dh2303

Joined: Nov 2025
Subscription: Core

Admissions profile

LSAT

Not provided Goal score: 180

CAS GPA

Not provided

1L START YEAR

2026

Applications

Arizona State

In process

Discussions

LSAT · New feature: AI Coach

dh2303

17 hours ago

@J.Y.Ping respectfully, that is not really responsive to my concern. My concern is not that seeing your AI product is going to harm me specifically, but that without an atypical measurement strategy, your AI product is going to harm 7sage. Management attention is finite. AI products are likely to generate compelling business metrics that are easy to measure (engagement and satisfaction), and, in my opinion, likely to produce harms on the 7sage system in general, which are harder to measure (reduced teaching effectiveness and reduced objective quality in total content, when compared to the counterfactual without an AI product). It's your company, of course you can do what you want, but I think your commitment to quality created a uniquely powerful education product. I've made a specific suggestion about how to protect that despite the pull of AI in edTech.

LSAT · New feature: AI Coach

dh2303

Yesterday

To me, LSAT studying has two goals: 1. achieving a desired score, and 2. sharpening my reasoning skills. I subscribe to 7sage because the content is high quality and is more likely to serve my goals than the alternatives. I realize the drive to release an AI product is very difficult to resist, especially for a company that sees creating new products as one of its fundamental roles, but I am still disappointed to see it. In my experience, LLMs universally degrade quality relative to human work product. Empirically, they can produce higher preference scores in some contexts, but that is not quality. It's related to sycophancy. I think this will be exciting for a number of people and might even lead to more customer satisfaction, but I do not expect it to produce a measurable benefit for the primary goals of LSAT studying, and it may produce a harm. I expect the sort of metrics you will use to evaluate this product are engagement and satisfaction, in addition to net cost. Please think carefully about how to test the effect of this tool on the quality of the content you produce and the effectiveness of your product on student achievement as well.

LSAT · New feature: AI Coach

dh2303

Yesterday

@icedshakenespresso text can have attitude. See, e.g., RC questions, which ask you to infer attitude from text. It is entirely appropriate to share, in a forum about a new product, the attitudinal qualities of text output from that product.

LSAT · New feature: AI Coach

dh2303

Yesterday

@icedshakenespresso No LLM has "scored a 180 on the LSAT." It has scored a 180 on an exam that includes question sets present in its training data, by virtue of being on the open internet. LSAC has not offered any AI company a live look at question sets not published.

LSAT · New feature: AI Coach

dh2303

Yesterday

@antitrust_fan I don't want to get in a flame war, but you're drawing an unwarranted conclusion about what LLM AIs can do well. They do well on benchmarks that test on exam material that is present in their training set (in other words, they are good at knowing the answer if they already have that answer in their training set). This is not a useful measure of being able to produce output that properly analyzes, e.g., the conditional logic. It's polluted with the answer choices and it's polluted with the large volume of low quality analysis

PrepTests ·

PT124.S1.Q3

dh2303

Yesterday

Great set theory question: P1 Many A are B (B is a subset of A), P2 claim C, some characteristic about B, Therefore, C is relevant to another subset of A. Weaken this by attacking the difference between the subset with the characteristic and the subset in the conclusion (here, diseases that are genetically based and diseases in common with cats)

PrepTests ·

PT129.S1.Q2

dh2303

Yesterday

Great example of reasoning about motive, modeled as causal reasoning. Randy claims a motive as cause for a phenomenon (only one source of news, mayor gets another source, the cause/motive is major interested in diversity of news sources). Marion attacks the cause/motive, posits the sole cause/motive of mayor's behavior is mayor's interest in AZCO (political supporter). The credited answer as "logically strongest counter" to the claim of bias as a sole motive/cause is the the same behavior occurring contra interest. As causal reasoning, this would be less strong because the effect is a different, analogous phenomenon (from a causal reasoning perspective, Randy and Marion are reasoning about the cause of a specific phenomenon, the mayor advocating for AZCO. Answer B gives an analogous phenomenon, the mayor advocating for, call it BZCO). As reasoning about motives, answer B is the classical rhetorical counter to a claim of bias, and that makes it strong examining it directly, without comparison to the answer choice set. It should be what you hunt for. It is also the strongest, relative to the answer set.

PrepTests ·

PT131.S3.Q2

dh2303

Yesterday

This is a good question for understanding what the LSAT means by bias.

