TAR in eDiscovery, short for Technology Assisted Review, is a method of sorting large volumes of electronic documents during litigation using machine learning instead of relying entirely on human reviewers. The software is trained to recognize which files are likely relevant to the case and which are not, then ranks the entire document population by that likelihood. One comparison cited in the research found that reviewing 130,000 documents took 27 days manually, 10 days with TAR 1.0, and roughly five days with newer AI tools. Federal courts have accepted the approach for more than a decade, and in most large matters it is now the default rather than the exception.
How TAR Actually Works
Once a document collection is loaded into a review platform, the software assigns a numerical relevance score to every file. A document scoring 95 out of 100 is almost certainly responsive. A document scoring 12 almost certainly is not. The interesting work happens in the middle, where human judgment still matters.
The algorithm builds those scores by analyzing text content, metadata, and the coding patterns from documents reviewers have already looked at. It identifies which combinations of terms, phrases, senders, recipients, and date ranges correlate with relevance. Every time a reviewer codes another document, the model updates and reranks the pool.
The output is a ranked list of the full population. You then set a cutoff score. Documents above it get produced or go through a final human quality check. Documents below it are presumed non-relevant and excluded, though a sample from that discard pile is tested later to confirm the cutoff was reasonable. Choosing where to draw that line is one of the most consequential decisions in the whole process, because it directly controls how much relevant material you find.
Over the course of a review, the files being surfaced to reviewers become progressively less relevant. That declining yield is one of the signals used to decide when to stop.
TAR 1.0 and TAR 2.0
Two approaches share the TAR label but handle learning differently.
TAR 1.0 separates training from review. A subject matter expert manually codes a sample of documents, and the algorithm trains on those decisions over several rounds. After each round, the expert checks a control set to gauge accuracy. Once the model hits a target recall rate, training stops and the review team works through whatever the algorithm ranked as likely relevant.
TAR 2.0, usually called Continuous Active Learning, collapses that distinction. The algorithm learns from every document a reviewer codes, continuously reranking the remaining pool and pushing the most likely relevant files to the top. There is no separate training phase and no fixed seed set. The model improves with every decision the review team makes. The court in Rio Tinto PLC v. Vale S.A. specifically noted that continuous active learning “eliminates issues about the seed set and stabilizing the TAR tool.”1Justia. Hyles v. City of New York et al
For most modern reviews, TAR 2.0 is the default. It needs less upfront time from senior attorneys, adapts better when the definition of relevance shifts during litigation, and surfaces important documents faster. TAR 1.0 still has a role where review criteria are narrow and well-defined from the outset, but the industry has largely moved on.
The Seed Set in TAR 1.0
In a TAR 1.0 workflow, the quality of the seed set matters enormously because the entire model builds on those early judgments. If the expert miscodes documents or the sample is unrepresentative, the algorithm inherits those blind spots. In Da Silva Moore v. Publicis Groupe, the parties agreed to a 95% confidence level with a margin of plus or minus two percent to draw a random sample of 2,399 documents as the initial seed set, and the court approved seven iterative training rounds with the understanding that more could be ordered if the model had not stabilized.2Justia. Da Silva Moore v. Publicis Groupe et al
TAR 2.0 largely sidesteps the seed set debate. Because the algorithm learns continuously from every reviewed document, there is no single batch of training data an opposing party can attack as biased or incomplete.
Validating the Results
Courts do not expect perfection, but they expect you to prove the review was reasonable. That proof is statistical. Four metrics do most of the work:
- Recall is the percentage of all truly relevant documents the system actually found. If the collection contains 10,000 relevant documents and your production includes 8,000 of them, recall is 80%. Courts care about this one most because it measures completeness.
- Precision is the percentage of documents flagged as relevant that actually are. High precision means a clean production without padding.
- Elusion is the rate of relevant documents hiding in the discard pile, measured by sampling the files the algorithm rejected and checking how many should have been produced.
- F-measure is the harmonic mean of recall and precision. Because the two pull in opposite directions, F-measure captures the overall balance.
Industry experience suggests a recall rate between 75% and 85% strikes a reasonable balance in most cases, though the facts can push the target higher or lower. No court has set a rigid floor, and the EDRM’s TAR guidelines state that “there is currently no black letter law or bright-line rule as to what constitutes a reasonable review.”
To measure elusion, you draw a random sample from the documents the algorithm classified as non-relevant and have human reviewers code it. Standard practice is a 95% confidence level with a confidence interval of plus or minus 5%, which sets the minimum sample size for statistical reliability.3EDRM. Draft TAR Protocol – Governing Production of Relevant Information Using Technology Assisted Review If the sample contains very few relevant documents, elusion is low and the review is defensible. If it turns up more than expected, you either lower the cutoff score and review more documents or run additional training rounds.
In TAR 1.0, a separate control set is drawn at the start, coded by the subject matter expert, and never fed back into training. It serves as an answer key the algorithm’s scores are compared against as training progresses.
Are Courts On Board With TAR?
