This is going to take more than one. Sometime around 1980 I was a co-op in a steel mill. There was this huge machine that pressed slabs of hot steel into thin strips. It was overheating. 1/
The machine spec said it was up to a root mean square (RMS) amperage of 400 or so. Damn, that's a lotta juice. 2/
Anyhow, the thing had an amperage data trap, so they plugged in an HP data recorder for like 12 hours. Then they needed to compute the RMS. 3/
They had a PDP 11/10 with a place you could plug in the data recorder and read a value out of an absolute memory address. And it had a FORTRAN compiler. 4/
Who knew FORTRAN? The snot-nosed hippie co-op did! So, I wrote a program that fetched the amperage and accumulated the data and computed the RMS. 5/
(I kind of wish I remember how you dereference an absolute memory address in FORTRAN, but I digress.) 6/
Anyhow, the output showed the RMS starting kinda high but drifting down and being well under 400, so we thought we were inside the envelope. 7/
Unfortunately I had stored the sum of squares in a REAL variable which in FORTRAN (that version anyhow) meant 32-bit float. 8/
You don't have to accumulate 400-or-more squared *that* many times to overflow a 32-bit float. 9/
The RMS was wrong. The report went to the steel mill's General Foreman, a god-like creature. They told the mill operator to carry on. 10/
Fortunately the mill operator was a grizzled old steel guy who could tell when several million bucks worth of molten metal squisher was about to self-incinerate. 11/
He slammed it off on instinct with a steel slab halfway through and it only cost a few shifts downtime to remediate, which is to say many times my pathetic co-op salary. 12/
Trying to explain "double precision" to the General Foreman really wasn't fun. 13/13
