Contract review tolerates zero invention: clauses packaged under strict grounding, obligations cited by section, and missing terms reported as missing — never assumed.
Contracts are where AI review is most useful and most dangerous: the language is dense enough that help matters, and an invented term costs real money. This packaging puts clauses under the strictest discipline — only the delimited text exists; every obligation, deadline, and cap cited by section; missing terms reported as "The source does not say." rather than filled from typical-contract knowledge, because THIS contract's deviation from typical is exactly what review exists to find. The task makes the shape explicit: list obligations, deadlines, and caps, citing the section for each.
Forbid the typical
General contract knowledge is the enemy here — strict grounding shuts it out.
Cite every obligation
Uptime, credits, notice periods — each claim carries its section number.
Treat absence as a finding
What the contract does NOT say is reported as exactly that — often the review's most valuable line.
The Strict Grounding (grounding level: Maximum) packaging confines the model to text between <<<SOURCE START>>> and <<<SOURCE END>>>, forbids outside knowledge "in any form," and requires every claim to cite its [§N] marker. Typical-contract fill-in is exactly what hides a clause's deviation, so the grounding block shuts it out rather than letting the model assume standard terms.
For anything the source doesn't state, the packaging requires the exact answer "The source does not say." — then stop, with no speculation about what it might say. Absence becomes a first-class finding instead of a silent fill-in, which is why a missing cap or notice period surfaces as its own reported line rather than being assumed from typical contracts.
It doesn't review anything — long-input-formatter reshapes the text you paste into the packaged block (delimiters, [§N — title] labels, grounding and TASK instructions). You copy that block into your own assistant like ChatGPT or Claude to actually run the review, then own the legal judgment yourself. NewPrompt structures the reading; it isn't legal advice and doesn't guarantee correctness.
Yes — swap the content between <<<SOURCE START>>> and <<<SOURCE END>>> for your clauses; the sample Sections 4–6 (uptime, termination, Data Return) are just placeholders. Re-label your sections as [§N — title] so citations resolve, and if the auto-detected source type looks wrong, override it in the tool as the packaging note says. The grounding rules and TASK carry over unchanged.
Pasting a document raw mixes material with instructions. Package it: explicit delimiters, citable [§N] section labels, and grounding rules — the source travels verbatim.
Transcript analysis fails when speakers blur. Transcript mode packages the conversation with attribution rules: who said it, when it was revised, and no words in anyone's mouth.
A grounded workflow starts with a grounded source: package the reference once with strict rules, then run every question of the session against it.
Token budget planning for real workloads: how much of the window a transcript actually consumes, what is left for the answer, and how much headroom remains.
The "message too long" error has a structural fix: split at paragraph boundaries into sequenced chunks with wait rules, instead of pasting fragments and hoping.
Formats fuzzy agent instructions into a structured prompt with objective, available tools, constraints, success criteria, and failure handling.
Package source material with delimiters, citable section labels, and grounding rules — material and instructions stay separate.
AI pulls the fields you asked for, but hands back a flat list with no way to tell which values it read from the document and which it guessed. Here's how to make each extracted value carry its source quote, location, and a review flag — so you can check the result instead of trusting it.