"Produce" vs. "Testing" efforts

[paulca]

In software a blunt finger point is 20/80.  20% writing the code, 80% testing and managing it.

Now a "one man band" hobbyist might not come close to that figure, certainly not formally, but if you include the "Why this no work?" and "Why does this work?" loops during manual testing, it's probably not that far off either.

Anyway my question is more around complex(er) digital designs I am now working on.  It seems the effort you need to go to to "unit test" a circuit increases linearly or possibly geometrically with the board complexity.

So if you have a "latch" circuit, you can basically breadboard it.  If you have a RAM chip with 32 pins and a bus with 60 signals thats not an option.  Breadboards are out once you get anything complex.  The more connections the more probability of a bad one until you end up with a circuit where you are all but guaranteed at least one at a time... and ... thats not a test reference anymore its a liability.

So next is protoboards and where my question arrives.  It took me longer to solder together a test harness for a circuit than it did to solder the UUT itself.The test code was probably twice as many lines in C as it was verilog under test.

I am now looking fondly at the 1-2week lead time and $35 cost of a PCB for just a test rig.

For a larger board, like a CPU board... I can even see potential in designing, printing and testing the test harness first, then the actual PCB...

So I gather this comes down to "Cost" at it's bottom, "Cost" vs. "Correctness".  At what point does the test overhead outweight the cost of the failure mode they test for.