V1 is an amplifier.
V2 is an amplifier.
V3 is an amplifier.
V4 V5 is a power output stage. It uses push-pull parallel because this box has a HIGH rated output: it may have been asked to drive long lines in high-RF fields with heavy line-EQ networks. +30dBm was routine on somewhat beefier broadcast limiters.
V1 gain is pretty much constant, and it also derives push-pull signals from the single-ended input (transformer and gain pot).
V2 is a variable gain amplifier. Any amplifier will lose gain as you change bias to a lower DC current. In most amplifiers you don't want that. Here, R11 R12 go to a varying DC voltage which can reduce V2 gain.
V3 V4 V5 are all tied together in negative feedback with R31 R21 R32 R24, so we can treat this as a single stage.
V6 rectifier compares the negative peaks of the output voltage to the positive voltage on R47 wiper. When output voltage exceeds the set level, the excess ends up on V2's grids, in a direction to reduce its gain. That's how it limits.
If the control voltage followed the peaks, we would just be clipping. What we really want to do is: if peaks are high, drop the gain quick, if peaks are not high let the gain rise slowly. V6 diode action and R40 C16 R39 appear to set the different time constants.
What is really exceptional is that the control voltage also appears on coupling caps C3 C5 C4 C6 via the high-value R11 R12. So we really have a complex multi-step 3-pole attack time constant. And the dirty fact is: C3 C4 must pass bass, which means attack time constant must be slower than bass waves. Up around 50mS. That is slow enough to allow audible distortion on sudden transients. And it is locked into the design. Other similar designs manage to avoid this tradeoff and can have much faster attack times. I presume this was more a safety limiter than a "sound effect", though the combination of clipped attack and slow limiting may have application in highly processed music.
Limiting level is a function of several voltages. V7 supplies a stable voltage to those points.
Connect a true VU meter to chassis connector pins 7 and 3. Switch S1 reads current in each tube; they should all be nearly the same indication. If not, note which tube reads "different" and schedule a tube swap when the radio station goes off the air for the night. A true-VU meter from pin 5 to 3 will indicate amount of gain reduction.
Overall gain is VERY high, possibly 10,000 or 80dB. It may actually be mike-input. But I suspect they expected you to pad the input according to your local levels. For modern studio use, you might even swap the primary and secondary on T1: this gives a more manageable gain of 40dB and a hi-Z input.
Intended output limiting level is around +18dBm. You can trim R47 for lower level but the limiting action will be very soft. For studio use, you will probably want a pad on the output, so it can run hard without smoking your inputs.
Frankly, this beast is strange even within its genre.