A replay, not a time machine
Imagine pausing the market on a past date, making a decision using only the information available then, and moving forward to see what happened. That is the basic idea of a backtest. The rules, universe, prices and timing need to be clear enough that somebody else could repeat the same replay.
Historical testing is valuable because waiting twenty years to examine every long-term idea is not practical. It lets researchers find obvious weaknesses, compare approaches and form better questions. It does not recreate the pressure, costs or unknowns of a real decision perfectly.
The clock matters
Using information that arrived later makes an earlier decision look smarter than it could really have been. A company may revise a figure, an index may remove failed members, or a closing price may not have been available when the supposed trade occurred. This is often called look-ahead bias.
A defensible test fixes observation cut-offs, market sessions and price availability. If earnings were released after the market closed, a result measured at that same close cannot pretend the news was already public. Time boundaries are part of the method, not a technical footnote.
The research choices can overfit too
Trying many variations and reporting only the strongest result can turn ordinary chance into an impressive chart. The problem is not experimentation itself. The problem is forgetting how wide the search was and acting as if the winner was the only idea ever tested.
Tradour keeps exploratory work separate from later tests. An observation recorded before its outcome is known faces a cleaner challenge than a pattern discovered after inspecting the whole archive. The two records should not be blended into one flattering success rate.