Matheus,
thanks for your suggestions:
- spray Ar close to the needle valve, maybe there is a leak there: I did, nothing
- check if the LF connection of the conflo is connected to any instrument: no LF instrument... but I didn't pay too much attention to that port, I'll try
- you said you checked all capillaries, did you check all the way to see if none of them is split apart?: I can't say I visually checked all of them all the way, but if the leak was from a cracked cap, I would assume moving it should alter the leak rate, and if a cap is completely broken, then no flow...
- change the helium tank, maybe this one is contaminated: if Ar was in He, then changing He dilution shouldn't change Ar. And we just tried with another tank, no change.

You mentioned the sample capillary. All things considered, it looks like the most likely culprit

Julien Foriel
Johnston Lab, Harvard U.