PrepTests ·

PT134.S1.Q4

dh2303

Yesterday

Generally good explanation by Ardaschir Arguelles for a study design causal reasoning question, but it misses on one subtle and one not so subtle point.

Not so subtle point: E is directionally wrong, not irrelevant. There being wide variation in the specific causes of chronic back pain suffered by the patients in the experiment. Very relevant. It's back pain, it's in the patients in the experiment. That makes it relevant. This is describing heterogeneity in the baseline status of the outcome the study is meant to measure. In lay terms, this wide variation increases the noise in the study results, and may make it harder to detect a difference. A difference was nonetheless detected, so E is directionally wrong, not irrelevant.

Subtle point, but with implications for other questions: Magnets are not necessarily a placebo, the placebo effect, however, is the strongest counter to a study that doesn't have a control. Answer choice A is a restatement of the placebo effect. It's correct not because we know a priori that one of the treatments is a placebo, but because we CAN'T know, as there is no control. Direct the issue to the lack of a control, not to some a priori knowledge that a treatment is placebo. If you assume that lack of control MEANS the measured difference was placebo effect, you may be caught by more difficult trap answer. Lack of control prevents you from detecting a placebo effect, the explanation here commits the fallacy of unproven -> proven false.

PrepTests ·

PT137.S4.Q10

dh2303

Yesterday

Great question and good explanation by Eleanor Roberts. It is difficult academic prose that obscures the key structure of the argument: A is [list]. [list] is Characteristic. Therefore A is B. The argument depends on the idea that [list] and Characteristic are sufficient to designate A to be B.

PrepTests ·

PT137.S3.Q4

dh2303

Edited Yesterday

Another causal reasoning question with issues in the explanations. The credited answer is quite clearly the correct answer. It's an example of group allocation being associated with disease severity. It's supposed to be addressed by randomizing group allocation, which didn't occur here. However, C is also a threat to the study, it is just LESS of a threat. C is about blinding, but it's a possible threat and, unlike with B, we don't have a textual reason to know that it created an issue with the results. The study not being blinded is a threat. The study not being randomized is a threat. But the threat is clearly realized in choice B, because it tells us that group allocation is associated with disease severity. The explanation in D is also wrong. The study is about the effects of the scents on falling asleep, but the conclusion is about insomnia (which means difficulty sleeping, and includes falling AND staying asleep). D is not a strong weakener because it doesn't tell us anything about whether there was a differential effect, but the explanation saying "the study and conclusion are only about the effects of the scents on falling asleep" is unequivocally wrong.

PrepTests ·

PT118.S1.Q12

dh2303

Edited Yesterday

Generally good explanation in the video of a nice challenging question, but I have to quibble with the characterization of A. A is clearly wrong, but JY Ping says "just because the study was funded by paper manufacturers doesn't mean it was biased. It could be objective." No, it couldn't be objective. Objectivity is a state regarding the relationship between an observer and what is being observed. If an entity has an interest, they are not objective. Iniquum est aliquem sui rei esse judicem. Being objective means you don't have an interest in the outcome. The ad hominem fallacy is taking that lack of objectivity as formally sufficient for a conclusion about what someone claims, without examining the claim on its merits. I formalize it thus: A makes a claim, p. B makes a claim, c, about A's character or interest, and THEN USES c ALONE to draw a conclusion about p. Ad hominem is distinctly NOT identifying that someone or something is not objective because they have an interest. It is using that lack of objectivity to fail to examine a claim on its merits. This isn't just a philosophical or philological position. It's LSAT relevant. You'll often see ad hominem flaw questions where it's clear the LSAT requires precisely that step: using the source attack as sole foundation for discarding an argument without addressing its merits. Identifying competing interests is not the ad hominem fallacy. Using that alone to draw conclusions is. You will also see questions (PT131.S3.Q2) where a claim denying bias in a decision is refuted without even examining the merits of the decision. This provides some utility for understanding the meaning of bias in the LSAT corpus.

Lessons · Assumptions - Trash Bin Argument

dh2303

Edited Monday, May 18

@listening I'm glad my comments have been helpful! I didn't do any prior LSAT prep from outside sources. I am an older student, and come to this with some background a few different fields of study that ended up being useful. Here, that's probability and statistics via epidemiology and research design. I expect that's not particularly helpful, unfortunately, as it doesn't make sense to go and spend a few years studying something only peripherally related. I wish I had more actionable advice I could give you!

