To sum up there are two different types of pool performances:
Performance of mining and availability
Performance (luck) of minted blocks in relation to the expected number of blocks
1. Performance of mining and availability:
If a pool is offline at the moment it should mine a block, the block will be counted as a missed one. Therefore the overall performance would be bad. From this we can conclude that the main task of a SPO (stake pool operator) is to make sure that the pool is online all the time. This is the only way to ensure that each block assigned is being processed.
The biggest challenges in order to stay online with the nodes are during update processes or due to limited resources (like too little available memory space for example). Besides, missed blocks can be caused by wrong configuration settings.
How is P₳C Stakepool able to handle this?
First of all, the hardware is scalabe during runtime. Shutdowns/restarts are not necessary. Nevertheless, P₳C Stakepool has different redundant nodes in the background as a safeguard. All nodes are updated one after the other, so that at least one node is always available to maintain mining. All things considered P₳C Stakepool is able to reach the full availability (24/7).
2. Performance (luck) of minted blocks in relation to the expected number of blocks:
The Cardano protocol assigns blocks to a pool in proportion to its stake-size. It is important to be aware that this is completely random. The assigned blocks merely reflect a trend in the expected number of blocks per epoch. But depending on the probability, the number of blocks reached in an epoch can drift up and down.
This is not the responsibility of the SPO! The only thing he can ensure is to make the blocks when they are assigned to his pool!
Due to the great work of Marcel Baumberg [TITAN], Andrew Westberg [BCSH] and Papacarp [LOVE], it is now possible to estimate how many blocks can be mint in an epoch. You can find this on my homepage under Rewards--> calculator
The graph below illustrates that, depending on the number of epochs (samples), this behaviour tends towards a normal distribution. From this we can conclude that sooner or later epochs of bad luck are compensated by epochs of good luck (and vice versa):