The Mega Society asks for one in a million, and measuring it is the hard part

0 The short answer

The Mega Society is a high IQ society founded in 1982 that admits people who scored at the one in a million level, and what makes it unusual is that no standard norm sample can directly show that rarity, so its qualifying scores rest on untimed take home tests normed by estimation. The society accepts only the scores its own tests page lists, such as 43 correct on the Mega Test taken before 1995, and the critical literature, including a 2020 peer reviewed renorming by David Redvaldsen, disputes where those raw scores really sit on the scale. ACIS is one concrete option for a different job: an online, adult normed battery (Quick at 25 dollars and about 45 minutes, Optimized at 35 dollars and about 110 minutes, Full Scale at 59 dollars and about 175 minutes, prices read November 4, 2026) that shows where you stand by index. It is not accepted by any high IQ society and does not predict a pass, and its limit is that it is online, unsupervised and not a high range instrument.

1 What is the Mega Society, and who belongs to it?

The Mega Society is a small society, founded in 1982 by Ronald K. Hoeflin, that is open only to people who have scored at the one in a million level on a test of general intelligence. Its home page (read November 4, 2026) says it was founded to facilitate psychometric research, and describes membership as open to people who have scored at the one in a million level. It also quotes Guinness, which at the time of its entry counted 26 members at the 99.9999th percentile, and adds that the number is copied from that entry and is not a current count. The same page states that as of October 2024 there are 36 active members, a figure that excludes members who have died and people who qualified without joining. Those are the only membership counts the society publishes there.

The society's own account of its history (Noesis 149, "About the Mega Society", read November 4, 2026) adds that it absorbed an earlier group. Christopher Harding's 606 Society, whose name means 6 in 106, was incorporated into the new society, and members of the 606 Society with scores of 173 or more on Hoeflin's Langdon Adult Intelligence Test were invited to join. The same page says that membership later voted to make the Langdon test and Hoeflin's Mega Test the sole official entrance tests, and that the qualifying score on the Langdon test was raised to 175. The journal is called Noesis. The home page describes it as the journal of the society, distributed by email and archived on the society's website, and lists issue 217 of July 2026 as the latest.

Hoeflin also founded the Prometheus Society, whose standard is lower, and in the 2020 response quoted below he mentions serving as editor for the Triple Nine Society. That context matters for a reader because the Mega Society is not a rung above the other societies in a single ladder run by a single test. It is a separate group with its own admission rules, its own approved tests and a membership small enough to count by hand. The page that compares the requirements of the main societies puts the cutoffs side by side, and this page stays with the Mega Society and the measurement question it raises.

Two things this page will not do. It does not rank named individuals or repeat claims about who holds which score, because that belongs to the pages on Christopher Langan's claimed IQ and Marilyn vos Savant's IQ, where the sources for each claim are examined. And it does not reproduce a single item, sample question or answer from any society test, or link to any site that does. People who type "mega society test questions" are usually looking for practice material, but the tests are untimed take home exams whose value depends on the questions staying private.

2 What does the Mega Society require, and which tests count?

The society admits a person on a qualifying raw score on one of a short list of its own approved tests, and the list has shrunk because older tests were retired by date. Its tests page (read November 4, 2026) says it accepts only the qualifying scores listed there for admission purposes and lists these tests.

By those dates, the Hoeflin Power Test is the only test on the list with no date limit. The page gives a scoring fee of 50 dollars per test, payable through Stripe, and the Ultra Test page (read November 4, 2026) repeats that fee. Both were read on that date and can change, since the society's pages show no last updated date. The Ultra Test page describes a test with verbal and nonverbal sections that is untimed and unsupervised ("There is no enforceable time limit," in its words) and designed by Hoeflin. It links to a separate page for its norm instead of describing the norming itself, so this page cannot say how it was normed.

The tests page does not say why the Titan Test or the Mega Test stopped being accepted. The society's journal gives the reason for the Langdon test and the Mega Test: both were compromised, which means answers had spread. The 2020 response from Hoeflin (Noesis 206, read November 4, 2026) says that the Mega Test was similar until there was a spate of very high scores, which in his words may have been due to leakage of answers on the Internet. That is his own conjecture, offered as a possibility, and this page treats it that way.

