A couple of years ago one of MT staffers explained the way Tx power worked at that moment. It was in heat of discussion about why Mikrotik removed possibility to set antenna gain (via GUI) and why there was hard-coded minimum antenna gain set on most devices[*]. I can't find that forum thread right now ....
Anyway, here's how Tx power is limited:
- country regulations of maximum EIRP. This figure includes all Tx chains (so in case of massive MIMO, power per chain and thus beacon gies down considerably) and crucially it includes antenna gain.
Thus value is set with selection of country, can be channel-specific and is legaly binding for both equipmnent nanufacturer and device owner/administrator. While local radio spectrum inspectors might not visit every homeowner (unless somebody important complains because of interference, such as local ATC or meteorological service), regulatory agencies (such as FCC) will push on manufacturers to make things as correct as possible.
- Tx power amplifier (often integrated in SoC) power limit.
This limit is often rate-dependent due to low quality of low-cost power amplifiers. This low quality means imprecission of power output (smearing of QAM symbols) which means self-made interference. Excerpt from the list is published on product pages of all MT wifi devices - the table(s) under "Wireless specifications".
- user set maximum Tx power
This can be set per rate.
And the way Tx power control works is to frequently evaluate all 3 limits and then adjust actual Tx power to the lowest value. Which means it should be safe to set Tx power to "all rates fixed" with value exceeding some of figures from the table mentioned in bullet #3 above.
There was some discussion about why it's not fine to have Tx power for all rates set to the same value. The effect of setting Tx power for higher rates at lower values is that when a device approaches cell border, throughput will reduce. If power for high rates is set to same value as for low rates (and is lower than chipset limit), then client will enjoy higher data rate right to the point of loosing cell coverage.
In reality it's not like that. Receiver sensitivity plays equally important role and for most devices (client devices as well as APs) it drops at higher rates much more than Tx power.
[*] People traditionally used antenna gain setting as means of reducing Tx power ... as explained in bullet #1 antenna gain is part of equation of EIRP which is limited. With higher setting of antenna gain Tx power has to be reduced not to exceed legal limits. It also seems that when using Tx power table, ROS actually dismissed other bullets in the list above. Meaning that indeed it was possible to push Tx powef amplifier over its limits (as @pe1chl warns in post preceeding his last one). However if Tx power logic was indeed changed as was expkained by MT, then this consideration is not valid any more.