Build Status: Released v0.1.0

Prompt Laboratory

See why a small wording change flips model behavior, by comparing prompt structures side by side.

Runs in your browser. Nothing you enter is sent to a server, and nothing is stored unless you save it to this device.

Version A 0 tokens

Version B 0 tokens

Findings

Detected by local pattern analysis. Select a finding to highlight the exact text it refers to.

Idle

Stale Inputs changed after this result was produced. Run it again to see numbers that match what is on screen.

Prompt anatomy

Where the prompt spends its budget.

Idle

Stale Inputs changed after this result was produced. Run it again to see numbers that match what is on screen.

Side by side diff

Word level difference from version A to version B.

Idle

Stale Inputs changed after this result was produced. Run it again to see numbers that match what is on screen.

Improved draft

A rule based rewrite of version A. Every change is listed with its reason.

Idle

Stale Inputs changed after this result was produced. Run it again to see numbers that match what is on screen.
Note

This is not a model rewrite

These edits come from the same local rules that produced the findings above. Where a fix needs your judgment, the draft inserts a bracketed marker rather than inventing a requirement you never stated.
Export
Note Why does a small wording change alter behavior?

A prompt is not read as prose. Every clause either narrows the space of acceptable answers or leaves it open. A word like appropriate does not narrow anything, so the model fills the gap from its own priors, and those priors shift between versions, temperatures, and even reorderings of your text.

The findings above are the places where your prompt leaves the decision open without saying so. That is usually where behavior moves when you did not expect it to.