Four down to three isn't much of a reduction if four is 'way too much'. How about two? Would leave you spare kit to carry out swarm management, aka the AS.
there was a lot of swarming
At your stage of the game, seven-day checks should alert you to swarm preps and allow time to manage the event, so aim to stick to that schedule and have the kit ready.
I had no real structure for my first three years and was reactive to colony development rather than proactive through observation, didn't have enough kit and hadn't embedded the swarm cycle into my head. Winter is a good time to start a plan.
First for the chop, by the JBM method above. There is no benefit in keeping her if you already know she produces a defensive colony that makes your work uncomfortable and spreads her genes far and wide. Once you've killed her, unite the lot to the placid colony immediately.
Now you're down to two: the big united colony is almost certain to want to swarm or supersede to replace the third-year queen. Summer supersedure means swarming, so don't leave them to it if you see 3 or fewer QCs.
When you see QCs, make an AS and leave one good QC in the split box; make a couple of nucs with sealed QCs as reserve; check all colonies thoroughly six days later and remove any EQCs.
When the splits have mated and are laying well, remove the old colony queen and unite a nuc. Same with the third colony, which you haven't described. This will leave you with 2023 queens heading two colonies by midsummer, plus an insurance nuc. Sell it then if you wish, or the following spring.
The trick to maintaining colony numbers is to unite surplus - keeping the best or newest queens - at or just before the main flow, which will give you strong colonies (so plenty of honey) and new queens. Alternatively, overwinter or sell surplus.
Enrico's advice will make life easier: keep only good temper and cull bad in spring before drones are about to spread genes.
concentrate more on learning rather than honey production
Good plan, but remember that bees are keener on honey production than learning.