The one idea that connects the workshop

When a problem has no exact answer and no true inverse, you do not give up. You find the best stable approximation instead.

Sections 07, 09 and 10 state it. You watch it happen three times:

1. The pseudoinverse on two columns that are copies of each other, where it picks the smallest of many exact answers (section 07).
2. Tucker on the taxi tensor, 480 counts down to 102 numbers at 6.7% error (section 10).
3. Richardson-Lucy on a blurred photograph, where the direct inverse is unusable (deep dive 13).

Stating that connection out loud at the wrap-up is what makes the second half feel like one lesson rather than four.