After this receiver was bought, I noticed that the AVC didn't really exist as such, what there was was something similar to the injection of two voltage levels into the AVC, a medium one for medium signals in 'Fast', a high one for strong signals in 'Slow' (in this position the sensitivity of the receiver was significantly reduced), and for normal listening the radio had to work with AVC 'Off'.

When drawing the existing AVC circuit it became clear that it had been badly modified with who knows what intention. It is also not clear that mod comes from the army (51J-5s are certainly from military use), IMO it was modified by some previous owner who had the usual objections on SSB reception and didn't like messing with RF gain. There is a lot of creativity in adapting AGC circuits designed for AM to SSB reception, but there is no easy trick based on changing a couple of components and making a couple of connections (if it were that easy, the manufacturer's engineers would have already included it).

Several AGC designs were initially tested in order to restore standard AVC operation, but it was concluded that the one best suited to the 51J-5 was the one based on the existing remains of the original design (similar to 51J-4). The restoration of the AVC circuit in the 51J-5 has been based on the clue provided by these remains, the comparison with that of the 51J-4, and the search for possible compatible designs.
 

The AVC system of 51J-4 combines AVC with RF gain, 51J-5 has the AVC system separated from the RF gain circuit, same as 75A-4. Another peculiarity is that the AGC output in the 51J-4 is through the plate of the DC amplifier, and in the 51-5 through the cathode adding an output diode to prevent any possible positive voltage, just like the AGC systems shown above.

Pure DC amplification means that the only thing that will be processed are voltage levels (more positive or less negative, more negative or less positive), therefore a DC circuit will work more electrically than electronically, the 'signal' concept is different here. Therefore, by making point measurements on a resistor placed between a positive and a negative pole, the only thing that will happen is that the voltage at that point will be more or less positive, or more or less negative, regarding "something" (ground in the circuit below). This is the key to understanding the DC amplified AGC system. Do not confuse this with DC coupling in AF amplifiers.
 

You have to think that the tube is not such, but a variable resistor controlled by the grid voltage, placing a series resistance value from the positive source to its plate in order to place a calibration point at the cathode with some positive volts relative to ground (the delay voltage), and a load resistance at the cathode to the negative source. The less negative the input voltage on the grid, the tube resistance value goes down and more current flows within it (this is what happens when the RF signal goes down), so the total resistance towards the positive side goes down and the connection point at the cathode becomes more positive with regard to ground and the initial adjusted positive voltage at this point. The more negative the voltage applied to the grid, the tube resistance value goes up and less current flows within it (this is what happens when the RF signal goes up), making the connection point at the cathode less positive (relative to the initial positive voltage) or already totally negative with regard to ground (what will happen with medium to high RF signals). The amplification occurs because in the series circuit between the positive and negative poles formed by R0, the triode (its 'Ri' characteristic), cathode output, and R4, one unit at the plate of the D1 corresponds to more than one unit at the cathode of the triode. D2 acts as a polarity controller having the function of allowing only negative voltage to pass to the AGC line.

The working parameters and delay voltage have been adjusted based on the references for the 51J-4 and 75A-4, but note that the AGC time constant has been increased with reference to 75A-4 (.680 µF combined with the 330 Kohm for FAST and 3M3 ohms for SLOW). R6 and R5 values also have effect on the AGC time constant since they are in parallel with R7 and R8, that's why the value of R5 has been chosen very high so that it has little effect on the C4/R7-R8 time constant, mainly in SLOW position. The value of R6 (2M2 ohms) has a direct influence on the degree of "compaction" of strong signals (less dynamics), the value chosen is a balance between the dynamics of strong and weaker signals (e.g. choosing 1 Mohm compacts strong signals too much). Both AM and 'normal' SSB stations can be received very well in the FAST position (that's in fact a MEDium position), and the same for local high power 1 KW SSB stations in SLOW position, so usually there is no need to tweak RF gain. The price to pay is that the audio becomes a bit 'packed', with less dynamics. Audio can be incrementally 'unpacked' by reducing the AVC time constant, but it will increasingly be necessary to act on the RF gain.

The attack (incoming) and release (outgoing) times can be controlled to some extent (as permitted by the AGC source and AGC line impedances) by adjusting the values of C4 and R7/R8. C4 primarily controls the attack time, as it value increases it absorbs the incoming syllabic pulse (attack) and delays the outgoing syllabic cadence (release). The release time is specifically controlled by the values of R7 and R8 which should be adjusted taking into account the time between words in a phrase and the existence, or not, of carrier voltage.