It is difficult for fee estimators to accurately predict the minimum required fee because the time until the next block is always unknown. If a block was just found, the mempool will have fewer high-paying transactions, but it is reasonable to expect that high-paying transactions will appear in the mempool over time until the next block is found. This makes estimating the required fee difficult because you don't know when the next block will be found, and in cases when a lot of mining capacity is being taken on/offline, the fee estimator may not even know how long until the expected time of the next block.
That is the "human behavior" that we cannot predict not the time between blocks. Even if blocks were mined at fixed intervals (like every 60 seconds) we still wouldn't have been able to predict whether in the next seconds there will be a surge of new transactions paying higher fees or not.
This has the potential to cause confusion to some of the fee estimators out there.
Fee estimators should work on size not count and shouldn't try to predict the future.
Basically they should tell you that with current mempool situation when the next block is mined it will clear ~4MB from mempool and will contain down to x sat/vbyte fee and anything lower will remain unconfirmed. That way it won't matter if someone was paying very high fees for 100 transactions.