Useful research
shows its assumptions.
MethodologyUnderstand what a model result represents, how to inspect its supporting record, and which limitations matter before using an allocation.
From inputs to allocations
01Define the work
A model starts with a natural-language request or supplied Python. Declared inputs, a supported universe and the model runtime contract determine what can execute. Review generated logic as carefully as code you write yourself.
02Inspect the data
Available price histories and metadata vary by asset. Examine coverage dates and missing fields before drawing conclusions. Data updates and corrections can change later runs; a stored data-as-of date does not imply a frozen copy of every underlying input.
03Identify synthetic history
Backfills extend observed histories using a source and an explicit transformation. Their assumptions can materially affect a model’s results. Distinguish backfilled periods from actual observations and examine the splice boundary.
04Execute within constraints
The platform validates the script and executes it in an isolated runtime with resource and time limits. A successful build demonstrates execution under those inputs; it does not establish that the strategy is sound or suitable for an investor.
05Read the result in context
Model backtests produce hypothetical return series, positions and metrics. Inspect the script’s signal timing, allocation rules and execution assumptions. Gross history and any requested fee overlay are different views; do not assume costs, taxes or market impact were measured unless the output explicitly models them.
06Follow changes
Builds preserve scripts and outcomes, and runs retain inputs, results and script references. Compare like-for-like windows and assumptions. Changed data, optimizer code or model dependencies may alter results even when the top-level script is unchanged.
07Check freshness before use
Scheduled refreshes operate on supported trading days when a model is ready and account limits allow execution. Inspect the latest successful run and its data date. The schedule itself is not evidence that an allocation is current.
08Keep execution explicit
Allocation downloads and broker-format files support your next step. They neither place orders nor confirm broker eligibility. Verify the intended weights, account restrictions and file format before taking action at your broker.
Research, with appropriate limits
Backtests are illustrative. Past performance does not guarantee future results. ENSEMBLE is a software platform, not an investment adviser.
Hypothetical results benefit from hindsight and do not reflect actual capital at risk. Liquidity, financing, taxes, transaction costs and market impact can differ from modeled assumptions. No account is assured results like those of a backtest.