Survivorship bias
Testing on a universe of currently listed companies excludes everything that went bankrupt or was delisted. The strategy is then evaluated only on names that survived, which no strategy could have known in advance. This alone can turn a losing system into a winning chart.
Look-ahead bias
Using information at a timestamp before it was actually available — restated fundamentals, a closing price to make a decision at that close, an earnings figure on the date it applied to rather than the date it was published. Each is easy to introduce and invisible in the results.
The defence
Point-in-time data, a strict rule that every decision uses only information timestamped before it, and walk-forward validation on data the model has never touched. This is unglamorous engineering and it is the difference between a research result and a tradeable one.
Why it belongs on this page
Every strategy described across these five markets can be made to look excellent by either bug. The most valuable automation any operation builds is the harness that makes these two errors impossible, rather than the strategy it validates.