Worsley AI

Metrics

Each Worsley metric measures a different part of working with AI. Find the one that fits how you work, then read its full page.

Which One Fits You

ROIP

Return on Iterative Prompting

How effectively you work with AI across an entire conversation.

Where It Applies
AI chat sessions, captured in your browser and your terminal.
What It Looks At
Which of seven criteria each prompt needed, given its task and its place in the conversation, and whether the AI had them from it or the conversation before it, not how the AI answered.
What You Get
A score, and feedback on turns and whole sessions: why something affected the interaction and how to do it better next time.
Example

You ask ChatGPT for a customer email without saying who it's for or what changed. You add both in the next turn. ROIP marks Context as missed on the first prompt and covered on the second.

What Is ROIP

ROD

Return on Decisions

How effectively you and your code use decision models, such as TypeSafe’s Jev.

Where It Applies
Decision models: typed state and questions in, typed decisions with probabilities and confidence out.
What It Looks At
The decisions: how they were asked, and how they were used.
What You Get
How well your decisions pay off, and what to change.
Example

A support tool asks a decision model whether to approve a refund, passing the order total, the days since delivery and the reason. It gets back approve at 86%, escalate at 11% and deny at 3%, with high confidence. Worsley shows the decision with the state it was made from.

What Is ROD

Side by Side

ROIP

Return on Iterative Prompting

What It Measures
How effectively you work with AI across an entire conversation.
Where It Applies
AI chat sessions, captured in your browser and your terminal.
What It Looks At
Which of seven criteria each prompt needed, given its task and its place in the conversation, and whether the AI had them from it or the conversation before it, not how the AI answered.
What You Get
A score, and feedback on turns and whole sessions: why something affected the interaction and how to do it better next time.

ROD

Return on Decisions

What It Measures
How effectively you and your code use decision models, such as TypeSafe’s Jev.
Where It Applies
Decision models: typed state and questions in, typed decisions with probabilities and confidence out.
What It Looks At
The decisions: how they were asked, and how they were used.
What You Get
How well your decisions pay off, and what to change.