In re Class Action Settlement Administration Litigation, Misc. A. No. 25-179 (JDB), MDL No. 3162 (D.D.C. Aug. 19, 2026), full opinion (PDF)
Judge John D. Bates declined to require two of the three sampling steps plaintiffs wanted written into a multidistrict ESI protocol before a producing party could validate that its production was substantially complete. His August 19 order leaves them one, the sample of documents coded non-responsive. His reasons for declining them give litigators a test for any validation provision they are asked to sign.
What Happened
The parties had submitted a second proposed ESI protocol after the court gave guidance on the first. Their disputes had narrowed to a single question, how a producing party must validate that its productions are substantially complete.
Plaintiffs proposed sampling in three strata. One went to documents coded non-responsive, another to documents coded responsive, and a third to documents pulled out of the review universe before merits review ever began, which the court called the Excluded Set. Defendants accepted the first stratum, objecting to the other two.
The Court's Analysis
Citing Rule 26(g)(1)(A), the court began from the requirement that a party responding to a discovery request certify that its response is "complete and correct as of the time it is made". Completeness is where the risk sits, the court reasoned, because outside of forgery a document production is not false. Translating that requirement into statistical terms, the court stated that a producing party has to confirm its false-negative rate falls within acceptable limits.
The second stratum failed on that reasoning. Confirming that a responsive coding call was correct "reveals only the production's false-positive rate", in the court's words, a proxy for overproduction rather than for anything missing. Because the Federal Rules do not prohibit overproduction, the court declined to make defendants pay to re-review documents they had already coded and intended to produce, privilege claims aside.
Material excluded before review, the court noted, could hold responsive documents erroneously left out, so sampling that material could reveal a false negative. Sampling the Excluded Set still failed, because the protocol already required a producing party using search terms to sample the non-hits before merits review and confirm they were non-responsive. Validating that material again at the end would, in the court's view, be duplicative and unnecessary. Plaintiffs replied that the two efforts "answer different questions at different points in time", the first testing the search terms before review, the second testing the whole process through to production. The court rejected that distinction, reasoning that the Excluded Set is carved out before merits review begins, never passing through the coding decisions, review workflow, or production choices the later sampling was said to test.
Under the order, a producing party escapes the end-stage sampling only where it has already statistically validated that the documents excluded from merits review do not contain responsive material.
The court also declined to treat plaintiffs' approach as settled practice, collecting three other cases that resolved the question different ways. One of them required the very sampling plaintiffs proposed in this litigation.
Why It Matters
Judge a validation provision by the error it is capable of catching. Sampling that cannot identify a document missing from the production does not advance a Rule 26(g) certification, however rigorous it looks on the page.
Then measure it against what the protocol already does. The court declined to require the Excluded Set sampling because a producing party using search terms already had to validate the non-hits before merits review. Coverage at that earlier stage is not automatic. This protocol committed to no sampling method for technology assisted review, leaving it to a later meet and confer. A party that negotiated no such sampling into its collection parameters has nothing to point at when the substantial-completeness demand arrives at the end. This order hands the requesting side that argument.
The court left the drafting to the parties, quoting EEOC v. George Wash. Univ. for the proposition that courts are "not normally in the business of dictating to parties the process they should use when responding to discovery". It ordered them to submit a third proposal conforming the protocol to this ruling. Drafting that language remains their job, against the conclusions this order sets out.
The full opinion is available as a PDF.
