ravine-api

free 10k/day no signup AUC 0.77

One POST endpoint. Predict whether a Solana pump.fun memecoin is about to rupture (drop ≥10% in next 5 ticks).

Trained on 15,724 real trajectories captured from live pump.fun activity. Returns calibrated probability + decision tier.

Try the endpoint →

Quickstart

curl -X POST https://YOUR_HOST/v1/predict \
  -H "Content-Type: application/json" \
  -d '{
    "history": [
      {"ts":1700000000,"mcap":5000,"buy_count":8,"sell_count":1,"buy_sol":4.0,"sell_sol":0.4,"bond":0.05},
      {"ts":1700000010,"mcap":5500,"buy_count":7,"sell_count":2,"buy_sol":3.5,"sell_sol":0.5,"bond":0.06},
      {"ts":1700000020,"mcap":6200,"buy_count":6,"sell_count":4,"buy_sol":2.8,"sell_sol":1.2,"bond":0.07},
      {"ts":1700000030,"mcap":6400,"buy_count":4,"sell_count":6,"buy_sol":1.5,"sell_sol":2.5,"bond":0.07}
    ]
  }'

Response:

{
  "p_rupture": 0.92,
  "decision": "QUANTUM_EXIT_NOW",
  "model_loaded": true,
  "quota_remaining": 9999
}

Why?

Helius, Birdeye, Moralis give you raw Solana data. They do not tell you what to do with it. ravine-api gives a calibrated rupture probability — exit-or-hold decisions for memecoin trading bots.

Decision	p_rupture	Meaning
`HOLD`	< 0.30	No rupture signal
`QUANTUM_CAUTION`	0.30–0.60	Consider partial exit
`QUANTUM_EXIT_NOW`	≥ 0.60	Exit recommended

Model

Logistic regression on 11 microstructure features (bonding, buy pressure decay, sell damage, callout divergence, mcap velocity, flicker count, etc.)
Trained on 15,724 (history-snapshot, future-rupture) pairs from real pump.fun tokens
Test ROC-AUC: 0.77
Backtest: +4.35pt edge vs hold-to-end on 1,500 replayed trajectories

Limits

Built for pump.fun specifically — does not generalize to other DEX
Probabilities are rank-reliable but Brier-imperfect
Model snapshot from early 2026; refresh periodically as memecoin meta evolves

Companion endpoint: /v1/score — the Phantom Score

Think PageRank, but for pump.fun memecoins. GET /v1/score?mint=X returns a 0-100 calibrated score representing P(homerun) based on the token's T+0/T+5 features. Includes per-feature contribution so you can see why a token is rated high or low.

curl "https://YOUR_HOST/v1/score?mint=ABC123pump" -H "X-API-Key: YOUR_KEY"

Tiers: COLD < 5 · WATCH 5-14 · WARM 15-29 · HOT 30-49 · FIRE 50+. The model was trained on 35K labeled trajectories — AUC 0.66 base tier (Pro tier with alpha features reaches 0.74+).

Companion endpoint: /v1/trajectories

Labeled lifecycle dataset of recent pump.fun memecoins. For each mint we capture features at T+0, T+5, T+15, T+30, T+1h, T+24h and tag the outcome (homerun +50%+, win +5–50%, loss <+5%).

curl https://YOUR_HOST/v1/trajectories?label=homerun \
  -H "X-API-Key: YOUR_KEY"

Free tier: 24h window, 50 newest labeled mints, public features only (bonding, replies, mcap, ROI, label). Alpha features (deployer reputation, top holder concentration, alpha-wallet count) are Pro-tier.

API spec

Swagger UI · OpenAPI 3.1 JSON · health

Code

Source on GitHub. MIT license.

If this saves you money, send some SOL: YOUR_WALLET_HERE · 0.001 SOL is plenty.