I should mention, I have found 7sage to be extraordinarily helpful in providing a method for approaching different types of LSAT questions. I find wrestling with boundary issues to be useful, so most of my comments are quibbles about some specific fine point. But I hope that doesn't come across as disparaging the course. It has been immensely helpful to me, and I recommend engaging deeply in it. Both JY and Kevin Lin are excellent. When there are two explanations, Lin's approach is usually the most helpful for me, but that's probably just because my way of thinking tends to align a little more with him. So, I do actually have actionable advice. Engage deeply with the 7sage material. Wrestle with it where you need to :)

PrepTests ·

PT126.S4.Q21

dh2303

Sunday, May 17

This is a useful example of a case where one needs to read the answers broadly. Typically, when asked to put our inference hat, we need to be sure we're using the stimulus as the source of an inference, rather than making inferences about what an answer choice might mean. Here, however, because all of the other answer choices are firmly not examples that illustrate the proposition, we're forced to make inferences about what the answer choice says. Answer choice C does not explicitly provide 'evidence that an act will benefit other people'. It provides evidence that an act will make a furnace run better (which is agnostic, as far as the evidence is concerned, as to whether it benefits ones self or other people). We have to assume that Betsy acting to benefit other people based on the neutral evidence of general benefit is sufficient to illustrate the proposition.

PrepTests ·

PT126.S3.Q25

dh2303

Sunday, May 17

This is a useful question because the argument uses an analogy, and an 'attack the analogy' answer choice is a trap by virtue of being descriptively inaccurate. I think the key here is 'any policy', vs. a specific policy, making E descriptively inaccurate. The dean presumes that a specific policy that applies to history courses is also justified with regard to math courses. JY says in his video that this isn't a policy, but clearly it is: history department policy "courses approaching its subject from a historical perspective need not be taught by a history professor. get your own department to teach it". That's a policy.

Lessons · Lesson 7 - A Trip to Mars

dh2303

Monday, May 11

@LiaWang they did :)

PrepTests ·

PT136.S1.P4.Q27

dh2303

Friday, May 8

@soleluna883 glad it helped! I think the skill of knowing when to get creative and think about possibility and when you be strict and read narrowly is precisely the thing that separates a good score from an excellent score.

Lessons · You Try - Auto Mechanics

dh2303

Edited Wednesday, May 6

@SerayaTalbott-Carey coming back to this, I think Ping is correct, and I was importing a requirement that is not necessary, and not part of the chaining requirement. Usually when we chain A -> B -> C, B can be conceptualized as a superset of the B that are A, so it is reasonable for A -m-> B -m-> C to operate in the same way.

PrepTests ·

PT119.S4.Q20

dh2303

Wednesday, May 6

This is a lovely question because it invokes a less common flaw. P is possible. Q is possible. Therefore it is possible that P and Q. This is a deductive flaw of modal logic. P is necessary. Q is necessary, There for it is necessary that P and Q. This is valid. It doesn't matter what the content of P and Q are, you cannot go from the possibility of two things individually, to the possibility that they both happen.

Lessons · Fast Track - Parallel and Parallel Flaw

dh2303

Edited Thursday, Apr 30

@RafaelGahan It's not valid in the abstract because of the possibility of A being internally contradictory. A -> B is satisfied for any internally contradictory A. (the conditional is true if A is always false). You can't assert the modal possibility of something internally contradictory. Let A: squares are circles, B: circles are round. If squares are circles then circles are round is a true conditional. Assert 'circles are round' to be true. You cannot conclude "therefore squares might be circles". That is why it's not a valid argument unless you know what A is, and know it's not internally contradictory.

