AI output limitations
AI-generated content may be incomplete, misleading, stale, overconfident, or simply wrong.
Generated formulas, scan drafts, explanations, and summaries must be reviewed by the user before use.
Backtest limitations
Backtests, simulations, hypothetical rankings, and model outputs are hypothetical and have material limitations.
They may omit or simplify slippage, liquidity constraints, borrow availability, market impact, execution quality, halts, session behavior, partial fills, stale data, and operational failure modes.
Survivorship bias in historical results
Current-membership scan universes are built from currently listed symbols. Companies that were delisted, acquired, merged, or went to zero are not members, so they do not appear in historical scans or backtests that run against those universe sources.
Historical point-in-time scans using those sources therefore use present-day membership rather than the membership that existed on the date being tested. This biases results upward, and the effect grows with the lookback. Treat returns, hit rates, and drawdowns from current-membership universe tests as optimistic.
A chart-attached backtest is constrained to one explicit symbol and does not mechanically omit other securities from that calculation. Selecting a present-day survivor with hindsight can still make the strategy evidence look too optimistic, but that is hindsight selection rather than a multi-symbol universe excluding failed companies.
No performance promise
A backtest or AI explanation does not represent actual trading and should not be treated as a promise, forecast, or expected outcome.
Live performance can differ materially from historical or simulated results.
Best use
- Use AI and backtests to generate ideas, test assumptions, and accelerate workflow.
- Do not rely on them as a substitute for independent review, risk controls, and execution discipline.