- What Technology-Assisted Review Means for the CEDS Exam
- Core TAR Concepts You Must Master
- Predictive Coding Workflow: From Seed Set to Production
- TAR vs. Traditional Linear Review: What CEDS Tests
- Quality Control and Validation in TAR
- Legal and Ethical Dimensions of TAR
- How Domain 10 Connects to the Rest of the CEDS Exam
- A Domain-Specific Study Schedule for TAR
- Frequently Asked Questions
- Domain 10 covers Technology-Assisted Review and Predictive Coding - one of eleven domains tested across the CEDS exam's approximately 100 scenario-based...
- ACEDS does not publish domain weighting percentages, so treat TAR as equally testable alongside processing, production, and legal hold domains.
- The CEDS exam uses 4-option multiple-choice scenarios; TAR questions frequently present attorney or project manager decision points, not just tool definitions.
- You must understand TAR 1.0, TAR 2.0, continuous active learning, and validation protocols - not just surface-level terminology.
What Technology-Assisted Review Means for the CEDS Exam
Technology-Assisted Review - universally abbreviated as TAR - represents one of the most technically demanding areas in the CEDS certification. Domain 10 is formally titled Technology-Assisted Review and Predictive Coding, and it reflects the real-world shift in e-discovery practice from purely manual document review toward machine-learning-driven prioritization and classification.
For candidates preparing for the CEDS exam, it is important to understand that ACEDS designs questions around practical application, not vocabulary lists. The exam's approximately 100 multiple-choice scenario-based questions - each with four answer options - are built to test whether you can make sound professional decisions under realistic conditions. A TAR question might describe a case with 2.5 million documents, a tight production deadline, and a disagreement between outside counsel and the review vendor about validation methodology. Your job is to identify the most defensible course of action, not simply define a term.
This is why Domain 10 demands more than memorizing what "predictive coding" means. ACEDS, governed by BARBRI and adhering to ICE credentialing standards, expects certified specialists to demonstrate the judgment of an experienced e-discovery professional. The vendor-neutral nature of the certification also means you are tested on process and principle - not on any particular software platform's user interface or proprietary features.
Core TAR Concepts You Must Master
The Two Major TAR Protocols
The CEDS exam expects candidates to distinguish between TAR 1.0 and TAR 2.0 with precision. These are not simply version upgrades - they represent fundamentally different approaches to how a predictive coding model is trained and how reviewer input is incorporated.
TAR 1.0: Simple Active Learning (SAL)
In TAR 1.0, a small group of subject-matter experts reviews a carefully constructed seed set of documents. Those relevance decisions are fed into the machine-learning model, which then scores and ranks the remaining collection. The model is typically trained once or a limited number of times before being applied to the full document universe.
- Seed set quality is critical - garbage in, garbage out applies directly here
- Defensibility depends heavily on the quality and consistency of the senior reviewers training the model
- Validation occurs after the model is applied, often through statistical sampling of the low-scoring documents
- Courts have accepted TAR 1.0 workflows, but opposing counsel may scrutinize seed set composition
TAR 2.0: Continuous Active Learning (CAL)
TAR 2.0 uses continuous active learning, meaning the model updates dynamically as reviewers code documents throughout the entire review process. There is no fixed seed set - the model learns from every reviewer decision made during the review, continuously re-ranking documents to surface the most likely relevant content next.
- Scales well to large, complex document sets with heterogeneous content
- Allows any reviewer's decisions to contribute to model training, though quality control becomes more complex
- Stopping criteria and recall estimation are key areas of professional judgment
- Increasingly preferred in large-scale commercial litigation and regulatory matters
Recall, Precision, and the F1 Score
These three metrics are the statistical backbone of any TAR validation discussion, and CEDS candidates consistently underestimate how deeply the exam tests them. Recall measures the proportion of truly relevant documents that the TAR process identified. Precision measures how many of the documents flagged as relevant actually are relevant. The F1 score is the harmonic mean of recall and precision, used when you need a single composite measure of model performance.
For the CEDS exam, you need to understand not just the definitions but the tradeoffs. Optimizing purely for recall means you retrieve more relevant documents but potentially review a larger volume of non-relevant material. Optimizing purely for precision means you review a smaller, cleaner set but risk leaving relevant documents in the non-reviewed population. Exam scenarios will present fact patterns where you must recommend an appropriate balance based on case-specific factors such as stakes, proportionality, and the nature of the claims.
Elusion Rate and Null Set Sampling
The elusion rate - the percentage of relevant documents estimated to remain in the non-reviewed or low-scoring population - is a critical TAR validation concept. CEDS scenarios may ask you to evaluate whether a proposed TAR workflow is sufficiently complete based on elusion estimates drawn from statistical samples of the null set (documents the model scored as non-relevant and that were not reviewed).
Predictive Coding Workflow: From Seed Set to Production
The CEDS exam does not test isolated concepts in a vacuum. Domain 10 questions frequently require you to trace a TAR workflow from its starting point through to a defensible production. Understanding each stage - and how failures at one stage cascade into problems downstream - is essential exam preparation.