PrepTests ·

PT123.S3.Q15

dh2303

Thursday, Apr 23

There is a fundamental error in Kevin Lin's explanation. He's a fantastic teacher and has excellent, deep knowledge on LSAT content, but, with so much content, errors can slip in. Here he says, answer B cannot weaken the conclusion of the argument, because it’s reporting a proportion, and the proportion is independent of the number of people in the sample. That is a fundamental error. It suggests you can simply assume sampling to be properly done without anything in the text that supports that assumption. To put the error in terms that may be clearer than the traditional causal reasoning terms, Lin's explanation assumes without warrant that a subset has the same proportional representation of characteristics as a superset. This is the same error as assuming A -> B -m-> C allows the conclusion A -m-> C. You cannot make that inference without warrant, just as you cannot make the inference that a sample has the same proportion as the population it is sampling from. You need a fair amount of math to warrant that characteristics of two samples (subsets) of a population (superset) can be compared to draw conclusions about the population (superset). Fortunately, we don't need to do that math for the LSAT. We just need to know that, for the purposes of the LSAT, the assumption is warranted if the samples are random (if we want to get into the weeds, there are different ways to produce that random effect, but in every case, to warrant generalizability, the difference between a sample statistic and the population statistic needs to be primarily due to random error). There are other requirements, but I haven’t seen them tested. Broadly, the reason I’m harping on this is the underlying assumption is going to backfire on sampling flaw and weaken questions. We cannot assume a sample was done in such a way as to allow a generalizable inference without textual warrant for that assumption. What follows is a fairly involved discussion of the structure, the correct answer, and the weakeners B and D that do not weaken the argument as much as C (which destroys it).

Lets break down the stimulus and the answer choices to explain what I mean here. Before diving too deeply, this is a good test question. In a database of 7000 questions, despite very rigorous quality control for each question, we can expect a few stinkers. This isn’t one of them. It is well set up. It has a clear construct. C is the credited answer, and is clearly the correct answer. The error in Lin’s explanation is in the categorization of B and D. B is UNEQUIVOCALLY a weakener of the causal claim in the argument. It illustrates a classic error in causal reasoning. D is also a weakener, for similar reasons. C is the answer, not because it is the only weakener, but because it destroys the argument. I think it's important to understand why. The stem, as per usual, is instructive. It asks for the answer choice that MOST SERIOUSLY weakens the argument. So that is what we need to look for. Not the first weakener we see, but the one that MOST SERIOUSLY damages the argument.

The stimulus describes an observational study using survey data to evaluate two different exposures (or, here, we could call them treatments) within a sample, identifying an outcome correlated with those different exposures in that sample, and then draws a generalized causal conclusion about a population these samples were taken from. The premise here is, within the survey sample, reporting long term treatment is correlated with a higher proportion that endorse “made things a lot better”. The conclusion here is that, generally, in the population sampled from (people who sought a psychologist’s help), long term treatment is more effective than short term treatment. Importantly, this is a two step inference, generalization (sample difference between treatments to population difference between treatments), and causal from association (correlation of satisfaction to effectiveness). Effectiveness is a causal claim, and this is no difference. Answer choice C is reverse causation. It is, as always, the largest threat to causal inference for a cross sectional study (which a survey is). The MC of the stimulus says treatment length causes satisfaction. Answer choice C says satisfaction causes treatment length. That is it. We can all go home, because one cannot weaken a claim any more than this. This is another fundamental lesson in LSAT level causal reasoning. If you grant reverse causality, the original claim is destroyed. In human scale and human processes, this is only ever going to be an issue if you see warrant for a feedback mechanism of some sort. One side note -- I see this in some of the comments described as a sampling issue. It is not a sampling issue. Answer choice says nothing about who was sent a survey, who received a survey, who was badgered to return the survey, or anything of the sort. It grants there is a causal relationship between treatment length and effectiveness of psychotherapy. It just points from effectiveness to length rather than the other way around.

How about B and D? They target the generalization step. This is an important lesson. Rather than dig into the math of why it is a fundamental error to assume that the proportion of any sample, no matter how it is drawn, is going to be the same as the population, let me offer an example. Answer choice B posits that people in the population (those who sought a psychologist’s help) were more likely to be in our sample if they were in the longer term treatment group. That’s our exposure variable (the thing we are investigating as a potential cause of treatment satisfaction or effectiveness). This means that the proportion in the population of long term treatment is x, and the proportion in our sample is x + d. That is prima facie evidence that our sample is not sufficiently random to contain the same proportion of a relevant characteristic. Answer choice D says people in the population were more likely to be in the sample if they were dissatisfied with their treatment. That’s our outcome variable (the thing we are investigating as the potential effect of our exposure). This means that the proportion of people in the population who are dissatisfied is y, and the proportion in our sample is y + d. This is, again, prima facie evidence that our sample is not sufficiently random to contain the same proportion of a relevant characteristic. I'd note that this isn't the typical way to describe the type of weakening B and D do, but it doesn't require the math or the hand waving more typical of this point. Broadly (and here is the hand waving), in causal reasoning, if selection in a sample is associated with the outcome, that's a good indicator you have a sampling bias problem. If selection in a sample is associated with one of two exposure variables you're trying to compare, you have a very important sampling bias problem.