One more rule from the society's constitutional history is worth knowing, labeled as history. The bylaws of the pre merger Mega Society (Noesis 150, read November 4, 2026) said that the qualifying score could never be set lower than 4.25 sigma above the mean or higher than 4.76 sigma above the mean, and that a person could qualify with one approved test or with a combined score from two or more approved tests using the Ferguson Formula. The page is titled pre merger, so it cannot be confirmed here that these rules still govern admission. The relevant fact is the band: even the society's founding rules allowed a range of cutoffs, from about 1 in 93,000 to about 1 in 1,000,000 on a normal curve, which is our arithmetic and not the bylaws' wording.

Finally, the society points to a committee finding. Its tests page says that a Prometheus Society committee in 1998 or 1999 found that the Langdon, Mega and Titan tests "are able to discriminate at the 4.75 sigma (one in a million) level." The Prometheus Society page on this site covers that society's own, lower standard.

3 How rare is one in a million, in IQ points and standard deviations?

One in a million is about 4.75 standard deviations above the mean, which is an IQ of about 171 on a scale with a standard deviation of 15 and about 176 on a scale with a standard deviation of 16, and the choice of scale alone moves the number by five points. This is our arithmetic on the normal curve, not a figure from any society: the upper tail beyond 4.75 standard deviations holds about 1.0 in a million people. Multiply 4.75 by 15 and add 100 to get about 171; multiply by 16 and add 100 to get 176. The page on why a standard deviation of 15 matters explains why the two scales differ, and the answer is historical, since older Stanford-Binet scales used 16.

That explains a confusion in the society's own literature. Kevin Langdon, whose test the society once accepted, wrote in a piece reprinted in Noesis (Concerning Mega Admission Standards, read November 4, 2026) that he works with a standard deviation of 16 and that the Mega cutoff of 43 corresponds to an IQ of about 172 on that scale, not the 177 that Hoeflin's own fitting gives. Each author is entitled to a scale, and a reader who sees 172 next to 177 should ask which standard deviation each used before deciding they disagree.

The percentile view is more useful than the IQ view for this reader. The page on IQ score versus percentile explains why the two should not be mixed and averaged, and the IQ percentile calculator converts a score on a chosen scale into a percentile. At the percentile level the contrast looks like this:

The society's own tests page makes the key point about the 160 row. It notes that the range of the Stanford-Binet is 40 to 160, which is four standard deviations of 15 about the mean of 100, and that a score of 160 corresponds to a population rarity of 1 in 30,000. In other words the society itself states that conventional batteries stop about thirty times short of its standard. The page adds that timed, supervised tests "do not accurately measure at the one in a million level." That sentence sets the terms for the rest of this article: the standard needs a different kind of instrument, and the question is whether such an instrument can be normed.

4 Why can a norm sample not directly support a one in a million score?

A norm sample of a few thousand people contains, on average, almost nobody at the one in a million level, so any rarity that far out is estimated by extrapolation or by a model, not read off the sample. The normal curve says that about one person in a million lies beyond 4.75 standard deviations. In a hypothetical standardization sample of 2,000 adults the expected number of such people is 0.002, and the chance that the sample contains no one at all at or beyond that point is about 99.8 percent (our arithmetic). Even at 4.0 standard deviations, the level of a 160 on a scale with a standard deviation of 15, the same sample would have no one beyond it about 94 percent of the time. To expect even ten people at the one in a million level, a sample would need about ten million, which no test publisher assembles.

Psychometric research states the same problem in general terms. Lenhard, Lenhard, Suggate and Segerer (Assessment, 2018) write that the more extreme a test result and the smaller the standardization sample, the higher the probability of a gap in the transformation between raw scores and percentiles. Their worked example is the Wechsler Intelligence Scale for Children, Fifth Edition, in which each age bracket includes 200 participants. They give a probability of .58 that all 200 score within 3 standard deviations of the mean (an IQ between 55 and 145), which they express as only a 42 percent chance that a single participant supplies data from the extreme ends of the test. Our check of the arithmetic agrees: 0.9973 raised to the 200th power is about 0.58. Their remedy is "vertical" modeling, which means fitting a smooth function from raw scores to percentiles across the whole sample so that the sparse ends are estimated from the dense middle. That is an estimate, not an observation, and the authors present it as a way to close the gaps.

