For Andreas Design a single 3300µF-input-capacitor is used (or 2200µF + 1000µF).
The leakage of 105°C-rated caps appear to be higher compared to 85°C-rated caps, according to my limited sample observation of 10 105°C caps vs 10 85°C caps. Also there are low-leakage-specified caps like
https://eu.mouser.com/datasheet/2/293/e-kl-14230.pdf I tested about 10 3300µF/16V/105°C Panasonic-NHG-series capacitors and all of them were at about ~40nA after 48h at 10V, but i didnt form them at 16V before.
After that i bought 10 3300µF/50V/85°C Panasonic-M-Series (16V-capacitors were not available at that from moment from the seller), formed them at 50V for about a day, so the aluminium-oxide-layer could heal/stabilize.
Then i discharged them to 10V and let them stay at that voltage for 48h, i measured leakage-currents of <=5nA for 6 of them i think.
I used a calibrator to generate the stable 10V and 1Meg-resistors in series with each capacitor. My 6.5-Digit-DMM 34465A was used to measure the voltages over the 1Meg-resistors, 1nA -> 1mV over the resistor. But a stable voltage-supply might also be used to generate the test-voltage.
You can also use the discharge-current method you mentioned, but i guess it takes at least the same amount of time as the charge-current-measurement-method, since were basically waiting for the dielectric absorption in the cap to stabilize.
If youre using the PCBs made by mimmus78: the pcb fits tightly inside a TEKO 393, if you cut of a few mm on the width of the pcb with lever scissors. Also the mentioned 3300µF/50V/85°C are bigger than the 16V-types, but still fit on the pcb and inside the TEKO 393.
TL:DR:
I suggest you get 10 3300µF/>=16V/85°C-capacitors, form them at their rated voltage for a day and then let them stabilize at 10V for 48h. That should get you at least one very good capacitor with <=5nA leakage-current. Remember to cool the capacitor-leads during soldering, otherwise its leakage might jump up due to the temperature-shock...at least thats what ive heard.
I tried to measure the leakage-current first with my Keithley 2500 Picoampmeter, but it wouldnt show me reasonable readings (100Giga-Ohm-resistor and Keithley220-current-source was tested fine with it). I guess it didnt like the heavy capacitive component? Maybe your Keithley 196 might show this behavior also, thats why i used them 1Meg-resistor-method.