For B and D, in no case (in the stimulus or the answer choice) are we given any information to warrant that the proportion of people in our sample who endorsed “made things a lot better” is similar to the proportion in our population. That’s a substantial threat. Suppose in both cases the true proportion in the population was 0.5 and the population sampled was, say 1000 US adults (it’s not, but we don’t need background knowledge of the number of US adults in psychotherapy to construct this hypothetical). We sampled 100 of each group. There are many arbitrary samples that would produce the survey data we see (20 short term endorse the phrase, 80 short term do not, and 36 long term endorse the phrase while 64 long term do not). If you think, well, that would be highly unlikely for it to be so far away from the sample population, NO IT WOULDN’T. Examine that assumption. It’s based on the math of the central limit theorem, and the assumptions of the central limit theorem, which include random sampling. WE NEED A WARRANT to assume our sample proportion is representative of the population proportion. We do not have it. In B and D, in fact, we have warrant to assume it is NOT representative, as we can already see that our sample proportions are not the population proportions. Broadly, I think the point here is that moving from conditional reasoning to causal reasoning doesn’t mean we can just all of a sudden say that subsets are representative of supersets. The claim is quite extraordinary, in fact, and requires a warrant. Induction isn’t loose deduction. It’s more involved math that allows a broader range of conclusions.

In conclusion, I’d like to highlight two things: first, watch for reverse causality. Reverse causality can be tricky to spot at first. If you approach any causal claim by identifying the exposure variable and the outcome variable it becomes much easier. Reverse causality destroys any causal claim, and is especially a threat in any scenario where the premises don’t clearly declare the exposure variable to occur before the outcome variable. Second, do NOT make the mistake of assuming subsets look exactly like supersets. This applies just as much to samples and populations sampled as it does to any attempt to chain inferences through a ‘most’ or ‘some’ claim.

PrepTests ·

PT123.S3.Q15

dh2303

Thursday, Apr 23

@cappelldaniel400 this is a causal reasoning question. The conclusion is a generalized causal claim about the relative effectiveness of short term and long term treatment. D absolutely weakens the claim. It just doesn't weaken it more than C. C is reverse causality, and is the most serious threat to a causal claim from cross sectional data (which is what a single time point survey is). The stimulus says the exposure "longer treatment" causes the outcome "endorse made things a lot better". Answer choice C says the outcome (feel you are doing well in treatment) causes the exposure (stay in treatment longer). If C is true, the causal claim in the argument must be false.

LSAT

•

Edited Sunday, Apr 19

dh2303

Vacuous truths and the LSAT

I've heard the following claim (in one form or another) here, and elsewhere in LSAT discussions: vacuous truths are a quirk of conditional logic that makes for an interesting philosophical discussion, but aren't really important for the LSAT. I think this is a mistake, and it has gotten a bit under my bonnet, so I thought I would post about it.

First, what is a vacuous truth? Typically it's described as a universal conditional statement, which we would represent in LAWGIC as A -> B, where the sufficient condition is contradictory or impossible, making the necessary condition irrelevant to the truth of the conditional. If pigs fly on their own power, then I have a 180 on the LSAT might be an example. Pigs do not fly on their own power, so I can put whatever I want in the necessary condition and the conditional will still be true. We can also think of this in terms of set logic, given an empty set, A, we can make a true statement A -> whatever we would like, because there are no elements in A.

Why do I think this concept is important on the LSAT? First of all, I grant that one does not have to think in this way in order to get a good score (even a 180) on the LSAT. People can have strong intuitive reasoning capabilities, and so grasp that saying "if I had a million dollars, I'd buy you a fur coat" doesn't mean much if one doesn't have a million dollars. Nevertheless, if we're to take a formal and rigorous approach to conditional logic, I think it is CRUCIAL to examine the formal representation behind that intuition, a truth table, for example, where we can list out all the possible combinations of having a million dollars and buying a fur coat (but not a real fur coat, that's cruel). This may not be understood as vacuous, as I'm sure many of us here will go into big law and at some point have a million dollars, but in the domain of right now, for me at least, I do not have a million liquid in any account, so in the domain of here and now, right now, I could say whatever I want about what I would do if I had a million dollars and be under no obligation whatsoever. So, lets examine the different cases. HM is I have a million dollars, BC is buy you a fur coat.

