No look-ahead. Ever.
That's what you're paying for.
Every value is computed exactly as it stood at as_of — nothing from the future bleeds in. Train on the historical archive, deploy live, and the distributions match. The #1 thing quants pay for, guaranteed at the API seam.
166 signals, auto-expanded into 15,606 addressable features.
One base signal × every window × every transform. You address any cell directly: category.signal@window:transform
The top 20 by market cap.
Live today on BTC, ETH, SOL, XRP and BNB — 69 signals serving live data, 92 more shipping. We're rolling out the same leak-free, point-in-time signals to the rest of the top 20 by market cap, with Kalshi range markets landing alongside Polymarket.
Binance 451-geo-blocks its USDT-M futures feed from EU / UK / US / SG IPs (MiCA). We capture it from Tokyo — the full all-market liquidation firehose, 300+ perps. An EU-hosted data feed structurally can't serve this. Powers liquidations.liq_market_cascade + per-asset USDT-M liquidation pressure.
15 categories of edge.
Order book
Depth, imbalance and queue dynamics at the top of book.
Order flow
Trade-driven pressure: OFI, VPIN and aggressor balance.
Liquidations
Forced-deleveraging cascades and liquidation imbalance.
Funding & basis
Perp funding momentum and spot-perp basis structure.
Regime
Hurst, variance ratios and realized-vol regime state.
Positioning
Long/short ratios and crowd-divergence signals.
Options
IV-RV spread, skew and term-structure features.
Price discovery
Kyle's lambda and microstructure price impact.
Cross-market
Lead-lag and cross-venue premium dislocations.
On-chain
on-chainFlows, exchange balances and on-chain pressure.
Prediction markets
eventsYES/NO arb deviation and resolution-window edges.
Price action
Returns, momentum and realized moves across windows.
Volume
Participation, surges and volume-profile features.
Built for how you actually trade.
Trend follower, perp scalper, intraday window trader, event trader — pick your seat. Every feature below is one API call away.
Ride the trend, sit out the chop
For the long-term trend follower: only stay in when the regime persists. Gate the model on regime.hurst@4h + variance_ratio and flat the book the moment the tape turns mean-reverting.
Gate HL perp scalps on order-flow regime
For the perpetual scalper on Hyperliquid: size up only when the flow regime agrees. Read flow.ofi@1m and book_imbalance_top5, then confirm with basis_hl_vs_spot when funding flips against the crowd. (Second-by-second streaming is on the roadmap.)
Trade the BTC M5 / M15 windows
For the intraday window trader: time entries with liquidations.liq_imbalance@15m:zscore and flow.vpin@5m — fade the exhaustion print, ride the impulse that follows.
Arb the prediction-market book
For the event trader: yes_no_arb_dev@5m flags when YES+NO drifts off 1.00, while dist_to_half and context.tte_hours tell you when to press into resolution.
One playbook, end to end.
No abstractions — here is exactly what a subscriber does on day one: catch the post-liquidation-cascade reversal on the BTC hourly UP/DOWN binary. Four steps, every one of them a real call against this API.
A cascade is a forced seller.
When BTC dumps, over-levered longs get liquidated at any price — selling that has nothing to do with new information. When the cascade exhausts, price tends to snap back within the hour. The hourly binary reprices late: right after the flush, “UP this hour” is often cheaper than it should be. That lag is the trade.
Three features, one rule.
GET /v1/features/live?asset=BTC&features= liquidations.liq_imbalance@15m:zscore, flow.ofi@5m:zscore, funding.funding_avg@1h Authorization: Bearer fk_live_… # poll every minute
Backtest it without lying to yourself.
One POST /v1/features/historicalwith your rule’s timestamps returns every signal exactly as it stood at that minute— leak-free by construction. Measure the hit-rate of post-cascade hours against the 50/50 prior. Clears your bar → size it. Doesn’t → you killed a bad idea in an afternoon. Both outcomes are wins.
Run it live.
Set the alert on the trigger, poll the rule every minute with your key, place the order on Polymarket when it fires. Your plan decides how deep the same playbook goes:
You see everything.
You unlock what you need.
The whole grid is visible on every plan — windows and categories above your tier are dimmed, never hidden. Upgrade and the locks lift live, mid-session. No re-keying, no migration.
Raw feeds are public. The seam is the work.
Every venue will hand you a websocket. The distance between that socket and a number your backtest can trust is the part you'd be rebuilding.
Questions a quant will ask.
How do I know it’s really leak-free?
Every read filters ts ≤ as_of over a stored series: a historical value is computed only from data that existed at that timestamp, through the same path that serves live traffic — so backtest and live distributions match by construction, and nothing is revised after the fact. The methodology page documents the mechanism and includes a downloadable point-in-time sample you can audit against public data yourself.
Read the methodology ↗Can I trust the analytics?
Yes — and when you shouldn’t, we tell you. Analytics are computed over real stored series, with outcomes measured against real next-bar mark-price returns. Where a series doesn’t yet have enough real history, the response is explicitly labeled simulated: true and the dashboard badges it. Every analytics response also carries span_days and bars, so you see exactly how much history backs a number before you lean on it.
Why not build this in-house?
You can — the raw feeds are public. What it actually takes: harvesters across spot, perp, options and prediction-market venues running 24/7 (a capture gap is a permanent hole in your history), storage for every series, a point-in-time resolution layer so backtests only see what was known at each timestamp, and mark-price outcome series to evaluate against. That’s an ongoing infrastructure commitment, and every hour it absorbs is an hour not spent on your models. We sell that layer at the API seam so you start at the research step.
What’s the ROI?
We don’t promise returns, and you won’t find a fabricated performance number anywhere on this site. What you’re buying is input quality: point-in-time, leak-free features your backtest can actually trust. Whether that becomes alpha depends on your models — that part is yours. Whether the data quietly lied to you is the part we take off the table.
Am I locked in?
No. Exports are watermarked to your account and remain yours — downgrading or cancelling doesn’t delete them. You can revoke any of your API keys at any time, the trial requires no card, and plan changes apply mid-session without re-keying or migration.
The edge comes from you.
I keep it leak-free.
Spin up a key, pull a live feature in under a minute, and see your signals resolved point-in-time, leak-free.