SYSTEM 1 · DECISION MODEL API

Your task.
Structured probabilities.

Define a prediction question, possible outcomes and the information available at the time. Receive a typed probability distribution your application can use directly.

Developer access coming later

Explore the contract while public API access is prepared.

Documentation preview · v1.1

POST /v1/predict

Task-oriented contract

Three top-level fields: model, task and input. Define the answer options in task.criteria. The response always follows our Choice schema, using your exact option IDs. This defines the possible answers rather than an arbitrary JSON layout.

Prediction request fields
Field	Meaning
model	The requested model release. The response identifies its exact version.
task.type	`choice`: select from two or more caller-defined options.
task.instructions	The prediction question.
task.criteria	A map from option IDs to descriptions (or null). Provide 2–20 mutually exclusive and exhaustive options; the API does not invent missing options.
task.resolution	The criteria and deadline used to determine the eventual outcome.
input.as_of	The information cutoff, with an explicit timezone.
input.context	Dated text or structured observations. No observation may be later than the information cutoff.

Download request JSON Schema · Download illustrative request

Request · illustrative task

{
  "model": "forecasting-one-r13-c",
  "task": {
    "type": "choice",
    "instructions": "Will the project reach its agreed milestone by the deadline?",
    "criteria": {
      "met": "The milestone is completed by the deadline.",
      "not_met": "The milestone is not completed by the deadline."
    },
    "resolution": {
      "criteria": "Use the dated acceptance record. These two outcomes are mutually exclusive and exhaustive.",
      "resolves_at": "2026-10-30T18:00:00Z"
    }
  },
  "input": {
    "as_of": "2026-10-13T12:00:00Z",
    "context": [
      {
        "type": "text",
        "text": "Illustrative context only. This is a schema example, not a supported live task.",
        "observed_at": "2026-10-13T11:00:00Z"
      }
    ]
  }
}

cURL · local gateway

This project-milestone example explains the schema. It is not an executed prediction or a claim of support by the current model. Schema validation and model capability are separate checks.

API documentation preview. Public access is not enabled yet.

Your options. A typed Choice result.

The core result has three fields: choice selects the highest-probability option; probabilities maps every requested option ID to its probability; confidence is reserved and currently null. These are illustrative values, not a live prediction.

{
  "type": "choice",
  "choice": "met",
  "probabilities": {
    "met": 0.6,
    "not_met": 0.4
  },
  "confidence": null,
  "calibration_status": "not_independently_verified"
}

Probabilities sum to one. Exact ties select the lexicographically first option ID. Confidence has no validated definition in this release; it is not copied from the highest probability.

Complete response · illustrative only

{
  "schema_version": "1.1",
  "request_id": "illustrative-only",
  "model": "forecasting-one-r13-c",
  "model_version": "r13-c@14cd833ac22b02cd876efeefa8b3e9c39bcd823f",
  "task": {
    "type": "choice",
    "instructions": "Will the project reach its agreed milestone by the deadline?",
    "criteria": {
      "met": "The milestone is completed by the deadline.",
      "not_met": "The milestone is not completed by the deadline."
    },
    "resolution": {
      "criteria": "Use the dated acceptance record. These two outcomes are mutually exclusive and exhaustive.",
      "resolves_at": "2026-10-30T18:00:00Z"
    }
  },
  "result": {
    "type": "choice",
    "choice": "met",
    "probabilities": {
      "met": 0.6,
      "not_met": 0.4
    },
    "confidence": null,
    "calibration_status": "not_independently_verified"
  },
  "generated_at": "2026-10-13T12:00:00Z",
  "input_as_of": "2026-10-13T12:00:00Z",
  "input_id": "sha256:dd328076700d3314b9a6c50741984a71d71e38879051bc35b4cacf922d7390f5",
  "timing": {
    "gateway_queue_ms": null,
    "model_queue_ms": null,
    "inference_ms": null,
    "load_ms": null,
    "cold_start": null,
    "gateway_total_ms": null,
    "cache_hit": false,
    "client_end_to_end_ms": null
  },
  "mode": "illustration"
}

Illustrative values only; no model was run for this example. mode: illustration and null timing distinguish it from a real response. Real successful responses use mode: research and measured server timing. Calibration has not been independently verified.

input_id is a server-generated digest for traceability. Model inference, model queue, gateway queue and loading time are separate. The client measures its own end-to-end time; a server timing is not a public-network latency claim.

Download successful-response JSON Schema

Explicit errors. No substitute forecasts.

API errors
HTTP	Error code	Meaning
400 / 413	INVALID_JSON / REQUEST_TOO_LARGE	Malformed JSON or a body exceeding 32 KiB.
401	UNAUTHORIZED	Missing or incorrect configured API credential.
422	INVALID_REQUEST / INVALID_TASK / INVALID_CRITERIA	Invalid fields, incomplete task, missing criteria or invalid option IDs/descriptions.
422	INVALID_CONTEXT / FUTURE_INPUT / INVALID_RESOLUTION	Invalid observations, information beyond the cutoff or an invalid resolution deadline.
422	EXPLICIT_TIMEZONE_REQUIRED	Timestamps must include a timezone.
422	MODEL_NOT_SUPPORTED / TASK_NOT_SUPPORTED	The model release or task is not supported by the configured adapter.
422	INPUT_QUALITY_UNVERIFIED	The input has not passed the deployment's quality review.
429	RATE_LIMITED / QUEUE_FULL	Shared capacity is limited; retry with backoff.
503	MODEL_NOT_READY / MODEL_OFFLINE / AUDIT_UNAVAILABLE	Inference or required recording is unavailable.
504	MODEL_TIMEOUT	Queue or inference deadline exceeded.
502	INVALID_MODEL_RESPONSE / INVALID_MODEL_TIMING	Upstream output or timing violates the contract.

One contract.
Your decision workflow.

The v1.1 contract supports a single Choice task, with binary or multiclass answer sets. Text and structured context are input formats, not guarantees that a given model can use every input.

Model adapters enforce validated task scope. A schema-valid request may return TASK_NOT_SUPPORTED. The current research adapter accepts only privately reviewed task bindings; arbitrary tasks and new context inputs are not enabled. Score, Noul, batch questions and arbitrary response JSON Schemas are outside this version. The Choice fields are inspired by typed decision APIs; this endpoint is not a drop-in Jev SDK replacement.

Developer access is coming later.

No public prediction endpoint is connected. This is a preview of the contract, with a local gateway implementation.

Supported task profiles, authentication details and service availability will be supplied when access opens. Keep credentials in your backend.

Your task.Structured probabilities.