CORRECTION: The divider total resistance is 15 k Ohm (that 1 M I mentioned was for a battery voltage readout in another project that must have minimal leakage, in order to sustain months on a coin cell). I am sort-of forced into this order-of-10k resistance since I have to sense the contact with under-100-Ohm as good, and over that ignore.
The analog readouts on the bench are noise-spread in the range of about 300 units in the total 0-4095 range. I may try to record the lines readout in the field, that may show the noise higher than on the bench - thanks for the pointer. The thing is I did not see false readings in the field, e.g. indication of a contact while the object that creates the contact is physically not there, but that may be because the code asks for readout on both lines to be in range to call them "in contact", so spurious signal on all wires may bring one in range but will pull others out at the same time...