Methodology you can inspect
Gilito should never ask for trust it has not earned. This page describes the current signal contract and its limits.
Current version: signal-v1.0
Introduced in the focused private beta. Method changes require a new version and changelog entry.
One canonical signal
A signal is classified on the server from the highest-ranked completed strategy for an asset. A buy requires both a buy state and the minimum strategy-quality threshold; a sell follows a sell state; all other cases are hold. Clients render this result and do not invent their own classification.
Confidence is not probability
Confidence is a bounded relative research-quality score. It starts with strategy quality, is adjusted for the current market regime and walk-forward validation, and is reduced when research is stale. It is not the probability that an asset will rise or that a trade will be profitable.
Evidence and freshness
Each insight includes the calculation timestamp, backtest start and end dates, trade sample size, market regime, strategy rules, validation status, methodology version, and staleness warning. End-of-day provider data may be delayed.
Costs and benchmarks
The insight summary explicitly states whether commissions and slippage were included. When they are not, both are shown as zero and the limitation is called out. Full strategy detail carries the buy-and-hold benchmark and out-of-sample results where available.
Known limitations
Historical research can overfit and market relationships change. Taxes, liquidity, spreads, commissions, slippage, corporate actions, and data errors may alter real outcomes. Gilito is research software, not personalized investment advice or trade execution.