What is the difference between technology-assisted review (TAR 1.0) and continuous active learning (TAR 2.0)?

Technology-assisted review (TAR) uses machine learning to help legal teams classify documents during e-discovery — typically to separate relevant from non-relevant material in large collections. “TAR 1.0” and “TAR 2.0” describe two different workflows for training and applying that classifier. Both are recognized, defensible approaches; the right choice depends on the matter, the collection, and proportionality.

TAR 1.0: Train once, then classify

TAR 1.0 (often called predictive coding) follows a structured, front-loaded process:

• One or more subject-matter experts review a training set — sometimes a randomly drawn sample, sometimes a “seed set.”
• The system learns from those coding decisions to build a model.
• Reviewers validate the model against a control sample until it stabilizes.
• Once trained, the model is applied to the rest of the collection in a largely one-time pass.

Because training is separated from production review, TAR 1.0 depends heavily on assembling a representative training set up front and on consistent expert coding.

TAR 2.0: Continuous active learning

TAR 2.0, or continuous active learning (CAL), folds training into review itself:

• The model continuously retrains as reviewers code documents.
• It repeatedly surfaces the documents it currently predicts are most likely relevant (“active learning”).
• There is no fixed training phase — learning continues until the team has reviewed enough to meet its goals.

Practical differences

• Workflow: TAR 1.0 separates training from review; TAR 2.0 merges them.
• Adaptability: CAL adjusts as new documents and issues emerge, which helps with rolling or evolving collections.
• Expertise demands: TAR 1.0 leans on expert seed coding; CAL can incorporate decisions from a broader review team over time.
• Stopping point: TAR 1.0 stops when the model stabilizes; CAL stops based on richness and recall targets.

Why it matters

Courts in many U.S. jurisdictions have accepted TAR when parties act reasonably, transparently, and proportionally — principles reflected in the Federal Rules of Civil Procedure governing discovery. Whichever method you use, document your process, validation, and quality measures. Requirements and case law vary by jurisdiction (state courts, other countries), so confirm what applies to your matter.