pretickOutcome probabilities
Define the possible results of a question. Jev uses your evidence to assign each a probability.
This experimental endpoint complements numerical forecasts. It returns named outcomes, not future values or a price path. Try it in the playground · Model card and test results
Request
POST https://pretick.ai/v1/outcomes/forecast Authorization: Bearer YOUR_PRETICK_API_KEY Content-Type: application/json
{
"model": "typesafe/jev-1.13",
"question": "At the next daily observation, will the price be below 102, between 102 and 104 inclusive, or above 104?",
"context": "Daily price observations in USD. This is illustrative example data, not an actual market.",
"series": {
"data": [
100,
100.5,
101,
100.8,
101.4,
102,
101.8,
102.4,
103
],
"horizon": 1
},
"outcomes": [
{
"id": "below",
"description": "Price below 102"
},
{
"id": "within",
"description": "Price between 102 and 104 inclusive"
},
{
"id": "above",
"description": "Price above 104"
}
]
}Use your existing key and trial or paid credit. A successful request costs $0.02. Only Jev is called; no additional numerical-model forecast runs automatically.
Define the evidence and question
question specifies what to estimate and by when. Provide context, series, or both. Include units and relevant facts. Jev does not search for live data. Each outcome needs a unique ID and a description; make outcomes mutually exclusive and collectively exhaustive. Pretick validates their structure, but cannot prove that natural-language descriptions cover every possibility.
series accepts the existing numerical input formats: pasted numbers, timestamp/value pairs, multiple columns and comparable sessions. Omit model and domain inside it. Pretick normalizes observation keys, computes count, first/latest value, minimum, maximum, mean and change, and supplies requested future keys as context. There are no invented observations. The question must agree with that future horizon; the API does not resolve contradictions in prose.
Use ISO timestamps with an offset or Z. Timestamp cadence is inferred. Values-only data uses sequence keys and defaults to daily frequency; specify frequency when relevant. Horizon defaults to input length, so set it explicitly for a single future outcome. This reuses the numerical parser’s scheduling rules and limits; Jev does not run a TSFM behind the scenes.
Limits
| Field | Limit |
|---|---|
| Question | 1,200 characters |
| Outcomes | 2–12; descriptions up to 500 characters each |
| Outcome ID | Up to 40 lowercase letters, digits or underscores; starts with a letter; unique; constructor and prototype are reserved |
| Context | 10,000 characters |
| Series | 3–512 prepared observations, up to 4 columns |
| Body | 64,000 UTF-8 bytes; prepared provider input at most 48,000 bytes |
Oversized or unsupported input fails before billing. Data is never silently truncated.
Response
Read selected_outcome and probabilities. outcomes includes each original description and its probability. The distribution sums to one; rounding differences up to 0.0001 are normalized and disclosed. The response also includes model_version, status: experimental, calibration: not_validated, parsing metadata when series is supplied, request_id, cost_cents and balance_cents.
These are model estimates, not calibrated event odds or confidence intervals. Jev’s probabilities do not participate in the numerical consensus. Our initial 18-day gas-price test did not beat a historical-frequency baseline on probability quality; see the model card for the method and results.
Errors, billing and data handling
400: invalid request. 401: invalid key. 402: insufficient credit. 413: oversized body. 503: unavailable, failed inference or accounting pending. Provider errors, timeouts and invalid distributions trigger a refund. Check refunded; retain request_id if accounting or refund is pending. Allow a 60-second client timeout. Do not automatically retry: each successful request is separately charged. Check GET /v1/usage first; it contains request accounting, not recoverable results.
Evidence and question text are sent to TypeSafe through OpenRouter. Pretick does not store that content in its database. Your Pretick key, email and balance are not sent to the model provider. Data handling.
AI agents
Use the MCP forecast_outcomes tool with the same request shape, existing OAuth connection and balance. list_models distinguishes numerical forecasting models from outcome models. This is a separate tool from forecast and forecast_advanced. Refresh your connection’s tools if it is missing.