"I'm Winston Wolf. I solve problems." - which I stole as my tagline. No Acura NSX for me just yet.
Mine is approximately minus one instruction per week, and it was the best money the company ever spent. Hard real-time is hard. (It was a strange processor with, among other things, delay slots everywhere, so you could never make just one change without ripple effects.)
I just dabbled in embedded, once debugged an arduino I2C peripheral, went through the same process, but never fixed the problem :) I was easier to get a different peripheral.
It's sad but I'm pretty sure "types" like you all will be labeled "enemy of the state" soon... Oscilloscope ? Registers ?? ;)
That is, if things will keep going the way things are currently going - corporations, banning periodic tables from schools, elected stallers and clan mentality on all levels.
My personal record is two characters of code per two weeks of work.
Embedded device kept crashing when initializing GUI rendering library (a precompiled one, STemWin IIRC). Two weeks of disassembly, instruction-level tracing of the MCU core narrowed it down to the contents of external RAM not being correct (that is- what was being read did not match what was being written).
Some more poking with oscilloscope, reading the datasheets and low level memory initialization revealed an incorrectly set RAM configuration register that was related to burst writes. RAM test during bringup used word-by-word writes, but when DMA was used only the first word in transaction was written correctly.