- Collection and ingestion: TAR workflows begin after collection and processing. The quality of metadata preservation and deduplication at the Domain 3 and Domain 4 stages directly affects the integrity of the document universe presented to the TAR model.
- Pre-processing and culling: Before training begins, teams typically apply date range filters, keyword culling, or domain exclusions to reduce the document population to a manageable and relevant universe. The decisions made here can significantly affect model performance and must be documented.
- Control set creation: A statistically valid control set - documents coded by humans with subject matter expertise - is used to measure the model's performance throughout the review. This is separate from the seed set used to train the model in TAR 1.0 workflows.
- Training and review cycles: In TAR 1.0, training occurs in discrete rounds. In TAR 2.0, it is continuous. Either way, the CEDS candidate must understand how reviewer decisions are quality-controlled during this phase.
- Validation and stopping criteria: How does a team know when TAR is complete? Validation protocols typically combine recall estimation through null set sampling with control set comparisons. Stopping criteria should be agreed upon - and ideally disclosed to opposing counsel - before review begins.
- Production: Documents cleared through the TAR workflow flow into Domain 6 production processes, including format decisions, privilege logging, and load file preparation.
Key Takeaway
CEDS scenarios about TAR frequently involve a decision point at the validation or stopping stage. Know how to defend a TAR protocol's completeness using recall estimates and null set sampling results - this is the type of judgment the exam rewards.
TAR vs. Traditional Linear Review: What CEDS Tests
| Dimension | Traditional Linear Review | Technology-Assisted Review |
|---|---|---|
| Document prioritization | Sequential; no ranking by likely relevance | Machine-scored; high-probability relevant documents reviewed first |
| Scalability | Costs scale directly with document volume | Fixed technology cost; reviewer hours reduced significantly on large sets |
| Consistency | Subject to individual reviewer fatigue and variation | Model applies consistent scoring criteria; human inconsistency still affects training |
| Validation | Quality control through random sample checks | Statistical validation through recall/precision metrics and null set sampling |
| Proportionality argument | Harder to justify for massive collections | Directly supports Federal Rule 26(b)(1) proportionality analysis |
| Transparency to opposing counsel | Generally simpler to explain | Requires disclosure of protocol; courts vary on required level of transparency |
| CEDS exam focus | Appears in Domain 5 (Review and Analysis) primarily | Central focus of Domain 10; also connects to Domain 5 and Domain 6 |
Quality Control and Validation in TAR
Quality control in a TAR workflow is substantively different from QC in traditional review, and the CEDS exam expects candidates to know both. In traditional review, QC typically involves a senior attorney reviewing a random sample of documents coded by junior reviewers. In TAR, QC operates at multiple levels simultaneously.
Inter-Rater Reliability
During the training phase, it is critical that the reviewers coding documents for model training apply relevance criteria consistently. Inter-rater reliability - the degree to which two independent reviewers would code the same document the same way - is a measurable quality indicator. Low inter-rater reliability during training will degrade model performance. CEDS scenarios may ask you to identify the appropriate response when QC data reveals significant disagreement among reviewers training the model.
Iterative Testing Against the Control Set
A well-designed TAR protocol tests the model's performance against the control set at regular intervals during training. If the model's recall and precision against the control set plateau or decline after additional training rounds, that signals that further training iterations may not improve performance - a key input to the stopping decision.
Legal and Ethical Dimensions of TAR
Court Acceptance and Seminal Cases
The CEDS exam does not require you to cite case names from memory, but you must understand the legal landscape in which TAR operates. Key judicial decisions have established that TAR is an acceptable and in some circumstances preferable method of review, provided the methodology is transparent and validated. The concept of cooperation between parties regarding TAR protocols - including disclosure of the seed set composition in some jurisdictions - is a tested topic. Know that courts have differed on how much of a producing party's TAR methodology must be disclosed to opposing counsel.
Privilege Review in TAR Workflows
One area where Domain 10 intersects with professional responsibility is privilege review. TAR models can misclassify privileged documents if the training data does not adequately represent privileged content patterns. CEDS candidates must understand how to structure a TAR workflow that includes a separate, human-driven privilege review layer rather than relying on the model alone to identify attorney-client communications or work product.
This connects directly to the broader review and analysis concepts in CEDS Domain 9: Cross-Border Discovery Study Guide 2026, where data protection laws in certain jurisdictions may restrict how TAR models can be trained on personal data - adding another layer of complexity for international matters.
Proportionality and Rule 26
TAR's most powerful legal justification is proportionality. Under Federal Rule of Civil Procedure 26(b)(1), discovery must be proportional to the needs of the case. TAR directly supports proportionality arguments because it can dramatically reduce review costs on large collections while maintaining or exceeding the recall rates of linear review. CEDS candidates should be prepared to articulate this argument in a scenario context - for example, when advising a client on whether to propose TAR to opposing counsel as the review methodology.
