Three places recurring value gets promoted to, named directly: Projects (shared context and instructions — Module 5), Skills (packaged, reusable capability — introduced in Module 6's permissioning discussion, revisited here as something to build, not just something to permission carefully), and standing instructions (persistent behavioral defaults).
Three specific things worth promoting, named directly: shared context (recurring reference material, i.e., Module 5 knowledge sources), format (recurring output-shape requirements — Module 2's format component, made persistent instead of retyped), and verification steps (recurring evaluation checks — Module 3's calibrated verification, made a standing part of the workflow instead of re-decided every time).
Optimization is proactive, not reactive. Root-cause diagnosis (§1) and the promotion judgment (§2) are both triggered by something already going wrong. Optimization is the same judgment applied ahead of time — recognizing durable value before a recurring correction becomes annoying enough to force the issue.
This is the course closing its own loop: Module 5's maintenance discipline, Module 6's Skill judgment, and Module 3's verification calibration all show up again here, not as new material, but as the things this module teaches you to actually go invest in.