Calculate the deviation
If a category is expected 20 times in a 100-round model and appears 16 times, its relative count deviation is (16 − 20) / 20 = −20%. The definition of the percentage must be stated next to a chart.
Read the sample size first
The same percentage deviation may come from two hits versus an expectation of 2.5, or 200 hits versus 250. Those observations have different amounts of data behind them.
Cross-check chart windows
Comparing 1-hour, 6-hour and 24-hour samples can reveal whether an observation is local to one period. These windows overlap, so they are not independent confirmation of a trend.
Working with the historical data
For a useful comparison, specify the recording date, timezone, game table and observation window. A historical count without those details can easily be interpreted outside the circumstances in which it was measured. Distinguish theoretical probabilities described in game rules from estimated frequencies measured in a particular data source.
Record-based tools are most useful when they make observations reproducible. If a chart cannot be reconciled with the underlying event history, it needs a data-quality review before anyone treats its values as authoritative.
Frequently asked question
Does cold mean due?
No. Cold describes recorded frequency below the selected benchmark in the stated window.