HM. BC. HM -> BC.

T. T. T.

T. F. F.

F. T. T.

F. F. T.

The conditional is satisfied in any case where one does not have a cool million, and in those cases where one does, only when one buys the requisite coat. That's what a conditional MEANS, and one must understand that to properly deploy them. If we're operating in a scenario where the conditional must be true (say it's a premise in a MBT question), we're limited to three rows of that table (the ones where the conditional is true, rows 1, 3, and 4). This is where we get modus ponens (assert the sufficient, conclude the necessary), and modus tollens (deny the necessary, conclude the sufficient is false). One MUST understand that the truth value of the necessary is irrelevant if the sufficient is false in order to do well on the LSAT, either formally or intuitively. This is precisely the concept of a vacuous truth. In a restricted domain, where nothing can satisfy the sufficient condition, the necessary can be whatever we want. That's where "the oldest mistake in the book" comes from (confusing the necessary condition for the sufficient condition).

Conceptually, understanding the empty set satisfies any conditional comes into play very clearly as an illustration of that oldest mistake in the book, for example, in PT 159.S1.Q21, which I won't spoil here, but might recommend for anyone questioning the relevance of the strongest form of a vacuous truth. To be clear on the lesson, I think it is pretty legible to moderately well prepared students that the stimulus is a necessary for sufficient error. But when you go hunting for the answer, you're left scratching your head UNLESS you understand that the reason necessary for sufficient is an error is because the conditional is satisfied in cases where the sufficient condition is an empty set.

PrepTests ·

PT159.S1.Q21

dh2303

Tuesday, Apr 14

This is an important question to demonstrate that vacuous truths are tested on the LSAT. A vacuous truth is a conditional that is true because the sufficient condition is never satisfied (i.e., A -> B is true when A is an empty set). You can, of course, reason your way through this question without thinking about vacuous truths or the empty set, but it is precisely testing that concept.

The argument is:

Domain: "my students"

HfB = "heard Mercado's lecture from the beginning"

TF = "thought it was fascinating".

P1: HfB -> TF

P2: Assert TF

C: Some HfB

Identifying this as a form of a necessary for sufficient error is relatively easy. Identifying that it is making a special type of that error is a little trickier. The Professor is saying, given TF, surely at least one of my students (not necessarily the same one) heard it from the beginning. To show this doesn't follow from the premises, HfB within the domain must be the empty set (none of the Professor's students heard Mercado's lecture from the beginning). This allows the premises to remain intact (because HfB -> TF doesn't require there to be anything that satisfies HfB). Answer choice D doesn't say it in precisely that manner, but they are logically equivalent, given the premises of the argument are true.

PrepTests ·

PT117.S4.Q17

dh2303

Edited Thursday, Apr 9

@Tumptytumtoes By a taxonomy question, I mean the correct answer choices match members of an existing set of categories of "supports causal inferences". By being a representative member of a category within the class "supports causal inferences", they are interpreted as an answer that "lends support to the climatologists' (causal) hypothesis". It's a "does this shape fit in one of the available shape holes" type question. That's the way people taught and understood causal inference a few decades ago, and is still the way some professionals think about it.

The way I approach studying and analyzing LSAT questions is to attempt to derive the principles LSAT question writers use to develop questions. Any given test question depends on a construct they want to test and a discriminant function they want to use to separate people who understand the construct from people who do not. Interesting questions challenge prior assumptions, provide new information, alter the boundaries of what I understand about those elements (the constructs and the discriminant functions). Unsatisfying questions challenge them in ways I can't satisfyingly resolve. This is an unsatisfying question. If I run into another one with a similar issue, I can build a better mental model of the constructs and discriminant functions they use. With this approach, post hoc defenses of wrong answers being wrong and right answers being right are very important to identify (and very tricky), because, to be precise in discriminanting among test takers, the correct answer MUST be the correct answer and each wrong answer MUST be the wrong answer, using a set of rules that do not conflict with any other question in the question bank.