How many proxies you actually need
This takes two minutes to calculate and is almost always guessed instead. Here is the calculation and the things that ruin it.
Start from the request rate
The platform does not see your concurrency. It counts requests arriving from one address per minute.
Take the total rate: requests divided by the window in minutes. Divide that by the rate one address is allowed. You now have the floor.
A worked example
Forty thousand requests in four hours is 240 minutes and about 167 requests a minute.
At eight requests a minute per address that is 21 addresses. This is the floor and you cannot run on it: the working pool is twice that.
Why the spare half is not optional
Addresses drop out during a normal run: some hit limits, some answer slower than the rest, some catch a check.
A doubled pool holds the schedule even when a quarter of it is gone. Going four times over is waste: idle addresses cost the same as busy ones.
What breaks the maths
Addresses from one subnet: forty in a row read as one. A gigabyte package: it runs out mid-job. A shared pool: the limit is split with strangers.
And the hour of the day. At night the same pool sustains a higher rate, because the total load on the platform is lower.
total rate = 40000 / 240 = 167 requests per minute per address = 8 requests per minute floor = 167 / 8 = 21 addresses working pool = 21 x 2 = 42 addresses