Please help troubleshooting Advantest r6144

[n_haku]

After start my experiment with jref, I searched on one auction for some useful equipment, and accidentally found this damaged Advantest R6144 voltage/current generator


and almost passed by, but one of seller photos showed it working without errors. After quick search any internal structure pics, it seems that with this damage at high probability I lose only GPIB interface. Well, it's not that big deal with its price (approx $60), so I bid.
After receive, I removed crushed connector, hold switch and ready out BNC connector. Quick test shows any manual function work well, output pretty withing specs, but when I tried long cold-start run logging with diy DVM (inspired by nanodvm challenge), I got

Awww fiddletsicks! what the....Hmm, that oscillation aligned to normal voltage, and have approx 30s period, so it's probably some bad contact somewhere...When I searched internals pics, I run into this blog
https://damien.douxchamps.net/elec/equipment/advantest_r6144/stability_fix/
so relay can be an issue. Ok, quad already no longer available, so at least I can order TQ2 normal and latched to replace G6H-2. To be short, done up until now: removed and short K9 (polarity), K4 (volt<>millivolt) replaced temporarily (TM) with 2 relays and some wires, replaced K1, K2, K3 to new ones, short sense and force traces at connector, linearity pot replaced with 2 resistors, renew solder on critical path. Well, all useless, this behavior almost not changed.

This is well temperature-dependent, when relatively cold in my room, I may obtain up to days w/o oscillation, and if temp rise or covered with some blanket, output start to oscillate. Also if temp is on edge, it can oscillate with much lower amplitude, say 25uV, not usual 250-300uV. Current mode also shows same.
Any advice, what else can I try?
(I found manual here, but seems no interesting info about this problem).

[n_haku]

On side note, while all that repair, I test a bit this generator as I intend to use - null voltage generator to observe uV change in jfet voltage with diy TEC chamber, and notice pretty high noise. Well, specs says 3mV in (20Hz-20MHz), pwm freq is 1kHz, but settling time to 0,1% only 50ms, so LPF tau should be pretty low, no wonder noise this high. A bit of dig in analog pcb results in this schematic of pwm-to-voltage block:

it's less complicated than I initially thought, ok, lets ltspice it:

Reference is 7V, so to obtain 16V, output scaled up, meaning noise also scaled up. Well, I don't need that ultramegafast settling time, noise much more importatnt, so I replaced caps to 680n, 470n, 220n in LPF and 120n+68n in active filter. Theoretically, this should push pwm nouse down to part of uV.