The Standards for Educational and Psychological Testing (AERA, APA and NCME, 2014), open access at testingstandards.net, speak to what a user is owed here. Standard 5.8 asks that norms refer to clearly described populations that include the groups with whom test users will ordinarily wish to compare examinees, and Standard 5.9 asks that reports of norming studies include precise specification of the population sampled, the sampling procedures and participation rates, any weighting and the dates of testing. Standard 2.14 asks that conditional standard errors of measurement be reported at several score levels where possible, and in the vicinity of any specified cut score. Taken together, those standards give a plain test for any one in a million claim: say who was sampled, how they were drawn, and how precisely the score is measured at the cut.

Two things follow, and neither is a criticism of the people involved. First, a one in a million label is a statement about a model of the tail, and it inherits the assumptions of that model, chiefly that the distribution is normal far out and that the raw to IQ mapping is correct there. Second, items are written to discriminate across the range of the people the test was built for, so a test whose hardest item is solved by half of its takers says little about the top fraction of a percent. The page on how IQ scores are normed shows the usual process for a standard battery.

5 How were the Mega Test and the Titan Test made and normed?

Both tests were published in Omni magazine, and their norms came from comparing readers' raw scores with the standard test scores the readers reported about themselves, which is a method that works for a magazine audience and cannot be checked the way a supervised standardization can. The most detailed account is David Redvaldsen's investigation, published in the journal Psych in 2020 and printed with a response from Hoeflin in Noesis 206 (September 2020, read November 4, 2026). It says that Omni published three such tests between 1979 and 1990, the Langdon Adult Intelligence Test, the Mega Test and the Titan Test, and that the designers compared the number of correct answers with participants' self reported performance on standard educational or intellectual scales. The data arrived by mail, and Redvaldsen notes that this is an experimental method because there could be no supervision of the test takers and no control over whether the reported standard scores were accurate.

The Mega Test appeared in Omni in April 1985 and the Titan Test in April 1990. According to Redvaldsen, the Mega Test has 48 items: 24 verbal analogies, 12 spatial problems, 6 number series and 6 other numerical problems. There is no time limit, although the test suggests spending no more than one month, and reference materials and pocket calculators are allowed. Hoeflin dispensed with time pressure, and Redvaldsen suggests that the choice may be justified by privileging intellectual power over speed. A reviewer he cites, Dr. Carlson, had criticized the lack of supervision and the use of self reported data, while recognizing that Hoeflin chose them for practical reasons, because he had no large pool of people who could take the test under controlled conditions.

The score reports tell the sampling story. Redvaldsen writes that the January 1986 issue of Omni reported about 3,200 readers had submitted answers and that the median score was 15. Reading the published graph, he counted 3,258 test takers and calculated a mean raw score of 15 and a standard deviation of 8.24. Omni also reported that the mean IQ of its readers on the Mega Test was 141, on the scale used by the Stanford-Binet, which had a standard deviation of 16. Hoeflin arrived at that value by collating the previous scores on intelligence and achievement tests that participants reported. For the Titan Test, Redvaldsen found only 391 Omni readers and a mean raw score of 11.65 with a standard deviation of 10.42, and in his norming the Titan Test is in turn normed from readers' scores on the Mega Test, which is a second layer of estimation.

The Mega Society's own tests page describes two norming approaches in a single sentence each. Hoeflin normed the Mega and Titan tests by extrapolating from reported scores on supervised, timed tests, and Grady Towers used the Rasch model of item response theory to norm them. The page then adds that there is controversy over whether such tests measure the same thing as timed, supervised IQ tests. A 1991 critique in Test Critiques, quoted on the same page and credited to Carlson, is more pointed: it calls Hoeflin's approach interesting, inventive and internally consistent, and it says the approach violates many good psychometric principles by overinterpreting the weak data of a sample whose members chose to take the test. The society publishes both statements on its own page.

Hoeflin gave his own account of how he calibrated the middle of the scale. In the Noesis 206 response he says he used volunteers from the Triple Nine Society and chose problems that about half of them missed, so that a score of 24 out of 48 seemed about right for the 99.9th percentile, and that for the first 3,200 people 13 scored 43 to 48 right. He adds that in one of his final normings he equated the Mega Test distribution with SAT scores, placing the 99.9999th percentile at the same rarity as perfect SAT scores. That is a creative bridge between a take home test and a different test taken by a different population, and it is an estimate. The page on the hardest IQ tests explains why difficulty is not the same as measurement.

