A poltergeist in my machine?

So I've mentioned how I've seen this wierd behaviour where QBUS memory boards that hadn't been used in a long time didn't work when first plugged in, but started working later. I just had something even weirder happen, and am curious if anyone has an plausible explanations. So I had a dead M8044 (MSV11-D), symptom was that you could write -1 to any location, it read back as 0. Quite repeatable, I can power cycle the machine, take the card in and out, etc, etc. So I throw it on an extender, and start chasing. I have a two instruction loop (write location 0, loop), and I'm watching the data going into the memory chips on the card, and it all looks good. So I add a third instruction (read location 0, after the write), and continue chasing. Data looks good coming out of the chips; then it goes to an octal latch. So I look at the latch enable, and that doesn't look so hot - just a tiny little ugly spike. So I look at the source of that, and it's a D flop. So I look at the D flop's clock input, and it's also a nasty little spike. So that comes from the output of a triple-AND, and so I start looking at the inputs of the 3-AND. And when I put my 'scope lead on the second input... the memory suddenly starts working! Well, I could see that - the added resistance or capacitance or whatever of the probe might have had some effect on a circuit that was right on the edge. But here's where the ghost enters the machine. I pull the 'scope probe ..... and the memory keeps working! I can power cycle the machine, leave it off for 15 minutes, power it back on - and the memory still works fine! Does anyone have _any_ idea WTF is going on here?!?! I feel like I'm in some sort of AI koan...