The Vrenberg Method
The science behind Vrenberg.
Vrenberg is built on retrieval practice, spaced repetition, interleaving, and desirable difficulties. This page shows how each principle maps to a feature, how questions move through the generation pipeline, and how the canon is calibrated against released NCBE material.
The research
Four principles. Each one mapped to a feature.
Passive review is the weak link
Re-reading notes and watching lectures are forms of passive review. Dunlosky et al. (2013) surveyed decades of learning research and found that active techniques outperform passive review for durable retention. The Vrenberg Method treats that finding as a design constraint, not a slogan.
Dunlosky, J. et al. (2013). "Strengthening the Student Toolbox." Psychological Science in the Public Interest, 14(1), 4-58.
Retrieval practice
The act of recalling information strengthens memory far more than re-studying it. Roediger and Karpicke (2006) demonstrated that students who practiced retrieval retained significantly more material over time than those who re-read. Vrenberg applies this directly: every question set is retrieval practice, and every wrong answer becomes another retrieval opportunity.
Roediger, H.L. and Karpicke, J.D. (2006). "Test-Enhanced Learning." Psychological Science, 17(3), 249-255.
Spaced repetition
Distributing practice over time produces stronger retention than massing it into a single session. Ebbinghaus first documented the forgetting curve in 1885; Cepeda et al. (2006) confirmed in a large meta-analysis that spaced practice reliably outperforms massed practice across domains. Vrenberg's review queue schedules each rule at expanding intervals calibrated to individual performance.
Cepeda, N.J. et al. (2006). "Distributed Practice in Verbal Recall Tasks." Psychological Bulletin, 132(3), 354-380.
Ebbinghaus, H. (1885). Uber das Gedachtnis [Memory: A Contribution to Experimental Psychology].
Interleaving
Mixing different problem types within a session forces deeper discrimination between similar concepts - exactly what the MBE tests. Rather than drilling one subject at a time, Vrenberg interleaves rules across subjects, mimicking the exam's own structure and making the obvious answer less obvious.
Desirable difficulties
Bjork's research on desirable difficulties supports a simple idea: struggle is not the enemy of learning when it is structured and recoverable. Vrenberg uses that principle in timed drills, trap-radar sets, and missed-rule reviews that make students work hard enough to remember what they just learned.
The pipeline
Five stages. Every question. No exceptions.
Ingest
Black-letter law sources are ingested into the canon: the 391+ tested rules the platform is built to model and surface.
Generate
Questions are generated on demand - per rule, per difficulty - by the V4 engine, which writes each fact pattern against the doctrine it is meant to test.
Fact-check
An automated legal-accuracy check runs on every question: a separate solver must independently work the problem and agree on the answer. If it does not, the question is rejected.
Gate
Surviving questions pass through style and difficulty gates that enforce MBE conventions - stem structure, option construction, and the intended difficulty band.
Bank
Only then is a question banked and served. Every banked question carries its rule citation and a seven-part explanation of why the credited answer is right and each distractor is wrong.
The canon
Proprietary terms, used consistently.
V4 engine
The generation engine that writes questions per rule and per difficulty.
Rule graph
The structure that links each rule to the topics, traps, and review paths around it.
Mastery modeling
The scoring layer that turns performance into the next best action.
NCBE-calibrated
The benchmark standard used for question shape, difficulty, and answer style.
Calibration
Measured in public, not asserted in marketing.
The primary calibration instrument is the public Real-or-Vrenberg challenge: a blind test that mixes official released NCBE sample questions with output from our engine and asks visitors to tell them apart. Its population statistics - total attempts and average accuracy - are computed live from every submission and shown on the page. If people could reliably distinguish our questions from the real exam, that number would show it.
During development we also ran blind discrimination testing with frontier AI models, which reached coin-flip discrimination between our questions and official samples. A formal human-rater study is planned, and the results will be published either way - including if they embarrass us.
Error accountability
When something is wrong.
No pipeline is perfect, and a legally wrong practice question is worse than no question. If you believe a question, credited answer, or explanation is incorrect, report it to support@vrenberg.com. Reports are reviewed against the cited rule, and confirmed errors are corrected or pulled from the bank.
Test the output yourself.
Real questions in the challenge are official sample MBE questions publicly released by the National Conference of Bar Examiners (ncbex.org). NCBE is not affiliated with and does not endorse Vrenberg.