6 What did critics, and the designer, say about the cutoff of 43?

The sources disagree about where a raw score of 43 sits, with estimates running from about 1 in 33,000 to one in a million, and the disagreement shows that the cutoff rests on a norming model and not on a counted sample. Three views are on the record in the society's own journal, and the table sets them side by side. The IQ values use different scales, as the sources say.

Langdon's article is a plain internal argument. He writes that the membership's standard is 43 on the Mega Test, or 175 on the Langdon test, and he argues that the society should either raise its standard to one in a million or accept that it is admitting at a level of roughly two and a half to three per million. His own recommendation is a qualifying level of 46 on the Mega Test, which includes one point of what he calls ceiling bumping, an allowance for the fact that no one had scored perfectly and scores near the top of any test are compressed. The point is not who is right, but that the author of a test the society once accepted argued in print that its cutoff sat near 1 in 300,000.

Redvaldsen's 2020 paper reaches a mixed conclusion. In the abstract he writes that the official scores reported to test takers are too high, but that it is likely the Mega Test does stretch to the one in a million level, and that the Titan Test does not. He reports that the mean of all scores was IQ 137 on the Mega Test and 138 on the Titan Test, and that test takers who had previously taken standard tests averaged 135 to 145 on them, so that they had room to find their level without hitting a ceiling. In the Noesis 206 version, he reports that the highest Omni scorer on the Mega Test solved 45 correctly, which he equates to an IQ of 170 or slightly above, and that one in a million is reached at a raw score of 46, 47 or 48. In his conclusion he says Hoeflin's most recent Mega norming is too high by 11 IQ points at a raw score of 40, and that the Titan norming is too high by 13 points there. If his mapping holds, then a raw 43 on the Mega Test lies near the 1 in 33,000 to 1 in 53,000 range, which is our arithmetic from the percentiles in his table.

Hoeflin, who says he is not a statistician, answered that he had not read the whole report, found it counterintuitive that removing easier items would raise the ceiling, and suggested that the normal curve is known not to apply with much success to tests with high ceilings. He still allowed the possibility that a raw 43 is only at the 99.999th percentile. He also wrote that only one person in 30 years had a perfect score on the Titan Test, and that only a handful had 43 right. None of that settles the matter. It does show how the evidence is built: a mapping from raw scores to rarity that rests on assumptions, which several members of the society have examined in public.

Redvaldsen also records limits of his own study. No data had been published showing Titan Test correlations with other intelligence tests, he writes, and the Mega Test correlated 0.374 with the Stanford-Binet and 0.137 with the Wechsler Adult Intelligence Scale in the data he had. He describes the highest scorers as a group that selected itself, and concludes that the tests cannot be used on their own by psychologists because of the lack of supervision and the very long procedure.

7 Do untimed, unsupervised tests measure the same thing as an IQ battery?

An untimed, open book test of verbal analogies, number series and spatial puzzles measures reasoning and knowledge together with persistence, and it removes the speed demand that most modern batteries include, so its score is not interchangeable with a Full Scale IQ. In the vocabulary of the Cattell-Horn-Carroll model (the CHC model page lists the abilities), the Mega Test as Redvaldsen describes it draws, broadly, on crystallized ability (Gc) through verbal analogies, fluid reasoning (Gf) through number series and other problems, quantitative reasoning (Gq) through numerical items and visual processing (Gv) through spatial problems. A supervised battery covers a different spread, including working memory (Gwm) and processing speed (Gs) with time limits.

The two sets are related and not the same, and the society says so on its tests page. Redvaldsen reports correlations of 0.374 with the Stanford-Binet and 0.137 with the Wechsler scale for the Mega Test, and 0.565 and 0.562 with the Army General Classification Test and a Cattell test. Those come from self reported scores of a special group, and a restricted group depresses correlations, so they are not a verdict. They do show that the evidence a standard test provides, a battery of correlations in a described sample, has not been published. The Standards for Educational and Psychological Testing (AERA, APA and NCME, 2014), in their discussion of criterion evidence, note that data collections employing test takers selected for extreme scores (extreme groups) typically cannot provide adequate information about the association between variables.

Time matters because speed is part of what many batteries measure. The page on processing speed testing separates the cognitive efficiency that timed tasks capture from the reasoning that untimed ones do. An untimed test lets a person substitute persistence and library work for speed of thought. Redvaldsen defends the choice as favoring power over speed, and he also writes that the lack of a time limit rewards persistence and interest in the subject over capacity in a real world setting. Both readings can be held at once.

