We can't understand the problem without writing the software

If only software thought leaders embraced this. It appears to be a facet of whatever mathematics underlies software, so it's not going away, and it cripples any hope of reliably estimating development time.