But For. The Report

Formats · The working guide to every question type

Every question format on the NextGen UBE, and what each one rewards.

Five question shapes carry the whole exam. One of them, the six-option select-two, is the one candidates understand least and the one where points are most quietly left behind. This is the working guide to all of them, written for someone actually preparing.

The shape of the exam

The NextGen UBE runs nine hours over a day and a half: three three-hour sessions on your own laptop through NCBE's secure platform, reported on a 500–750 scale with passing lines set by each jurisdiction. Standalone multiple choice takes roughly forty percent of exam time, in two formats. Integrated question sets take just over a quarter. Performance tasks carry roughly a third of the score, in two species.

Underneath sit the skills the examiners name openly: legal research, legal writing, issue spotting and analysis, investigation and evaluation, counseling, negotiation and dispute resolution, and client relationship management. The new question shapes exist to reach them. Each format is a container for a lawyering operation, and the operation is what earns the point.

One structural change cuts across all of them. The exam regularly hands you a statute, a rule, or an edited opinion and expects you to apply that text, including where it departs from the rule you memorized, and it can test subjects outside the memorized syllabus by supplying all the law needed. Family law and trusts and estates enter in July 2028, largely through those resource-provided formats. The MBE reflex, recognize the doctrine and retrieve the majority rule, can now cost you points.

Select-one: the familiar shape, doing unfamiliar work

One stem, four options, exactly one credited. Early administrations deliberately resemble MBE-style questions, and the format variety grows from there. Select-one is not obsolete; it has become a container for a wider set of tasks.

What it still tests

The same three operations, under a clock: filtering the facts that carry legal weight from the facts that merely glitter, recognizing the doctrine a pattern summons before the options frame it, and mapping a rule onto facts precisely enough to survive an option that is almost right.

The shape of a well-built one, from an item of ours already public as a sample: intending only to frighten a rival, a man hurls a rock inches past the rival's head, and it breaks the nose of a bystander he never saw. Liable for battery? Yes, on transferred intent operating twice over, assault intent toward one person and battery result on another. Two wrong options say no for reasons that feel entirely fair, which is why they are there. The distance between the key and the best distractor is usually one element, one party, or one word in the call.

Where it departs from the MBE playbook

Three departures matter. The call is increasingly a lawyering move rather than a legal conclusion: what would you research next, which statement represents what the court actually decided, which fact most undermines the defense. Provided law overrides memory, and a stem that quotes a statute quotes it because the micro-detail carries the key. And you are given a role, so the question is what that lawyer should do.

The adjustment is worth building into muscle: read the call before the stem, every time. When one fact pattern can support a doctrinal, research, or counseling question, the call tells you which operation you are performing.

Select-two: six options, exactly two keys, real partial credit

This is the format with no legacy archive behind it, and where the most points are quietly left on the table. Six lettered options, A through F. Exactly two are credited. You select exactly two, and the partial credit is real: one correct selection earns points, and a blank earns nothing.

The scoring, plainly

On our bench the select-twos are scored on the scale the exam's partial credit describes, two points per item, awarded 2 / 1 / 0. Both keys, two points. One key and one miss, one point. Two misses, zero. Unanswered, zero. The last clause carries the whole strategy.

What a blank actually costs

The arithmetic is small enough to check by hand. Six options taken two at a time yield fifteen pairs. Exactly one is the credited pair. Eight contain exactly one key, since each key can be paired with each of the four non-keys. The remaining six pair two non-keys and score zero. Everything else follows from those counts.

Exhibit · Select-two · Expected points on the 2 / 1 / 0 scaleIllustrative arithmetic
Expected points on a select-two item, by how much the field has been narrowed
Where you are on the item Pairs still live Expected points Of available credit
You leave it blank none selected 0.00 0%
Pure guess, nothing eliminated 15 0.67 33%
Two options eliminated, both keys still standing 6 1.00 50%
One key identified, second pick a guess among five 5 1.20 60%
Both keys identified 1 2.00 100%

Derivations, so you can check them. Pure guess: one pair scores 2 and eight pairs score 1, so 10 points spread across 15 equally likely pairs, or 0.67. Field of four containing both keys: one pair scores 2, four score 1, one scores 0, so 6 points across 6 pairs, or 1.00. One key known: of the five options left, one completes the pair for 2 points and four leave you at 1, so (2 + 4) ÷ 5, or 1.20. This is arithmetic on the raw 2 / 1 / 0 scale, not a model of scaled scores; the conversion from raw performance to the 500–750 report is a psychometric matter.