Yes, and they have been since 2012. The first federal court to formally approve TAR was the Southern District of New York in Da Silva Moore v. Publicis Groupe, where Magistrate Judge Andrew Peck wrote that “computer-assisted review is an acceptable way to search for relevant ESI in appropriate cases” and told lawyers they “no longer have to worry about being the ‘first’ or ‘guinea pig’ for judicial acceptance.”2Justia. Da Silva Moore v. Publicis Groupe et al The district judge later affirmed, observing that manual review “is prone to human error and marred with inconsistencies.”4Justia. Da Silva Moore v. Publicis Groupe et al – Document 175
Three years later, Rio Tinto PLC v. Vale S.A. declared that “it is now black letter law that where the producing party wants to utilize TAR for document review, courts will permit it.”5Justia. Rio Tinto PLC v. Vale SA et al Then in Hyles v. City of New York (2016), Judge Peck stated plainly that “for most cases today, TAR is the best and most efficient search tool,” particularly when the methodology uses continuous active learning.1Justia. Hyles v. City of New York et al Those three decisions form the backbone of TAR’s legal legitimacy.
TAR also fits comfortably within Federal Rule of Civil Procedure 26(b)(1), which limits discovery to what is “proportional to the needs of the case” weighing cost against likely benefit.6Legal Information Institute. Federal Rules of Civil Procedure Rule 26 – Duty to Disclose; General Provisions Governing Discovery When a document population runs into the hundreds of thousands, purely manual review often fails that test, which makes automated alternatives not just reasonable but expected. The Rule 26(f) conference early in a case is the natural moment to raise your intended use of TAR and negotiate a protocol.
Can the Other Side Force You To Use TAR — or Stop You?
Neither, as a rule. The responding party chooses its own review methodology. In Hyles, the requesting side tried to force the City of New York to use TAR instead of keyword searching. Judge Peck denied the request, writing that “cooperation principles do not give the requesting party, or the Court, the power to force the responding party to use TAR.”1Justia. Hyles v. City of New York et al The same result followed in In re Viagra Products Liability Litigation and In re Mercedes-Benz Emissions Litigation, where the special master acknowledged TAR would likely be more efficient but still refused to order it.
The reverse is also true. Courts have declined to force parties off TAR and onto manual review or keyword searching. If you are producing documents and want to use TAR, courts will back you. If the other side objects, the burden falls on them to show your methodology is producing inadequate results, not merely that they would prefer something else.
Judge Peck added a forward-looking note in Hyles: “There may come a time when TAR is so widely used that it might be unreasonable for a party to decline to use TAR. We are not there yet.”1Justia. Hyles v. City of New York et al
Protecting Privileged Documents
TAR helps screen for privilege but does not replace a dedicated privilege review. Most workflows run a separate privilege analysis using natural language processing to identify attorney names, legal department domains, and contextual markers of legal advice. Flagged documents then go to senior attorneys who understand the nuances of attorney-client privilege and work product protection.
Mistakes still happen when you are processing hundreds of thousands of files. Federal Rule of Evidence 502(d) is the safety net. Under it, a federal court can order that any privileged document accidentally disclosed during litigation does not waive the privilege, and that protection extends to every other federal and state proceeding.7Legal Information Institute. Federal Rules of Evidence Rule 502 – Attorney-Client Privilege and Work Product; Limitations on Waiver The explanatory notes to the rule reference the “prohibitive costs” of record-by-record privilege review in large matters.
Getting a 502(d) order should be among the first things you do in any case involving TAR. The order enables clawback agreements, where parties return inadvertently produced privileged documents without the disclosure counting as a waiver. Most experienced eDiscovery practitioners treat it as non-negotiable.
What TAR Costs
TAR is almost always cheaper than manual review, but it is not cheap. The expenses break into several categories that are easy to underestimate if you only look at the software license.
Platform licensing in 2026 typically runs between $150 and $250 per user per month for standard tiers. Enterprise plans with advanced analytics can exceed $400 per user. In multi-party litigation with six or more attorneys on the platform, monthly costs can reach $1,500 or more before separate data processing fees are added. Many vendors charge additional fees for ingestion, processing, and production on top of the per-user price, so the license cost alone is rarely the total bill.
Human review still represents the largest expense. Even with TAR cutting the reviewable population significantly, someone has to code the training documents, review the highest-ranked files, handle privilege screening, and perform validation sampling. Contract attorney rates for document review work average around $23 per hour nationally, though billing rates for TAR-trained reviewers at major firms run considerably higher.
The math almost always favors TAR over the alternative. Manual review of a million-document collection at even modest hourly rates would cost several million dollars. TAR can reduce the number of documents needing human eyes by 70% to 90%, compressing both the timeline and the budget. The main eDiscovery platforms in 2026 — Relativity, DISCO, Everlaw, and Reveal among them — all offer continuous active learning features, though implementation quality varies and the platform choice affects reproducibility if you later need to show the court how the software works.
The Attorney’s Competence Duty
Running TAR does not let you delegate the thinking. The ABA’s Model Rules of Professional Conduct, in the comment to Rule 1.1, require lawyers to “keep abreast of changes in the law and its practice, including the benefits and risks associated with relevant technology.”8American Bar Association. Rule 1.1 Competence – Comment A majority of states have adopted that language or something close to it.
In practice, that means the supervising attorney needs to understand how the algorithm works, what the validation metrics mean, and where the process can go wrong. Handing the technical work to a project manager or vendor is fine. The legal judgment calls — relevance definitions, privilege decisions, cutoff thresholds — belong to the lawyer.