What they do
The adaptive threshold raises the bar when recent signals turned out badly and lowers it when things go well. Dynamic weights do the same for individual factors: a factor that has been wrong lately starts voting more quietly.
The mechanics are ported verbatim from the working bot, limits included: the threshold stays within [2, 5], the step is one point, the window is the last 50 signals.
Why they do not apply in a backtest
Both settings adjust from the results of already closed trades. Inside a historical run, "already closed" means trades that would not yet have closed at the moment of the decision. Applying them in a run gives the strategy knowledge from the future, and the result comes out better than is possible in life.
This limitation is fundamental, not temporary. That is why the interface shows a "live signals only" mark next to these toggles — and shows it at the moment you enable them, not after twenty minutes of computation.
When to enable
Once you have checked the strategy on history and are moving to live signals. Adaptation makes sense on a stream: it accumulates statistics from your own signals and adjusts to the current market regime.
What not to confuse
Adaptation does not turn a bad strategy into a good one. It softens bad streaks, and that is all.