Resolving TAR and Search-Term Disputes With a Neutral

Few things stall a complex case faster than a search dispute. The volume of electronically stored information has made TAR and search-term disputes a routine feature of modern litigation, and they have a way of metastasizing — one side's keyword list or technology-assisted review (TAR) workflow becomes the other side's grievance, and a meet-and-confer about culling turns into a motion about adequacy. A neutral who is fluent in both the information-retrieval science and the discovery rules can resolve these fights at the front end, before they harden into a production impasse.

Why search becomes a fight

Keyword search and technology-assisted review are not optional luxuries; they are necessities. No party can manually review millions of documents, so culling by search term or by a trained predictive-coding model is how a producing party makes discovery proportional at all. The trouble is that every culling choice is contestable. The requesting party suspects the terms are too narrow or the model under-trained; the producing party calls the demands a fishing expedition. What looks like a dispute over documents is usually a dispute over how the documents were found — these are methodology fights dressed up as discovery motions.

What the dispute actually turns on

Strip away the rhetoric and most of these fights reduce to a short list of technical questions. How complete is the result — the recall rate — and how much junk does it sweep in — the precision? How was the result validated: by what sampling method, against what reviewed control set, to what confidence level and margin of error? For a TAR workflow, was the seed set representative, and was the model's training documented and stable? For keyword culling, who proposed the search-term list, were the terms tested against the data, and were hit counts and false-positive rates shared? These are questions of statistics and information retrieval. They do not resolve cleanly through dueling expert declarations and a half-day hearing, because the court is being asked to grade competing methodologies it has no practical way to test from the bench.

Why courts struggle with TAR and search-term disputes

Courts have accepted TAR for well over a decade. Da Silva Moore v. Publicis Groupe (S.D.N.Y. 2012) is generally cited as the first judicial opinion approving the use of technology-assisted review, and Rio Tinto PLC v. Vale S.A. (S.D.N.Y. 2015) confirmed, in Judge Peck's words, that TAR is an acceptable way to search for relevant ESI. But approving the tool is not the same as supervising its execution. The recall targets, the sampling plan, the seed-set arguments — these are technical and iterative, and a busy court cannot easily babysit a validation protocol across weeks of back-and-forth. Meanwhile the governing standards point only to a destination, not a method: proportionality under Federal Rule of Civil Procedure 26(b)(1) asks whether the effort fits the case, and counsel's certification under Rule 26(g) puts a signature behind the reasonableness of the search. Neither rule tells the parties what recall rate or sampling method satisfies it. That gap is exactly where disputes live.

What a neutral does

A neutral fills that gap by managing the method rather than refereeing its aftermath. In practice that means setting a defensible validation protocol the parties can both stand behind — defining how the result will be sampled and measured before anyone runs it. It means mediating the recall and sampling targets so they are negotiated to a number, not litigated to a stalemate. It means evaluating seed-set and training disputes on their technical merits, and giving the parties a documented methodology they can live with instead of reopening the same argument at every production. More than a decade ago, in E-Discovery Mediation & the Art of Keyword Search (Cardozo Journal of Conflict Resolution, 2011), Daniel B. Garrie and Edwin A. Machuca argued that keyword search is necessary but costly and inefficient, and that the workable answer to both problems is a mediator, neutral, or special master — one who ideally has a solid background in both technology and law. The point holds with equal force for TAR: the value of the neutral is the rare ability to sit credibly with the engineers and the lawyers at the same table.

The procedural vehicles

The neutral's authority can be sourced several ways. A court can appoint a master under Federal Rule of Civil Procedure 53 to oversee the search and validation process. In California, the parties or the court can use a reference under Code of Civil Procedure sections 638 and 639. And in many matters the cleanest route is a stipulated ESI mediation clause — built into the protocol at the outset — that sends any search-methodology dispute to a named neutral before it becomes a motion.

Practical takeaways

Three habits keep these disputes from derailing a case. First, agree on the search protocol and the validation method up front, in writing, while the relationship is still cooperative — recall targets and sampling plans are far easier to negotiate before either side has a result to defend. Second, bring in a neutral early, when the question is still how to search rather than who failed. Third, document the methodology as you go: the terms tested, the seed set, the training decisions, the sampling results. A well-papered method is its own defense, and it is what lets a producing party answer a Rule 26(g) challenge with a record instead of an argument.

This article is general information only and is not legal advice. Engagement of Daniel Garrie as a neutral is administered exclusively through JAMS. Nothing here is a solicitation.

Case citations are illustrative and should be independently confirmed before use.