Measure against expectancy, not against last month

[Amanhasnoname]

Comparing this month's results with last month's is natural and misleading. Results fluctuate randomly, and a good month raises expectations that the next month is likely to disappoint.

A more useful benchmark is the system's expected performance, based on the longer history and the cost model. Define a range: the expected monthly result plus or minus a multiple of the standard deviation. If the actual result falls inside, the system is behaving as designed.

A month below the average but inside the range is not a failure. A month above it is not a triumph. Both are samples from a distribution.

This framing reduces emotional swings and prevents hasty changes. It also provides a clear signal when something is genuinely off: a result far outside the range, or a persistent trend of below-average months.

Update the expected values periodically using live data, since the original estimates came from backtests.

Process metrics, such as rule adherence and cost control, deserve equal attention, since they are controllable and the results are not.

What range of monthly results does the system's history suggest as normal?