Take a comparable sample
Write down the platform and version label, owned upgrades, food and collection behavior. Start a timer with your starting balance. Record each purchase during the interval, then note the ending balance and elapsed time in minutes. Enter zero spending only when no purchases occurred.
For the second run, keep those conditions comparable and change one thing. Collect in the same way at both interval boundaries. A single interval can be noisy; repeat it before attributing a difference to a purchase. Never mix balances in different units.
Worked arithmetic example
These are invented worksheet inputs to explain the arithmetic, not measured game results: starting balance 100, ending balance 160, spending 20 and duration 2 minutes give a net change of 30 units/minute and receipts of 40 units/minute. Spending is added back to reconstruct receipts; subtracting it again would count the cost twice.
A second interval from 100 to 140 with zero spending over 2 minutes gives 20 units/minute. The comparison is −10 units/minute. This says what happened to the entered balance; it does not prove an upgrade is worse.
When the result is unsuitable
A reset, refund, transfer, one-off reward, unrecorded purchase or change of balance units breaks the intended comparison. Restart the observation or keep those events in a separate record. The worksheet does not know game prices, skill effects, offline earnings or a best reset threshold.
Negative net change is possible when spending exceeds receipts. Negative reconstructed receipts are inconsistent with this worksheet. Missing fields, nonpositive time, negative inputs and calculations beyond the supported numeric range show an input explanation instead of a result. Inputs are not saved across reloads.