Read the first two rows together. A blank is the only choice on a select-two certain to be worth zero, and even a coin-flip pair returns about a third of the item's credit on average. There is no state of ignorance deep enough to make abstention rational.

A second, less obvious reading. A blind guess on a four-option select-one returns a quarter of that item's credit; a blind guess on a six-option select-two returns a third, despite the larger field, because this format pays for being half right and the four-option format never does. It is not a harder select-one, it is a format that meters partial knowledge, in both directions.

How the field of six is built

Six options do not feel generous in the room, because the four wrong ones are not filler. Each incorrect option is a specific, named failure mode, and knowing the taxonomy makes them visible under time. These are the families that recur in our own bank.

Because two options are credited, the call has a shape, and reading it is the fastest route into the item. Many select-twos are two-key locks: the credited pair is the two halves of one conjunctive test. Unconscionability is the cleanest example, needing oppressive terms and an oppressive process, so a well-built item plants a key for each prong and surrounds them with price complaints and regret. Others want two independent supports for one position, in which case the second key should look nothing like the first.

What partial credit changes about how you work the item

Because the item pays for being half right, the goal is no longer to find the answer. It is to convert whatever you know into the most credit that knowledge supports, and then leave.

Four moves, in order.

  1. Read the call and name the lock. Two halves of one test, or two independent supports? That decision narrows the field before you have evaluated a single option.
  2. Bank the key you are sure of. Expected value climbs further between "nothing" and "one key" than at any other step, from 0.67 to 1.20. Certainty about one option beats hesitation about two.
  3. Eliminate against the taxonomy, then choose. Score each remaining option as inert, adjacent, one element short, inverted, or merely sympathetic. Whatever survives is your second selection.
  4. Leave with the point. Our select-twos carry a time target of around two and a half minutes. When the second key will not resolve, take the one point and go: the item you never reach scores like a blank.

Integrated question sets: one matter, unfolding

This family is genuinely new. One client matter, one role, built on realistic documents: interview notes, a lease, a police report, a deposition excerpt, sometimes an excerpt of statute or an edited opinion. The set asks a mixed sequence of multiple choice, short answers, and medium-length written analysis, with facts revealed progressively as the matter develops. Some sets are doctrinal; others center on drafting or editing a document, or on counseling a client whose stated objectives and constraints you are expected to actually use.

The shape of a set

An illustration, from a set of ours already public as a sample. A software founder pays a twelve thousand dollar deposit on a thirty thousand dollar venue contract, cancels seventy-five days out, and meets a forfeiture clause; a provided statute makes retention above fifteen percent presumptively a penalty, rebuttable by proof of actual damages. The set opens with a select-one on statutory application, then a short answer asking for two specific facts supporting the penalty position. Then the record grows: a demand letter arrives with a further invoice, and a select-two asks which facts support the statute's separate cause of action. Mediation produces a settlement proposal, and a component asks for drawbacks to this client given what she has told you. Then a voicemail surfaces from the venue's manager, the set pivots from contracts into evidence, and it closes with a paragraph of written analysis carrying four scoring elements.

Six components, one continuous matter. That is the family's whole design: the facts do not restart, they accumulate.

"Given the facts now known"

That phrase, or its equivalent, is the mechanic. Each component is answerable on the record as it stands at that moment, and later reveals never make an earlier key wrong. The discipline runs the other way too: when new information arrives, earlier answers lock, and there is no revising a component in light of a fact you did not have. On our bench the runner enforces that lock rather than leaving it to the honor system, because a set worked with the ending visible is not the exercise the exam is running.

What the structure rewards

Performance tasks: two species, one clock

Roughly a third of the score, in two formats that ask for very different things.

The standard task

It descends from the MPT: a file, a library of provided authorities, and one extended writing assignment, on a sixty-minute clock. No outside law is wanted. What matters more than candidates expect is that the narrowing instructions and formatting rules are themselves graded. The assigning memo tells you what to write, in what form, and what not to address, and each of those is a scoreable instruction, not a courtesy.

The legal-research task

This one is new, with a different architecture: four multiple-choice questions, a short answer, and a medium-length writing assignment, worked over a small library. What it tests is research judgment itself. Which authority is primary and which secondary. Which binds and which merely persuades. What a case actually held, as against what it said in passing. And whether your library even answers the client's question.

