Two bad days are not enough data to change the plan

[Amanhasnoname]

Classic pattern: two losing sessions in a row, then a rewrite. A filter gets added, the stop gets moved, a new indicator joins the chart. Three days later the system looks completely different and none of the changes has any real sample behind it.

A rough sense of how noisy this is: a system with a 55 percent win rate and even payoffs will, over a long run, regularly produce streaks of five or six losers in a row. Two bad days of ten trades each hardly distinguishes between "edge gone" and "normal variance". Making changes at that point means fitting the plan to randomness.

A better routine is to pre-define what evidence would justify a change. For example: at least 100 trades under unchanged rules, with expectancy below a stated threshold, or a clearly identified execution error pattern. Anything below that goes into a "ideas" list, not into the live plan.

When a change idea comes up after a bad day, write it down with the date and test it on historical or demo data first. Look at the list a month later; most of the entries will look less brilliant.

What is your minimum sample size before touching the rules, and has a premature change ever bitten you?