Data & market intelligence
Market data for an AI strategy factory.
Give AI-generated strategies a traceable foundation: market prices, supplementary series and timestamped news, with data checks before the backtest.
Interactive illustration
See the gaps before testing the idea.
Move across a sample price series to reveal missing observations and a duplicate timestamp. Data checks make these issues visible before an AI strategy is evaluated.
Keep missing data visible.
A line across a gap would imply evidence that is not there. The example leaves gaps open and counts a duplicate separately; a repeated row is not an additional observation.
Illustrative price index
Expected bar checkpoint
- Unique observations
- Missing bars
- Duplicate rows
Count unique observations against the expected timeline.
Missing bars remain gaps until source data can be verified.
A duplicate timestamp does not repair an earlier gap.
Nine venues, each with its real rules
The engine charges each venue's actual maker/taker fees, accrues funding on the mark price, enforces margin and liquidation for leverage, and respects session calendars — a strategy that only works at zero cost never survives here.
| Market | Venue | Maker | Taker | Max leverage | History |
|---|---|---|---|---|---|
| Perpetual futures | Binance USD-M · Bybit · Hyperliquid | 0.015–0.020% | 0.045–0.055% | 20× | multi-year, 6 timeframes |
| Spot | Binance | 0.100% | 0.100% | 1× (cash, long-only) | multi-year |
| Inverse futures | Binance COIN-M | 0.010% | 0.050% | 20× | multi-year |
| Crypto options | Deribit · Binance | 0.030% | 0.030% | 1× | full chains incl. expired contracts |
| US equities | 560-symbol S&P 500 + ETF universe | $0 | $0 | 2× (Reg-T) | years of 1m/1h/1d on the XNYS calendar |
| Prediction markets | Polymarket | $0 | $0 | 1× | forward-only, 30k+ resolved markets |
Fee schedules are hand-verified against official exchange documentation and applied at backtest time. Optional square-root market impact and per-bar mark-to-market equity (unrealized PnL included) keep the accounting honest at every step.
Twenty-six series beyond price
Strategies subscribe to supplementary data the way a desk would — the loader injects only what the strategy actually reads, point-in-time aligned to its bars.
News, read at machine speed
A dedicated collector ingests tier-1 wires, primary government sources, the crypto press and a real-time fast feed — then an AI model scores every headline into per-asset sentiment indicators bucketed at five-minute resolution. Everything is timestamped by when we saw it, so a backtest can never read tomorrow's news.
Tier-1 wires
Reuters, Bloomberg, The Wall Street Journal, CNBC, MarketWatch and Yahoo Finance — the same headlines a terminal shows, minutes not hours.
Primary sources
SEC 8-K filings as they land, Treasury, White House, BLS and BEA data releases, the Fed, ECB, Bank of England and Bank of Japan, CFTC/DOJ/FTC actions, EIA energy reports and Nasdaq trade halts — the origin of the news, not the retelling.
Real-time fast feed
A persistent websocket lands breaking headlines — including exchange listing announcements — in the data plane within seconds of publication.
Scored, not just stored
Each article gets per-asset relevance, direction and surprise scores across 16 asset buckets plus any tickers it names. Sentiment becomes a strategy input, chartable live and testable historically.
Datasets nobody can buy
Some market history simply isn't for sale — so we record it ourselves, every day, and time works in our favor.
Prediction-market history
Polymarket markets are short-lived and vanish when they resolve. Our collector captures every market's full life — 30,000+ resolved markets so far — turning them into backtestable instruments no vendor offers.
US options surfaces
Free options feeds show only the present. We snapshot full chains — bid/ask, implied volatility, open interest and greeks for 18 major underlyings including SPX, SPY, QQQ and VIX — every 30 minutes of every session, building a history that is otherwise a paid license.
Live liquidation tape
Binance discontinued historical liquidation archives; our always-on stream has been recording every forced order since — a dataset that only exists if you were listening.
Point-in-time honesty
Forward-collected series carry ingestion timestamps, so backtests only ever see what was actually knowable at the time. No survivorship patching, no backfilled hindsight.
Audited continuously, not assumed
A standalone auditor daemon re-reads every catalog series on a loop, records gaps and duplicates, refills what the source still offers, and blocks backtests from running on unaudited or gap-ridden data. When data is missing at backtest time, the engine fetches it from the exchange on demand — and refuses to substitute a different symbol or timeframe.
See it live, no account needed
The market pages run on this exact data — cryptocurrency, US stocks, options chains, prediction markets and the news feed are all public.