The words your spec actually uses, pinned down — and the questions it cannot answer, sent to the person who can

Paste a spec, a PRD, a design note or a meeting transcript. Get its glossary: the canonical terms with tight definitions, the synonyms to avoid, the places it contradicts itself, and a CONTEXT.md you can commit. Then turn the ambiguities nobody in the room can settle into a questionnaire for the one person who can.

All three examples ship with a saved model run, so you can see the whole result — the terms, the contradictions, the CONTEXT.md, the questionnaire — without signing in and without spending a credit.

The document nothing pasted yet

Drop a file in, or Everything is read in your browser. Nothing uploads until you run.

What this is — optional, one or two sentences

Decisions already made — optional; paste the answers that came back and re-run

Paste a document to price the run.

verdict

terms

ambiguities

need a person

scan covered

groups

Ambiguities

Where the document is overloaded, undefined, drifting between names, contradicting itself, or unclear at a boundary. Each carries the quotes that show it, a proposal, and whether the text itself can settle it.

Terms

Term	Definition	Avoid	Group	Evidence

CONTEXT.md

A glossary and nothing else, in the domain-modeling format. Commit it at the repository root and let every later conversation read it first.

Scan reconciliation

questions

themes

needs covered

cannot proceed without

Context and how to answer

Questions

Anything else?

Which need each question serves

Need	Status	Questions	Note

The questionnaire, ready to send

Paste it into a ticket, a thread or an email, or open it on a shared screen and work through it together. “Email it” opens a draft in your own mail app; nothing here sends anything.

Raw model reply

The reply did not parse as the structured result, so it is shown as it arrived.

What this does, and what it does not

The scan is a reader of words, not of meaning. It counts every word under its singular form, notes which ones the document sets off as names — capitalised mid-sentence, quoted, bold, backticked — and looks for the two things a glossary exists to settle: several words for one concept (it knows the usual families: customer / account / user / client, plan / tier / package, cancel / terminate / void) and one word defined or ruled on two incompatible ways. Everything it finds goes to the model as facts it must reconcile, and the reply is checked against them afterwards — if the model skips a candidate, this page says so.

The glossary is opinionated on purpose: when several words compete, one is chosen and the rest go under Avoid, the way a CONTEXT.md wants it. The ambiguities are split into the ones the document itself settles and the ones only a person can — and those become the questionnaire with one button. The questionnaire grills the send, not the subject: it is aimed at what the recipient knows, ordered most-important-first because async means one pass, one idea per question, and it says out loud that "I don't know" is a useful answer.

Nothing here is sent anywhere. The page reads your document in the browser, sends it to the model only when you press the button, and hands you back text. A token or password in the paste is flagged for rotation and never repeated. Nothing to hand? Load the , the , or the . All three replay a saved run for free.