How Domain 10 Connects to the Rest of the CEDS Exam
No CEDS domain exists in isolation. TAR is deeply connected to several other domains you will be tested on, and understanding those connections helps you answer cross-domain scenarios correctly.
Domain 4: Processing
Processing decisions directly affect TAR performance. Deduplication approach (global vs. custodial), near-duplicate identification, and email threading all shape the document universe the TAR model trains on. A CEDS candidate must understand that processing errors - such as extracting text incorrectly from PDFs - can degrade model quality in ways that are difficult to detect until validation.
- Text extraction quality is foundational to TAR model accuracy
- Near-duplicate and email threading decisions affect which documents are included in the training population
Domain 5: Review and Analysis
Domain 5 and Domain 10 overlap significantly. The CEDS study material, which spans approximately 250 pages and covers the full EDRM model, treats TAR as an advanced review methodology that operates within the broader review and analysis framework. Candidates should understand when TAR is appropriate versus when targeted human review or other analytics tools are more suitable.
- Analytics tools such as clustering, concept searching, and email threading are complements to - not substitutes for - a formal TAR protocol
- Supervised machine learning (TAR) is distinct from unsupervised analytics
Domain 11: Project Management and Budgeting
TAR decisions have direct cost implications. A CEDS specialist must be able to model the cost-benefit tradeoff of implementing TAR versus linear review for a given document population. Domain 11 tests project management skills that apply directly to TAR workflow design, including vendor selection, timeline management, and budget forecasting for iterative training rounds.
- Technology licensing costs versus reviewer hour reductions
- Scheduling training rounds and validation cycles into the project timeline
For a broader view of how international data considerations interact with TAR workflows, review our companion guide covering CEDS Domain 9: Cross-Border Discovery Study Guide 2026, which addresses the GDPR and other legal frameworks that can affect how TAR is deployed on cross-border matters.
A Domain-Specific Study Schedule for TAR
Because ACEDS does not publish domain weighting percentages, it is tempting to distribute study time equally across all eleven domains. A more strategic approach weights your preparation toward domains that are conceptually dense, practically significant, and likely to appear in multi-domain scenarios. Domain 10 qualifies on all three counts.
Foundations and Terminology
- Master TAR 1.0 vs. TAR 2.0 distinctions
- Study recall, precision, F1 score, and elusion rate definitions and applications
- Read ACEDS study guide sections on review technology and processing (connects Domain 4 and Domain 10)
- Complete a set of TAR-focused practice scenarios at the CEDS practice exam platform
Workflow, Validation, and Legal Framework
- Map the full TAR workflow from collection through production
- Study null set sampling and stopping criteria methodologies in depth
- Review key legal and judicial concepts: proportionality, disclosure obligations, privilege review protocols
- Use the Feynman technique on the seed set construction process - explain it out loud as if briefing a client
Cross-Domain Integration and Exam Simulation
- Practice multi-domain scenarios that combine Domain 10 with Domain 5, Domain 6, and Domain 11
- Review international TAR considerations (connects to Domain 9)
- Take a timed full-length practice test and analyze any TAR questions you miss
- Revisit the ACEDS study guide's EDRM coverage with Domain 10 in mind
Frequently Asked Questions
No. ACEDS designs the CEDS as a vendor-neutral certification. Domain 10 tests your understanding of TAR principles, workflows, validation methodologies, and legal defensibility - not the features or interfaces of any particular e-discovery platform. You should understand how TAR works conceptually so you can apply that knowledge regardless of which tool is in use.
ACEDS does not publish domain weighting percentages. The exam covers eleven subject areas selected by a global taskforce and validated through an ACEDS survey. Candidates should prepare all eleven domains with seriousness. That said, Domain 10 is conceptually complex and frequently appears in multi-domain scenarios, which makes deep preparation particularly valuable.
TAR specifically refers to supervised machine learning workflows where human reviewer decisions train a model to predict relevance across an entire document population. Other analytics tools - such as concept clustering, email threading, or keyword search - are unsupervised or rule-based. For the CEDS exam, understanding this distinction is important because questions may present scenarios where you must recommend the appropriate tool for a given situation rather than defaulting to TAR in every context.
Both. The CEDS is designed for practitioners who sit at the intersection of technology and law. Domain 10 questions may present technical decisions (such as how to respond when model recall plateaus) or legal and ethical decisions (such as how much of the TAR protocol to disclose to opposing counsel). Treating TAR as purely technical or purely legal will leave gaps in your preparation.
The CEDS requires 40 qualifying credits from professional experience (up to 20 credits), training (up to 25 credits), and education (up to 15 credits), plus two professional references. Candidates with hands-on TAR project experience can count that work toward their professional experience credits, making Domain 10 one area where real-world practice directly supports both the prerequisite application and exam readiness.
Ready to Start Practicing?
Domain 10 is one of the most scenario-intensive areas of the CEDS exam. Reinforce your TAR knowledge with realistic, scenario-based practice questions built specifically for CEDS candidates - covering Technology-Assisted Review, predictive coding workflows, validation protocols, and all eleven exam domains.
Start Free Practice Test