Predictions

Individual Agent

One specialist agent's frozen prediction for the live window of a market.

Overview

GET/v1/predictions/{asset}/{timeframe}/{agent}

Developer PreviewAll plansapi.probora.xyz

Query any of the nine specialist agents for any asset and timeframe. Every response includes direction, UP/DOWN probabilities, confidence, model version and the instance's public track record.

Parameters

Parameter	In	Type	Description
assetrequired	path	string	Asset symbol. One of BTC, ETH, SOL, XRP, DOGE, HYPE, BNB, GOLD, SILVER, OIL.Example: `BTC`
timeframerequired	path	string	Window length. One of 5m, 15m, 1h, 4h, 1d.Example: `15m`
agentrequired	path	string	Agent slug: trend, momentum, mean-reversion, rsi-divergence, order-flow, volatility, liquidation, oi-funding, whale-flow or meta.Example: `order-flow`

Request

# export PROBORA_API_KEY=prb_live_…
curl "https://api.probora.xyz/v1/predictions/BTC/15m/order-flow" \
  -H "Authorization: Bearer $PROBORA_API_KEY"

Response

200 OK, returned as application/json. Example:

Response · 200

{
  "asset": "BTC",
  "timeframe": "15m",
  "market_id": "btc-15m-20261114-2000",
  "price_to_beat": 76230,
  "agent": "Order Flow",
  "agent_slug": "order-flow",
  "status": "published",
  "direction": "UP",
  "up_probability": 0.682,
  "down_probability": 0.318,
  "confidence": 0.84,
  "historical_accuracy": 0.638,
  "sample_size": 48291,
  "model_version": "orderflow-btc15-v12",
  "prediction_timestamp": "2026-11-14T20:00:00.412Z",
  "settlement_timestamp": "2026-11-14T20:15:00.000Z",
  "data_mode": "live"
}

Response fields

Field	Type	Description
asset	string	Asset symbol, e.g. BTC.
timeframe	string	Window length: 5m, 15m, 1h, 4h or 1d.
market_id	string	Deterministic market id {asset}-{timeframe}-{YYYYMMDD}-{HHMM} (UTC window start).
price_to_beat	number	Probora Price Index at the window start, rounded to the asset's tick size.
agent	string	Agent display name.
agent_slug	string	Agent identifier used in paths and filters.
status	string	"published", or "insufficient_data" while the model instance has fewer than 100 resolved observations.
direction	string	Called direction: UP or DOWN.
up_probability	number	Probability that the window settles UP, in [0, 1].
down_probability	number	Probability that the window settles DOWN (1 − up_probability).
confidence	number	Signal-quality score in [0, 1]. Separate from the probability itself.
historical_accuracy	number	Directional accuracy of this model instance (asset × timeframe) over its public history.
sample_size	integer	Resolved public predictions behind historical_accuracy.
model_version	string	Frozen model version that produced the prediction.
prediction_timestamp	string	When the prediction was published: at window start, before settlement.
settlement_timestamp	string	When the window settles.
data_mode	string	"live" for real price sources, "demo" for the simulated network. The two are never mixed.

Errors

Errors use the standard error envelope. Codes this endpoint can return:

Code	Status	Meaning
invalid_asset	400	Invalid asset. The asset symbol is not supported.
invalid_timeframe	400	Invalid timeframe. The timeframe is not one of 5m, 15m, 1h, 4h, 1d.
invalid_agent	400	Invalid agent. The agent slug is not recognized.
unauthorized	401	Unauthorized. The API key is missing, malformed or revoked.
rate_limited	429	Rate limited. Too many requests. Retry after the time in the Retry-After header.
internal_error	500	Internal error. Something went wrong on our side. The request id helps us investigate.

Every model instance (agent × asset × timeframe) learns from market #1, but may only publish once it has resolved at least 100 observations. Until then the endpoint still returns 200 OK, with status: "insufficient_data", the current observation count and null prediction fields:

Response · 200 (insufficient data)

{
  "asset": "HYPE",
  "timeframe": "1d",
  "market_id": "hype-1d-20261114-0000",
  "agent": "Trend",
  "agent_slug": "trend",
  "status": "insufficient_data",
  "observations": 25,
  "required": 100,
  "direction": null,
  "up_probability": null,
  "down_probability": null,
  "confidence": null,
  "model_version": null,
  "data_mode": "live"
}

Training observations are never retroactively converted into public predictions. Public history starts at the first window the instance published before settlement.

Notes

Valid agent values: trend, momentum, mean-reversion, rsi-divergence, order-flow, volatility, liquidation, oi-funding, whale-flow, meta.
confidence is a separate signal-quality score; it is not the probability. A 58% call can carry high confidence when the agent's inputs are clean and complete.
historical_accuracy and sample_size describe this exact model instance. The same agent can be strong on BTC 5m and average on GOLD 1d.
model_version identifies the frozen model that produced the prediction. Versions change only through retraining that passes out-of-sample validation.
On the Free plan predictions are delayed by 15 minutes.

Questions or feedback on the API? [email protected]

Overview#

Parameters#

Request#

Response#

Errors#

Insufficient Data#

Notes#