Victor Stanley v. Creative Pipe: the privilege review, run with Claira

Victor Stanley, Inc. sued Creative Pipe, Inc. and Mark and Stephanie Pappas in the District of Maryland, alleging copyright and patent infringement and unfair competition. In discovery the parties agreed a joint protocol for electronically stored information, with nearly five pages of search terms to find responsive material. The defendants first asked for a court-approved clawback agreement, then told the court one was unnecessary because they would review each document for privilege before producing it.

We take the file at that point, acting for the defendants, with close to 39 gigabytes to produce. About 4.9 gigabytes are text-searchable. The other 33.7 gigabytes are image files that a keyword search cannot read, among them engineering drawings.

In the real case, 165 privileged documents went out with the production, and Magistrate Judge Paul W. Grimm held in 2008 that privilege over all of them had been waived. This walkthrough asks how the same team could run the privilege review so that it would hold up.

The first job is to know what the production holds. Claira's early case assessment, opened from the Claira menu under More, reads the case's metadata and reports the document count, the custodians, the date range and the composition of the set, including file families, extensions and document types. It also charts documents per month, shows who writes to whom and suggests review segments. It reads no document content, so it can run before any review decision is made.

For this team the composition figures matter most. They set the email and office files with extracted text against the drawings and scanned images. In 2007 the defendants reviewed the image files by their titles alone, and the breakdown makes plain at the outset that most of the volume needs a method that reads the page itself.

The tables of correspondents do a second job. They point to the mailboxes that carry traffic with outside counsel, which is where a privilege screen needs its closest reading.

A privilege screen depends on knowing who the lawyers are. The team would open Case Context and start a context from the Litigation/eDiscovery template, which sets out the parties, the issues, and the relevance and privilege rules. Into it go the claims, the outside firms and each lawyer who advised the defendants, with the dates of each engagement. Claira can draft sections from up to five key documents selected in Nuix Discover, such as engagement letters, and it lists up to five questions the team still needs to answer. Once a partner clicks Approve, the context is locked and the prompt generator can use it.

From the approved context, Generate search terms produces a search term family of dtSearch queries, which the team downloads as an Excel file and imports into Nuix Discover. The terms serve as a cross-check on the AI screen, and the team keeps the list on file with its hit counts. Judge Grimm faulted the defendants for failing to identify the keywords they had used or the qualifications of the people who chose them.

The Privilege Review workflow starts from a prompt written for Canadian law, so the team adapts it to the attorney-client privilege and the work-product doctrine. Their version asks for a privilege call, the privilege types, the people involved, a short justification and the passages that support the call. They try it in single review on 10 to 25 documents first, including easy ones, hard ones and documents with little text, and revise the wording until the calls match what a senior associate would decide.

The bulk scan then runs the prompt across the production and writes the answers into a Discover field while the task is still running. The Scan as setting decides what Claira reads. In Auto mode, a document with a PDF or image file is sent as that file, so a drawing marked up by counsel would be read from the page itself. Image scan accepts TIFF, including multi-page TIFF, along with PDF and the common image formats. A long run can be sent to the background and keeps its Task ID.

Judge Grimm wrote that the only prudent way to test a keyword search was to sample both the documents found privileged and those found not privileged, to reach a comfort level that the categories were neither over-inclusive nor under-inclusive. The same discipline suits an AI screen. The team draws a sample from each side of the privilege field in Nuix Discover, and associates read and code every sampled document themselves.

Claira's QC Review workflow then helps with the comparison. It applies one criterion across the coded documents and flags those where its call differs from the human one, and when the prompt asks for a one-sentence justification, the reason for each disagreement can be read quickly. In a privilege screen the cause often lies in the inputs: a lawyer missing from the Case Context, an in-house adviser whose role was unclear, a thread forwarded outside the privilege. The team fixes the cause and scans again.

Each prompt change is kept as a version in the prompt history, and the version used for the production can be marked final.

Some decisions here belong to the lawyers alone. In the real case the defendants asked for a clawback agreement, let the request go, and relied on a review that the court later held they had not shown to be reasonable. Whether to seek that protection this time, however well the screen performs, is a judgment about risk for the partner running the case.

The same goes for the hard documents. The sample will turn up communications that mix business and legal advice, and documents shared with people outside the privilege. Privilege over a thread copied to a third party, and work-product protection for a document prepared before suit, are calls for the lawyer who will sign the log.

Judge Grimm weighed five factors in finding waiver: the reasonableness of the precautions taken, the number of disclosures, their extent, any delay in putting them right, and the interests of justice. Of the five, a documented and tested method speaks most directly to the first.

The court set out what a party claiming privilege must do: describe each withheld document with particularity in a privilege log under Rule 26(b)(5), and, if challenged, give an evidentiary basis for each element by affidavit or other evidence. The team builds the log as an Insight from the Privilege log quickstart template, reading the privilege field from the bulk scan. The table downloads to Excel or Word for the lawyers to finish, cited entries link to their source documents, and the log's sources can be found again in Discover by the Insight's reference.

The method needs its own record. The Bulk Task History Report gives the prompt text, the model, the document counts and the comments for each scan. The prompt history exports with its versions, and the approved Case Context downloads with its status badge. Together with the sample results, those papers cover the points the court found missing in 2008.

The real case ran on after the privilege ruling, through later sanctions opinions, to a judgment that the Federal Circuit affirmed in 2013.