AI Prompt Scorer
Drop in a prompt and get the real TypeGlish grade: an A–F score with the exact contradictions, hedges and vague rules to fix - the same compiler the CLI runs.
This is the real tg score.
This page runs the actual TypeGlish compiler on our server - the same grader the CLI runs locally, where tg score --min B fails your CI when a prompt PR drops below grade.
§ 1Questions
What do the grades mean?
A reads like a spec - ships. B is a few edits from spec-grade. C works on sunny days. D means the model is doing most of the deciding. F is a vibe, not a spec. The number is a weighted blend of the facet scores in the card above.
Is my prompt uploaded or stored?
Yes - to score it for real. Your prompt is sent to our server, run through the actual TypeGlish compiler, and discarded. Not stored, not used for training. Want zero network? Run npx typeglish score locally.
Why did my long, detailed prompt score badly?
Length isn’t rigor. Long prompts accumulate duplicates, buried rules and quiet contradictions - exactly what the Consistency axis punishes. The findings list names the lines.
Is this the same as tg score?
Yes - this runs the actual tg score on our server: real Z3 contradiction proofs, spelling, the full checker. The only difference from the CLI is that the CLI runs locally and gates CI with tg score --min B.