Your “hard to read” text might just be badly set

The free workflow

1. 01

   Measure before you fix

   Paste the text and register your guess before pressing the button. The audit reports characters per line, estimated eye stops per line, average words per block, and how many sentences run past 30 words — the reading surface, not the vocabulary.

2. 02

   Re-set the type

   The bench chunks the text into blocks of at most two sentences and ~40 words, sets them at a ~44ch measure with 1.75 leading, bolds each block’s first two words, and leaves the right edge ragged. Short measure shortens the return sweep; leading cuts crowding; bold leads anchor line entry; ragged right avoids the “rivers” justified text creates. It all runs in this page — nothing is uploaded anywhere.

3. 03

   Vocabulary, only if you need it

   The long-sentence count in the audit is the tell: if syntax and jargon are the barrier — not the setting — run the strict-JSON prompt below in a free chatbot and merge the result back into the bench. The chatbot never sees your text until this step, and only if you choose it.

4. 04

   Keep the re-set

   Copy it into any document, or print it. The reading surface you measured is the surface your reader now gets.

What this replaces

The comprehension-layer prompt

You are a plain-language editor. I will paste a dense document after these instructions. Return ONLY a valid JSON object — no prose, no markdown fence, no notes — matching exactly this schema:

{"plain_summary": "60 words or fewer: what the document actually decides or changes", "glossary": [{"term": "a jargon word that appears in the text", "plain": "what it means here"}], "hard_sentences": [{"quote": "the first eight words of the sentence", "why": "twelve words or fewer on what makes it hard", "simpler": "one plain rewrite of that sentence"}]}

Rules: glossary and hard_sentences are capped at 8 and 5 items; only include terms that literally appear in the pasted text; if the text is short, return fewer items rather than padding.

Document:
<PASTE THE DENSE TEXT HERE>

gemini · chatgpt · claude — free web tiers, no card · terms checked 2026-09-09 from training data (cold run — spot-check before republishing)

What breaks

• Lists, tables, and forms don’t re-set.Chunking is punctuation-based. Paste prose paragraphs; re-set a bulleted list by pasting the sentences around it.
• The measured words never change.The re-set fixes the surface. If the audit shows many 30-plus-word sentences, the syntax is the barrier — that’s what the chatbot layer is for.
• OCR text needs a scrub.Stray line breaks and hyphens from photo scans distort the measure; clean them first.
• Chatbots sometimes wrap the JSON.The parser will tell you and keep your input; ask again for “JSON only.”
• Long documents.Over 2,000 characters, the audit covers the prefix — the banner says so when it happens.