Feed a full contract into Grok's long context window for one pass, not five

You are running a risk scan across an entire long document in one pass, using Grok's extended context window to hold the whole thing at once rather than summarizing it section by section. The specific value of a whole-document read is catching a contradiction or inconsistency between two clauses that sit far apart in the document — a risk that a chunked, section-by-section review would structurally miss, since no single chunk would ever contain both halves of the contradiction.

DOCUMENT
The full text of a 40-page vendor services agreement, including all exhibits and schedules

DOCUMENT TYPE AND PURPOSE
A services agreement with a software vendor, being reviewed before signature by the business side, not by outside counsel

WHAT THE READER ACTUALLY NEEDS TO KNOW
Whether this can be terminated without penalty if the vendor's service levels slip, and whether liability is capped somewhere unfavorable relative to what was verbally negotiated

KNOWN PROBLEM AREAS FROM PAST DOCUMENTS OF THIS KIND
Past vendor contracts from this same vendor have had a liability cap in the main body that gets quietly overridden by a higher, less favorable figure buried in an exhibit

SCAN RULES
Read the entire document before flagging anything, and specifically check whether any clause, definition, or number appears more than once with a different value or different wording each time — a defined term used one way early in the document and a subtly different way much later, a liability figure stated once and a different figure appearing in an exhibit, a defined time period appearing once with a carve-out and once without — since this exact class of inconsistency is the reason a whole-document pass exists at all, and is invisible to any review that only ever holds one section in view. Flag every clause that materially favors the other party in a way that isn't obviously standard for this type of document — distinguish a boilerplate clause that appears in essentially every document of this kind from one that's been quietly tilted further than standard, and say explicitly which category each flagged clause falls into. Check the known problem areas named above specifically, even if nothing else about the document raises a flag — a pattern that's caused a real problem in a past document of this kind deserves a direct, deliberate check here, not just a hope that the general read happens to catch it again. Note anything defined and then never actually used elsewhere in the document, and anything used that was never defined — both are real drafting risks, not just tidiness issues, since an undefined term used in an operative clause creates real ambiguity about what was actually agreed to. State plainly where you are uncertain whether a clause is actually risky versus merely unusual-sounding to you without deeper knowledge of standard practice in this specific document type — a risk scan that reports every unfamiliar clause as dangerous is not more careful, it's less useful, because it can't be trusted to distinguish a real problem from an unfamiliar but perfectly normal structure.

OUTPUT FORMAT
1. Cross-document inconsistencies found: each one naming both locations and both conflicting values or wordings.
2. Materially unfavorable clauses: each tagged standard-but-present or unusually tilted, with the specific language quoted.
3. Findings against each known problem area named above, even if the finding is checked and not present here.
4. Undefined-but-used and defined-but-unused terms.
5. Anything flagged with genuine uncertainty about whether it's actually a risk, named as such rather than folded in with the confident findings.