Correct. And wk057 has said that in less than a year the module with the bad cell would get so far out of balance that the BMS would give up. But even more than that one long trip with a lot of Supercharging would cycle the module so many times that the BMS couldn't keep up in trying to balance the pack and cause it to give up. So in reality it could take as little as one or two days of driving with a disconnected cell before the pack fails.
That is for the 18650 based packs. Obviously it would be much worse for the 2170, or 4680, based packs.
My viewpoint differs from wk057 in terms of the pack balancing load. Note: this it theoretical, not necessarily how Tesla does it.
There is no requirement for all parallel groups (bricks) to stay at the same SOC all the time. The only two critical points are the max and min voltage limits. If the BMS targets all bricks to reach max voltage/ 100% SOC simultaneously, that maximizes pack capacity.
If they all reach min voltage at the same time instead, that is less total total energy due to lower average voltage, but still functional. As long as the undersized brick hits the high and low limit first (meaning all its energy is available), the pack will approximate the reduced brick's capacity.
If the bricks are still fairly close in self discharge, minimal balancing is needed, even though the spread may be 100%-100% when full and 22%-11% when drained. It just requires smarter BMS logic (or dumber since it can only focus on highest voltage, rather than full range).
Packs ultimately fail when the leakage differential of any two bricks is more that the BMS balance capacity. In this situation, drift will increase until one brick is critically overdischarged.
Recovery may take a looong time though. Following a cell short with brick now at 80% capacity.
80%-20% -> inital
100%-45% -> charge stopped on nominal brick limit
45%SOC*80%capacity=36% pack effective energy
If differential leakage is acceptable, it will bleed the nominal bricks down:
100%-100% charged
20%-0% discharged