So I'm wasting time on the net as usual when I run across an anecdote from
Operation Plumbob, a nuke testing program back in the 1950s.
It seems that their point criticality tests led to some interesting learning experiences. Follow along: Pascal-A was supposed to check that a single faulty detonator could not actually set off the whole bomb (an implosion type design counts on having a bunch of carefully timed detonators to, well, implode). So the test unit is lowered down a 147 meter shaft and set off, which is when they discover that someone's slipped a decimal point somewhere - the yield is about 55 tons, or 50,000 times what they expected. Oops.
Witness Robert Campbell said, "Biggest damn Roman candle you ever saw! It was beautiful. Big blue glow in the sky..."
Right. After this little learning experience, they redesigned things a bit for the next test.
Pascal-B was a near duplicate of Pascal-A, but after the fireworks display last time, the scientists decided to put a cap on the hole. (The military has a very good understanding of "tube with explosives at one end.") Starting with a piece of four inch thick armor plate, a one-ton steel cap for the hole was made and lowered into place for the test.
Like the first time, this test was predicted to yield about as much force as 1-2lbs worth of TNT. Like the first time, this wasn't particularly accurate. Pascal-B went off at around 1/3 kiloton, or 600,000 times what it should have. Oops, again.
The surface high-speed camera caught exactly one frame of that one-ton steel plate as it left the ground, headed upwards at what was later estimated as
six times Earth escape velocity, never to be seen again. A flying manhole cover moving face-on through the air is a spectacularly un-aerodynamic shape, and the best guess is that it vaporized long before it left the atmosphere.
While the cap was very, very gone it was not forgotten; this incident was brought up later in connection with the Project Orion concept. There's probably a moral in there somewhere.