Evaluation. Knowledge acquisition. Problem-shaping. Approach selection. These are not separate activities you do in sequence. They're facets of a single discipline. They inform each other. They loop back on each other.

This is the unified mental model.

The loop, not the sequence

It's tempting to think of this as a linear process:

1. First, understand the domain
2. Then, shape the problem
3. Then, pick the approach
4. Then, build
5. Then, evaluate

That's not how it works. The real process is iterative:

• You start shaping the problem, and realize you don't understand the domain well enough
• You try to define evaluation criteria, and realize the problem isn't shaped precisely enough
• You pick an approach, and realize it doesn't fit the data you actually have
• You evaluate early outputs, and realize your definition of "good enough" was wrong
• You learn the error shapes, and realize you need a different approach entirely

Each element informs and revises the others. The discipline is staying in the loop — not pretending you can get it right in one pass.

The core insight: evaluation and approach selection are the same decision

Choosing how to build and choosing how to evaluate are not separate choices. They're two sides of one choice.