I can construct a chain of events that would theoretically make detonation possible, but as the anecdotal evidence confirms, it's highly improbable on a random basis. The stars have to line up just right. The main problem for the laboratory is, once you close a case by seating a bullet, you loose control of the exact distribution of the powder charge inside, which I've come to suspect is critical.
As one writer was pointing out recently, one of the problems for the firearms industry has been that there is no NASA or other heavily funded outfit interested in firing millions of rounds in thousands of actions under controlled conditions to identify oddball problems and conditions, much less firing millions of rounds of reduced loads-only to experience the odd detonation. The military isn't interested in studying catsneeze rifle loads or, for that matter, reloads of any kind, so these kinds of things weren't done while the SBRL was still active.
There's an interesting comment by Wm. C. Davis, Jr., on page 133 of the out of print NRA publication, Handloading, in which statistical analysis and control of test conditions and interaction of components is being discussed (Interaction in this sense is when one component influences how another influences the load. For example, you rank primers by the velocity they produce with a fixed load, then change just the case or just the bullet or just the powder charge, and it results in the primer rank changing order.):
"As a matter of experience in ballistic testing, it often happens that two samples from the same lot of reference ammunition, fired on consecutive days, will show differences between averages that cannot reasonably be ascribed to the random shot-to-shot variation observed on the two separate occasions of firing. This is due apparently to some systematic difference in test conditions that cannot be eliminated, even by the most diligent efforts to control the procedures, and has plagued investigators in ballistics laboratories for many years."
Bottom line: things happen that you can't control, including in laboratory testing.
It follows from that and the nature of random distribution, that once in a long while some event occurs that you might have to fire a million rounds to see once. As I said before, nobody wants to fund that level of testing in a lab environment. But with many millions of rounds fired by civilian shooters every year, sooner or later, someone sees them.
I don't go through life ever expecting to experience a detonation and I don't recommend you get over excited about the possibility. It's small. But given that several strange things can happen in largely empty cases, why increase the odds of such things happening to you when that condition is easy to avoid?