when you ask for it (model: "typed-decisions") or when auto task detection is enabled and your question ids exactly match one of the four sets above. It is never a silent default. See typed decisions, and the invoice processing and security alert triage use cases.
Related terms
How Typed decisions connects to the rest of the vocabulary.
Go deeper
- ExplainerTyped decisions: asking a model for a value, not a paragraphWhat is a typed decision? You list the allowed answers up front and the AI returns one of them with a probability. No free text to parse, no invented labels.
- Use caseInvoice processing decisions with a fine-tuned decision modelAutomate accounts-payable decisions after extraction: purchase-order matching, duplicate detection, discrepancy severity and approve, hold or reject.
- Use caseSecurity alert triage with calibrated decisionsTriage SIEM and EDR alerts with a fine-tuned decision model: true positive, credential compromise, severity and next step, with calibrated confidence per alert.
- Use caseSupport ticket triage with a calibrated decision modelAutomate support ticket triage: one API call reads each ticket and returns intent, urgency, frustration and churn risk, with a confidence score per answer.
Get started
Try it on your own text
Send any text with your own typed questions and get a calibrated probability for every answer. 6k free tokens, no card needed; then $0.0357 per 1M input tokens — 15% below Jev.
Swiss-hosted inference. Nothing you send is ever stored.