A Half-Million-Dollar Review Estimate Bought a 5,000-Document Test Run

United Association National Pension Fund v. Carvana Company, No. CV-22-02126-PHX-MTL, D. Ariz. (July 15, 2026), full opinion (PDF)

Three companies served with document subpoenas in a securities class action told an Arizona federal court that the plaintiffs' search term would hit close to 200,000 documents and that reviewing them could cost more than half a million dollars. The court treated the figure as a real problem and declined to quash. It cut the subpoena to a sample of 5,000 documents and sent the parties back to the table to work out whose ESI would be searched to produce them.

Non-party burden fights usually end in an all-or-nothing ruling on a cost estimate nobody has tested. The Arizona court ordered the sample first and held the burden question open until it could see what the sample returned.

What Happened

The subpoenas went to DriveTime Automotive Group, Bridgecrest Credit Company, and Verde Investments, three entities related to defendant Garcia Sr. The underlying case is an investor suit against Carvana under the Securities Exchange Act of 1934, built on an alleged pumping and dumping of the company's stock. The requests reached documents and communications about the three entities' transactions with Carvana, everything they had given to or received from regulators and law enforcement concerning those transactions, and documents sufficient to show Garcia Senior's and Garcia Junior's control or ownership of the entities. The subpoenas issued in July 2025, and almost a year of conferral produced no agreement.

The three companies opposed on relevance, on duplication, and on burden. On burden they put numbers in front of the court. The search term the plaintiffs wanted run returned close to 200,000 hits, and the companies estimated review at $170,056 to $557,968 depending on the methodology, plus $920 a month to host the data.

The Court's Analysis

Relevance and duplication went nowhere. The court found the documents relevant to the plaintiffs' securities fraud claims and therefore discoverable. It also found them not cumulative of the defendants' own production, since they did not appear to be in the defendants' possession. The plaintiffs had further carved communications with Carvana out of the search, lowering the risk of retrieving material the defendants had already produced.

Burden was the closer question. Magistrate Judge John Z. Boyle acknowledged that the Federal Rules give non-parties added protection from the time and expense of complying with a subpoena. He then wrote that the court "does express concern with the potential half-a-million-dollar cost of this search". Weighing the plaintiffs' need for documents the defendants do not have against the expense to the three companies, and mindful of how broadly the requests were written, he found no burden severe enough to require quashing.

The remedy was a sample. The order modifies the subpoena to a test run of 5,000 documents from the three companies and their ESI custodians, with production due by July 24, 2026. The court framed the sample as a way of testing the burden claim rather than accepting or rejecting it, ordering the run to determine "whether any further disclosure would indeed present an undue burden upon the implicated third parties".

Scope went back to the parties. Counsel were directed to confer on the 5,000-document search, including "whether the search will be limited to certain custodians and how to limit the search of their ESI data". The same order resolved a set of work-product disputes over the defendants' communications with putative class members, which turn on doctrine unrelated to ESI.

Why It Matters

A hit count and a vendor quote predict what a collection holds. A sample measures it. Counsel on either side of a non-party burden motion should be ready to propose a test run, with agreed terms for what happens if the sample comes back rich and what happens if it comes back thin.

Nearly 200,000 hits and a ceiling approaching $558,000 bought modification of the subpoena rather than relief from it. The first 5,000 documents still belong to the non-parties to produce.

Arrive with a per-custodian breakdown. The court left open whether every custodian's ESI gets searched at all, and that is the argument a responding non-party wants to make. A total hit count across all custodians says nothing about which custodians carry the volume. Without the breakdown, a responding party has nothing to offer when the court asks how the search should be narrowed.

Sampling also gives a court a way to rule without guessing. A judge asked to weigh a six-figure estimate against a party's need is being asked to price a population neither side has seen. A small ordered production turns that guess into a fact both sides can argue from.

The full opinion is available as a PDF.

BlogeDiscovery Case in Focus