Ask questions that cut across 9,400 data-room files, without a single one leaving the laptop

9,400 files on a single SSD

What the seller handed over isn't an organized set. It's a dump of the folder tree as-is. Naming conventions differ by department, and the draft, the redline, and the executed copy of the same contract sit in different folders. The questions diligence has to answer cut straight across the pile.

• 9,400 files in the virtual data-room export. Contracts, articles of incorporation, permits, litigation records, and board minutes, all in one tree.
• About 2,700 contracts. Supply, distribution, license, lease. A good share are scanned PDFs, so full-text search doesn't reach them.
• One set of reps and warranties. The list of contracts the seller says it "disclosed" lives in an attached exhibit.
• One spreadsheet of top customers by revenue. Customer name, last three years of revenue, ten rows of ranking.
• Two years of board minutes, about 24 files. For some contracts, the approval record exists only in the minutes.
• The same company is written three different ways across contracts: "Taeyang Co., Ltd.", "Taeyang Components", "Taeyang Components Co., Ltd."

Today, six people split the folders

Six associates divide the folders between them. Each one runs workspace-wide searches for "change of control," "Change in Control," "change in ownership," opens every file that hits, checks the clause text, and copies the contract name, section number, and gist of the clause into a shared spreadsheet. It's a three-week schedule, and in the final week the senior re-checks that worksheet.

Three things break. Scanned PDFs don't hit in search, so people read the cover page and move on. If the search term doesn't line up with the contract's wording ("in the event of any change in a party's status under this Agreement"), the clause is there and never surfaces. And search only answers inside one contract at a time. "Which of those contracts has a top-ten customer on the other side" means keeping the spreadsheet open alongside and matching by eye.

A cloud document-review tool automates a good chunk of this work. This diligence proceeds on the premise that privileged documents are in the set, so nothing gets uploaded outside.

9,400 files go into the workspace

Prep starts by dropping the folder in whole. No sorting.

What you type into the chat box

"Which contracts have a change-of-control clause where the
counterparty is in the top 10 by revenue?"
→ 7 contracts. Each one quotes the clause text, with a
  [[Supply Agreement (Taeyang)]] reference and Section 3.2 attached.
  Below it, the mini graph "Where this came from in your notes" is drawn.

"Of the contracts the reps and warranties say were disclosed,
which ones have no actual file in the data room?"
→ 2. The board minutes carry an approval record; the contract file is missing.
  A [[2024-08 Board Minutes]] reference and Item 3 are attached.

"Of those 7, split the ones needing the counterparty's prior consent
from the ones that only need notice."
→ Consent 4 / notice 3. Each one quotes the sentence verbatim.

The first question works because the contracts and the spreadsheet sit on the same knowledge graph (a map that connects concepts as points and lines). A search that pulls back the passages closest in meaning gets you as far as "contracts with the clause" and stops at "and of those, the top-revenue customers." The counterparty in the contract and the customer in the spreadsheet have to be joined as the same entity for both conditions to meet inside one question.

The path to verifying an answer is short. On an entity page, hover a connection chip and you see the exact note and sentence the fact came from; click a [[bracketed]] reference in the answer and the source note opens. Once a conversation checks out, freeze it: in the session list, right-click → "Pin as case…". Days later, when the seller uploads more material and you reopen the case, a "New evidence on {count} pinned entities since your last visit:" strip points at exactly what grew.

What changes

People still do work in this flow. Judging which clause is a deal breaker, reading the clause text, choosing what to put to the seller. What shrinks is opening 1,200 hits one at a time to check whether the clause exists and copying it into a spreadsheet.