so this has been driving me nuts for about two weeks now. picked up an 817 off a guy at the club swap, worked fine at his place on the dummy load but when i got home and actually put it on a real antenna something felt off. checked it with my Bird and yeah, 40m is giving me maybe 0.8 to 1.2 watts depending on the frequency, everything else seems close to the rated 5w. 20m is fine, 80m is fine, even 2m and 70cm check out.

the previous owner says he never noticed but honestly i dont think he used 40 much. the ALC looks a little weird on 40 too, kind of sits higher than i'd expect before i even key up. not sure if thats related. i did the usual stuff, reseated the finals area, checked the band pass filter area visually and nothing looks obviously burned or damaged. voltage is good coming in.

anyone had something similar? wondering if its a PA transistor issue just on that band or maybe one of the LPF relays is flaky. i dont have a spectrum analyzer but i have a decent scope if thats useful here.

the LPF relay is probably your first thing to check on that. those little relays on the 817 are known to get flakey and on 40m specifically the relay switching can cause exactly that kind of power drop, and the ALC thing you're describing fits too because if the relay contact isnt fully seated the reflected stuff can confuse the ALC circuit. theres a known fix floating around on the QRP forums where guys just reflow the solder joints on the relay coil pins, apparently they cold joint over time from the thermal cycling.

grab the service manual if you dont have it, the LPF board isnt that bad to get to once you pull the top cover. i'd hit every relay joint on that board with fresh solder before assuming the PA transistors are shot, way cheaper to find out first.

had almost the exact same thing on mine a few years back, turned out to be a cold solder joint on one of the toroids in the 40m bandpass filter section, not the relay. scope showed the waveform was clean just low amplitude. took forever to find visually because it looked fine but there was basically no connection under load. if you've got a scope just probe around that section while you're keyed up into a dummy load and see if anything looks wrong upstream of the finals.