Mikrotik Capsman clients fall off, where to dig?

Question

WiFi

There are 3 pieces of hAP ac². The latest firmware is stable.
The first point is the main gateway and CAPsMan (and he is also a client), plus all the necessary VPNs in different directions.
The second and third points are in CAP mode.
Problem:
Clients lose connection.
It looks like this: everything works fine for 10-30 minutes, then packets stop coming.
From clients the gateway does not respond and that point on which the client hangs does not respond, and nothing further, respectively, too.
Packets to clients at this moment also do not go from the gateway or from the point, ping does not go through.
After some time (from 10 seconds to 5 minutes), the connection is restored. At the same time, the packets go for some time with a delay of up to 5 seconds (not for long), then everything returns to normal completely. If you reconnect WiFi on the client, then everything starts working fine, for a while.
During the fall, clients continue to show the full signal level with a dot.
In the logs of the points, nothing suspicious happens at the time of the falls (and nothing happens at all).
The problem is with all clients, at all points, regardless of their location and frequencies used. Either one client will fall off, then the second, then several at a time.
When connected with a wire, there is no such problem, everything is stable.
Wireless clients for each point from 3 to 10, the network is gigabit.
There are few other points in the district, the air is almost empty.
What we tried:
Enabled proxy-arp in the datapath, selected specific channels instead of automatic ones, set up an access list to cut off clients with a weak connection, changed group_key_update in different directions, took DHCP lease into space, danced with a tambourine.
Config (with overwritten external ip): https://gist.github.com/BATCOH/42a8b230b1ab9ec5970... A
similar question (there advice did not help): The Wi-Fi network on Mikrotik falls off. What could be the problem?
What can be done?

Answer 1

Look at what is in the log at this moment. Check the bridge settings on the points. This happens when the bridge setup and local forwarding are "inconsistent".
Sometimes interfaces remain in the bridge, even with localforwarding configured, then it is better to reset the point without the default configuration.