Agents

Agent Performance

Accuracy, probability quality, calibration and confidence buckets, plus the leaderboard endpoint.

Agent performance

GET/v1/agents/{agent}/performance

Developer PreviewAll plansapi.probora.xyz

Accuracy, Brier skill, calibration and confidence buckets for any scope.

Parameters

Parameter	In	Type	Description
agentrequired	path	string	Agent slug: trend, momentum, mean-reversion, rsi-divergence, order-flow, volatility, liquidation, oi-funding, whale-flow or meta.Example: `order-flow`
asset	query	string	Restrict to one asset.Example: `BTC`
timeframe	query	string	Restrict to one timeframe.Example: `15m`

Request

# export PROBORA_API_KEY=prb_live_…
curl "https://api.probora.xyz/v1/agents/order-flow/performance?asset=BTC&timeframe=15m" \
  -H "Authorization: Bearer $PROBORA_API_KEY"

Response

200 OK, returned as application/json. Example:

Response · 200

{
  "agent_slug": "order-flow",
  "scope": {
    "asset": "BTC",
    "timeframe": "15m"
  },
  "markets": 48291,
  "wins": 30810,
  "losses": 17481,
  "pushes": 12,
  "accuracy": 0.638,
  "accuracy_ci95": [
    0.6337,
    0.6423
  ],
  "brier": 0.2216,
  "brier_skill": 0.1136,
  "log_loss": 0.6342,
  "calibration_score": 91.4,
  "confidence_buckets": [
    {
      "bucket": "65-70",
      "markets": 1423,
      "accuracy": 0.681,
      "mean_probability": 0.674
    },
    {
      "bucket": "75+",
      "markets": 508,
      "accuracy": 0.79,
      "mean_probability": 0.783
    }
  ],
  "data_mode": "live"
}

Response fields

Field	Type	Description
agent_slug	string	Agent identifier used in paths and filters.
scope	object	Asset / timeframe filters applied to the statistics.
markets	integer	Scored predictions (wins + losses). FLAT and VOID windows are excluded.
wins	integer	Predictions that matched the outcome.
losses	integer	Predictions that missed the outcome.
pushes	integer	Predictions on FLAT windows, recorded but never scored.
accuracy	number	wins / (wins + losses).
accuracy_ci95	array	95% Wilson confidence interval for accuracy.
brier	number	Mean squared error of the called-direction probability. 0.25 equals a 50/50 forecast.
brier_skill	number	1 − brier / 0.25: improvement over an uninformed 50/50 forecast.
log_loss	number	Mean negative log-likelihood of the realized outcomes. ln 2 ≈ 0.693 equals 50/50.
calibration_score	number	100 · max(0, 1 − ECE / 0.10): 100 means perfectly calibrated.
confidence_buckets	array	Accuracy by called probability: 50–55%, 55–60%, 60–65%, 65–70%, 70–75%, 75%+.
data_mode	string	"live" for real price sources, "demo" for the simulated network. The two are never mixed.

Leaderboard

GET/v1/leaderboard

Developer PreviewAll plansapi.probora.xyz

Ranked agents for overall, asset, timeframe or asset × timeframe scopes.

Parameters

Parameter	In	Type	Description
asset	query	string	Asset scope.Example: `BTC`
timeframe	query	string	Timeframe scope.Example: `15m`

Request

# export PROBORA_API_KEY=prb_live_…
curl "https://api.probora.xyz/v1/leaderboard?asset=BTC&timeframe=15m" \
  -H "Authorization: Bearer $PROBORA_API_KEY"

Response

200 OK, returned as application/json. Example:

Response · 200

{
  "scope": {
    "asset": "BTC",
    "timeframe": "15m"
  },
  "data": [
    {
      "rank": 1,
      "agent_slug": "meta",
      "markets": 48291,
      "accuracy": 0.652,
      "brier_skill": 0.121,
      "calibration_score": 95.2,
      "overall": 88.4
    },
    {
      "rank": 2,
      "agent_slug": "order-flow",
      "markets": 48291,
      "accuracy": 0.638,
      "brier_skill": 0.114,
      "calibration_score": 91.4,
      "overall": 84.9
    }
  ],
  "updated_at": "2026-11-14T20:15:04.000Z",
  "data_mode": "live"
}

Response fields

Field	Type	Description
scope	object	Asset / timeframe filters applied to the statistics.
data	array	Page of results, newest first.
updated_at	string	When the statistics were last recomputed.
data_mode	string	"live" for real price sources, "demo" for the simulated network. The two are never mixed.

Fields of each item in data

Field	Type	Description
rank	integer	Position on the leaderboard for the requested scope.
agent_slug	string	Agent identifier used in paths and filters.
markets	integer	Scored predictions (wins + losses). FLAT and VOID windows are excluded.
accuracy	number	wins / (wins + losses).
brier_skill	number	1 − brier / 0.25: improvement over an uninformed 50/50 forecast.
calibration_score	number	100 · max(0, 1 − ECE / 0.10): 100 means perfectly calibrated.
overall	number	Composite score blending accuracy, probability quality and calibration (0–100).

Metric definitions

Field	Definition
accuracy	wins / (wins + losses). FLAT (push) and VOID windows are excluded.
accuracy_ci95	95% Wilson score interval: how much the accuracy could move with more samples.
brier	Mean of (p − y)² for the called-direction probability p. Lower is better; 0.25 is a 50/50 forecast.
brier_skill	1 − brier / 0.25. Positive means better probabilities than a coin flip; shown as “Prob. Score”.
log_loss	Mean −ln(p) for wins and −ln(1 − p) for losses. ln 2 ≈ 0.693 is the 50/50 baseline.
calibration_score	100 · max(0, 1 − ECE / 0.10), where ECE is the sample-weighted gap between accuracy and mean probability per bucket.
overall	Leaderboard score: reliability · (0.40 · accuracy score + 0.35 · probability score + 0.25 · calibration), with reliability = n / (n + 250) so small samples cannot top a leaderboard.

confidence_buckets group predictions by the probability of the called direction: 50–55%, 55–60%, 60–65%, 65–70%, 70–75%, 75%+. A well-calibrated agent's 65–70% bucket should win roughly 65–70% of the time. Definitions match the public methodology and the leaderboard.

Questions or feedback on the API? [email protected]