Seven steps.
One baseline.

Assemble a pipeline

Pick a product and model for each step, write the prompt, chain two to four steps into something that produces a finished deliverable.

Run it and see every stage

Execute end to end, see each step's output and the final result, and see what the run cost.

Declare a ground truth

Accept one run as the reference point for this creation. This is the version you're happy with.

Change one thing

Edit a prompt, swap a model, adjust a setting at any step.

See what your change did

After rerunning, see how far the output moved from your accepted version, measured at the step you touched and at the final deliverable, with both versions side by side.

Judge it

Thumbs up or thumbs down, with an optional reason. Thumbs up makes the new version the reference point.

Review the record

See every run for this creation, what changed between each one, and how you judged it.