That last one surprises people. In one of our research tasks, four sources are supplied and the credited answer to the final multiple-choice component is that the library does not resolve the question the client actually presents: the controlling case expressly reserves his exact posture. Being able to say "my authorities do not reach this" is a graded skill here, and it is the last thing most candidates have ever been asked to say on a test.

The two-pane workspace

Both species are worked in a document workspace rather than on a question screen: the file on one side, the library on the other, your answer in its own region, a clock running. The consequence is that reading becomes navigation. You are moving between documents, not holding one fact pattern in your head, and every minute spent re-finding a paragraph you already read is a minute not spent writing. Mapping the library on the first pass, which source says what and where the operative sentence lives, buys back more time than reading faster does.

What the graders reward

Element-level scoring, with the elements published alongside the task. One of our standard tasks runs twenty points: two for instruction-following and format, then six, eight, and three across the three issues, and one for synthesis. Look at the shape of that. A tenth of the credit is decided before a word of analysis is graded, purely on whether you wrote the section you were told to write, organized it as instructed, and left out what you were told to leave out.

What rubrics pay for, consistently.

Constructed response: written answers, graded by element

Written work reaches you in two sizes: the roughly one-sentence answer field inside integrated sets and research tasks, and the paragraph-to-memo analysis that closes a set or a task. Both are graded against elements rather than impressions, and the difference between those two things is where most written points are lost.

The one-sentence field

Deceptively hard. The instruction is usually about one sentence, one answer per field, graded against a list of representative correct answers. A field is a slot, so packing two facts into one does not earn two credits and risks earning one. And "specific" is the operative word: "the venue behaved unreasonably" is not an answer, while "the retained deposit equals forty percent of the contract price, far above the statute's fifteen percent presumption threshold" is, because it maps onto an element a grader can find.

What element-level grading actually looks like

A real grade from our own live grader shows it better than a description could. The task: identify two specific facts supporting that a deposit retention is presumptively an unenforceable penalty. The first answer cited the forty percent figure against the statutory fifteen percent threshold and was credited, the grader quoting the examinee's own words back as its evidence. The second argued that the venue made no effort to rebook, so it could not claim mitigation offsets. It reads well. It scored nothing, because the representative answers credit the seventy-five day notice, the all-cancellations forfeiture clause, or the forty percent figure, and this one asserted a fact about the venue's later conduct rather than the timing of the cancellation. One of two.

That gap, between an answer that sounds right and an answer that matches an element, is the whole of rubric grading. It is why a grader that flatters you is a liability: an encouraging score teaches you nothing about the sentence that failed.

Writing to the rubric

Length is not the variable; elements are, and a tight paragraph hitting four elements outscores a page hitting two. State the standard in the provided text's own terms. Quote the record, since specific facts are what elements are built from. Obey the exclusions exactly. And reach a conclusion, even a hedged one, because a supported prediction usually carries a point of its own and an unfinished analysis reliably does not.

Which format matters most

This page deliberately declines to rank them. The share figures are the honest guide: multiple choice takes the most time, performance tasks carry the most score, and the integrated sets are where the exam's genuinely new skills live. Whichever administration you are registered for, all three are on it, and skills transfer best to the conditions they were practiced in. That is the entire argument for working each format in its own grammar rather than translating from an exam that no longer exists.

Where this page comes from

Every structural claim above about the exam traces to NCBE's published materials and official statements, and where a figure describes our own materials rather than the exam, it says so. The scoring exhibit is arithmetic on the 2 / 1 / 0 scale, presented as arithmetic and not as a prediction of scaled scores. Illustrations are drawn from our own original items in the fictional State of Meridian; no real exam question appears here, and none ever will.

But For is an independent publication and study product of Ink Forms Rivers, d/b/a But For. We are not affiliated with, sponsored by, or endorsed by the National Conference of Bar Examiners or any bar examining authority; "NextGen UBE" is referenced solely to identify the examination. We publish examinee experience, never exam content.

support@butfor.ai reaches a human who reads every message, including about anything on this page you think is wrong.

The bench was built to exactly what this page describes: 534 original questions across both multiple-choice formats, 318 select-one and 216 select-two scored with real partial credit, all twelve integrated sets under an enforced lock, and both performance-task formats on the real clock, with rubric-transparent grading on written answers. The three-question calibration is free, the grader will grade an answer you write, and the editorial standard is published so you can hold us to it. Founding access is there when the evidence convinces you.