How does the automatic checking work?

For each instruction the tool runs lightweight heuristics — length and word-count rules, format checks like JSON or bullet lists, and forbidden-word checks for "do not" instructions. It is a first pass, and you can override any verdict, so the final score reflects your judgement plus the heuristics.

Why let me override the verdicts?

Many instructions are semantic ("be friendly", "stay on topic") and no heuristic can judge them reliably. Letting you set pass, partial, or fail per instruction keeps the score honest while the tool handles the mechanical checks for you.

What counts as an instruction?

The extractor pulls imperative sentences, numbered steps, and bullet points — the lines that tell the model what to do. It skips background and context. You can prune or add to the list before scoring so it matches what you actually care about.

Does anything get uploaded?

No. Extraction, checking, and scoring all run locally in your browser, so you can paste sensitive prompts and outputs without them leaving the page.

Instruction-Following Scorer

Instruction-following scorer

When a model output looks “mostly right” it is easy to miss the one rule it ignored — the length cap it blew past, the format it dropped, the word you told it to avoid. This scorer pulls the explicit instructions out of your prompt, checks the output against each one, and gives you a compliance percentage so drift is visible instead of buried in a wall of plausible text.

How it works

The tool first extracts candidate instructions: imperative sentences, numbered rules, and bullet points — the lines that actually tell the model what to do. It then runs lightweight heuristics on the output for each one: word and character limits, format checks (JSON, bullet lists, headings), and forbidden-word checks for “do not” rules. Each instruction gets a pass, partial, or fail verdict that you can override, because semantic instructions like “be concise” need a human eye. The compliance percentage updates live as you adjust.

Tips and notes

Prune the extracted list first. Removing non-instructions keeps the score meaningful.
Trust the mechanical checks, judge the rest. Length and format verdicts are reliable; tone and relevance are yours to set.
Use it to compare models. Score the same prompt across two models and the percentages give you a fast, concrete comparison.
Everything is local. Paste proprietary prompts and outputs freely — nothing leaves your browser.