Iteration Efficiency (IE)

Weight: 20% — measures how efficiently your session progresses toward a solution.

Convergence (0–10)

Do your prompts build toward a solution, or do they circle back?

High convergence:

• Each prompt builds on the previous result
• Scope narrows over time (broad → specific)
• Few corrections needed between prompts

Low convergence:

• Same issue addressed multiple times
• Scope keeps changing or expanding
• Frequent “undo” or “revert” requests

Examples:

Score	Pattern
9	Prompt 1: plan approach → Prompt 2: implement → Prompt 3: test → done
5	Prompt 1: implement → Prompt 2: fix bug → Prompt 3: fix same bug differently → Prompt 4: works
2	Prompt 1: do X → Prompt 2: undo X, do Y → Prompt 3: undo Y, do X again → …

Recovery (0–10)

When something goes wrong, do you recover efficiently?

High recovery:

• Adds constraints when a prompt fails (“try again, but only modify the return type”)
• Provides the error message in the follow-up prompt
• Changes approach rather than repeating the same request

Low recovery:

• Repeats the exact same prompt expecting different results (retry storm)
• No additional context or constraints in follow-up
• Gives up and starts over unnecessarily

Examples:

Score	Pattern
9	”That approach broke the tests. Instead, keep the existing interface and only change the implementation in processQueue()”
5	”That didn’t work. Try a different approach.”
2	[same prompt repeated 3 times]

Improving IE

1. Add constraints on failure — tell the AI what to avoid, not just what to do
2. Include the error — paste the actual error from the failed attempt
3. Narrow scope — if a broad change failed, break it into smaller steps
4. Use plan mode — for complex tasks, plan before implementing to reduce iteration
5. Avoid retry storms — if the same prompt failed twice, change your approach