Methodology
How ENSEMBLE constructs, executes, and measures strategies, and where the limits are.
Pipeline
Strategy construction
Your description is translated into an executable strategy script by a frontier language model working against a curated quantitative-finance context. The code is stored verbatim with the model: what ran is inspectable, it never changes unless you change it, and it downloads as a standalone script, so any assumption can be checked or handed to a reviewer.
Isolated execution
Every backtest runs in its own sandboxed container with CPU, memory, and time limits, isolated from other users and from ENSEMBLE's infrastructure. Generated code cannot reach anything beyond the market data it is given.
Market data
Daily historical price series for US equities, ETFs, and major crypto assets: more than 50,000 instruments, with decades of history for the longest-listed. Data refreshes daily so re-runs pick up the newest sessions.
Deterministic re-runs
A re-run executes the exact stored script against fresh data. No regeneration, no drifting logic: the same rules extended through the latest close, on demand from any surface.
Metrics
CAGR, volatility, Sharpe, maximum drawdown, and trade counts are computed from the model's daily return series using standard definitions: annualization at the calendar the series trades on (252 exchange sessions, 365 for 24/7 markets), sample volatility, geometric compounding. The series is exportable, so every figure can be reproduced.
Cost assumptions
Returns are net of simulated execution costs. Simulations assume end-of-day fills at close prices and do not model intraday slippage, market impact, financing, or taxes. Results are a research tool, not a forecast of live performance.
At a glance
Disclosures
Backtests are illustrative. Past performance does not guarantee future results. ENSEMBLE is a software platform, not an investment adviser.
Hypothetical performance has inherent limitations: it is prepared with the benefit of hindsight, does not reflect actual capital at risk, and may under- or over-compensate for market factors such as liquidity. No representation is made that any account will achieve results similar to those shown.
Questions: support@ensemble.markets