Why the same setup behaves differently in different regimes
Nothing broke. The conditions the pattern depended on stopped being present, and a chart cannot tell you when that happened.
A regime is a stretch during which market conditions are broadly consistent — trending or ranging, volatile or quiet, driven by one thing or another. Regimes change, usually without announcement.
This is the ordinary explanation for something that gets treated as mysterious: a setup that read well for months stops working, and nothing about the setup changed.
The dimensions that shift
Trending against ranging. Any rule built to detect persistence misfires when there is no persistence to detect. A moving-average crossover in a sideways market produces repeated reversals, each immediately undone. No setting fixes it, because the problem is the condition. See trend indicators.
High against low volatility. A fixed distance means different things at different ATR levels. Measures expressed in ATR terms adapt; measures expressed in points do not, and quietly become tighter or looser than intended.
What is driving the market. A period where everything moves with rate expectations behaves differently from one where individual names move on their own news. Levels matter more in the second and less in the first.
Who is participating. Holiday sessions, month-end, index rebalances, expiry weeks. Thinner or differently motivated participation changes how levels hold.
The structural changes too
Some shifts are mechanical rather than behavioural. A futures rollover changes the contract you are charting. A stock split changes every historical price. A change in tick size or trading hours changes the distribution of bar ranges outright.
These are worth checking first when something stops behaving, because they are verifiable rather than interpretive.
The trap
Do not conclude a method stopped working from a small sample. Runs of losses occur in anything with a win rate below certain, and they occur more often and last longer than intuition suggests. Ten poor trades is entirely consistent with nothing having changed.
The opposite error costs just as much: adjusting settings after every rough patch produces a method fitted to the most recent noise, and each adjustment feels justified at the time. That is curve fitting arrived at slowly.
What to do instead
Describe conditions in the same terms every day, using measurements rather than impressions — current ATR against its own recent range, whether the trend definition currently holds, whether the session is ordinary. Record it in your journal before the trades, not after.
Then, when you look back over fifty trades, you can ask whether the poor ones cluster in a particular condition. That question is answerable from records. It is not answerable from memory.