Electronic – Charging Li-Ion string with cells of different capacities

I was trying to understand what is the worst case scenario of using a string of 4 lithium ion cells (18650 cells, 3.6V nom) in series to form a battery pack, without balance circuitry, to understand when it is really needed, as long as with single cell monitor. I know that normally it is used a BMS, but I was thinking what are the conditions that are setting the threshold from a safe to unsafe system without a BMS (for over/under voltage and balance correciton).
During this process I actually learn a lot, or at least I think that. From what I've understood, there are some sources of unbalancing, like:

1. SOC unbalance, related to manufacturing imperfections, leading to a great self discharge and thus unbalance; but could be also related to anything else which brings this SOC variation, within cells of same impedance and capacity. This SOC variation seems to be the most famous and most documented.
   • From a pure theoretical point of view, I was thinking that with very light stress on the battery (like few % in DOD), with some safe limits and high quality cells from the same batch, seems to be not a big issue even after many hundred (light) cycles. Because I read also that this unbalance could be magnified with deep DOD and high currents.
2. Impedance mismatching: this is the most harmful when using balancers. Without balancing, the issue is not that big, if not considering under/over voltage protections during discharge/charge.

3. Capacity difference (which on its turn lead to a SOC variation): this is the most tough to find open information about that. The only paper/article which addresses the issue directly is the one from Yevgen Barsukov, but also here, is just estimating how much really impacts with respect to the pure SOC mismatches:

   It can be that a cells total chemical capacity, Q MAX , was different
   to start with. But even if all cells were discharged by an equal
   amount from a fully charged state, their chemical state of charge will be different. I ndeed, if all 3 cells are discharged by
   100 mAh, but cell 3 has different total capacity (eg: 2000 mAh
   instead of 2200 mAh), the resulting chemical states of charge will be 95.4 and 95%. This in turn will also cause
   different OCVs. As can be seen, 200 mAh difference in Q MAX
   causes only 0.4% difference in SOC. Because SOC correlates
   with voltage, this indicates that capacity imbalance causes less voltage difference than charge unbalance
   (cause 1).

   • This is not very clear to me, I was thinking that the capacity mismatch, if the battery is lightly used and kept to OCV maximum of 4.0V, it would keep both the strongest and the weaker within safe limits. In other words, I was thinking on this graph below, where a over/under voltage detection is used:
     
     Now, what I am writing may seems difficult, but I try to be clear. Translating this to a system without BMS and think what happens if I reduce the maximum SOC 100% to a maximum SOC 80% and with, for example, a light DOD of 10%. But ALSO cut the charge ONLY when the STRING gets the maximum float voltage related to a SOC of 80% (so a bit less than 4.2V) while never going lower than 70% SOC during discharge. In this case, is it correct to say that a 20% (100%-80%) of capacity mismatch is allowed? Explaining further, if the cells have the same SOC, when discharged of a given energy taken from the overall pack, the weakest cell will be 20% lower in capacity. On the other way, when charging, we may end up with the original situation, because we are putting back the same amount of charge, cutting the charge at the same amount of float voltage. But if we instead are charging from a middle capacity up to their maximum of 80%, the weakest (of 20% in this example) will reach now the 100% of charge. Thus why I though I can assume with this boundaries a 20% of margin allowed, in these use cases. So, the ideal situation is to prepare the batteries to have 80% SOC and use them from this moment.

The hypothetical graph will then be similar to the one shown, but now with the same amount of energy extracted from each cell, thus not reducing the delivered power (which is 10% only of the battery pack). Basically, let the ups and downs go above 80% and below 70%, but letting me to stay within 0% and 100%. In this case, if the assumption is correct, and according to the answers, the problems relates to how fast unbalance sums up over time and cycles.