For an ordinary reader the useful comparison is with how a battery builds a score. ACIS reports a Full Scale IQ from 20 subtests across six domains, its fluid reasoning index includes tasks such as Matrix Reasoning and Figure Weights, and its verbal comprehension index includes the Vocabulary subtest. Those pages describe the formats without a single item from a society test, and they are the right place to see what a timed, adult normed subtest looks like. The pages on general intelligence and on what an intelligence test is explain why batteries combine several abilities into one composite, and why a single test of one kind, however hard, is a narrower thing.

8 Does measuring the very top still matter?

It does, because the best longitudinal evidence shows that people in the top fraction of a percent still differ in ways that matter, and that evidence also says that measuring them takes more than a single ordinary test. Kell, Lubinski and Benbow (Psychological Science, 2013) followed 320 people identified before age 13 as having profound mathematical or verbal reasoning ability, the top 1 in 10,000, for nearly three decades. They examined awards and creative accomplishments by age 38, together with occupational details, and described leadership positions in business, health care, law, the professoriate and STEM fields. Their abstract adds that identifying such potential and forecasting differential development within such populations requires assessing multiple cognitive abilities and using atypical measurement procedures.

Two honest readings follow. The first is that the top of the distribution is not a wall. A person at the 99.99th percentile is not simply the same as one at the 99th, and a test that stops at a low ceiling will lump them together. That is the reason high range tests exist at all, and it is the fair case for what Hoeflin and the society have tried to do. The second is that the study reached its resolution by a different route from a self reported magazine score. Redvaldsen describes the Study of Mathematically Precocious Youth, the research program behind this kind of work, as giving adolescents achievement tests designed for adults, and the group was then followed for decades. That is the sort of design in which a rarity statement can be checked against outcomes, and the Mega and Titan tests have no published equivalent.

The study should not be stretched. It concerns a group identified at 13 by one design, it does not mention the Mega Society or any society test, and group findings are never individual forecasts. What a reader can take from it is narrower: differences exist at the top, measuring them is hard, and the field's own answer is multiple abilities and special procedures, not a single higher number.

This is also why record claims are poorly posed. A number far above the ceiling of every supervised battery depends on a norming model, a test, and conditions that few can check. The page on the highest IQ ever recorded takes up that question directly.

9 What are the real options for an adult who wants a normed picture?

The realistic choices are a psychologist giving a full battery, the society's own untimed tests, free screeners, job practice tests, and a normed online battery, and they differ in who supervises, who is the comparison group and what you receive. Prices and times come from pages opened on November 4, 2026 and can change.

The Pearson WAIS-5 page (read November 4, 2026) gives an age range of 16:0 to 90:11, and it does not state a price, so the cells say so. A supervised battery from a psychologist is the route when the question is clinical, when the result must be accepted by an institution, or when a rigorous account of a single person's profile matters. The page on where to take an IQ test lists the places that give each kind, and the guide to the best online IQ tests reviews the online field.

ACIS sits in one row of this table and sells what that row describes. Quick contains six subtests in three domains, Optimized 13 subtests in five domains and Full Scale all 20 in six domains; there is no subscription, breaks are allowed, progress is saved between subtests and there are 30 days to finish. The free trial, started from the Full Scale form, opens five subtests (Digit Span, Alphanumeric Sequencing, Visual Sequence, Coding and Symbol Search) and needs no card; finished trial subtests are kept if the person later unlocks a form that includes them. The IQ test for adults page explains the whole route in one place, and the free IQ test page covers what the trial offers.

For a reader curious about being gifted and not about a society, the gifted test for adults page describes how adults confirm where they stand, and the page on high range IQ testing covers the need to go beyond the range of standard batteries. The honest summary here is that high range instruments are a small, contested corner of the field.

10 How should you read a score that sits near the top of a scale?

A score near the top of any scale should be read as a band, not a point: it carries a larger uncertainty than a middle score, it means little without the norm group behind it, and a profile of indices says more than one number. Start with the vocabulary. A scaled subtest score on ACIS runs from 1 to 19 with a mean of 10 and a standard deviation of 3, and indices and the Full Scale IQ sit on the familiar scale with a mean of 100 and a standard deviation of 15. Each comes with a percentile, and the Full Scale IQ comes with a 95 percent confidence interval. The interval is the honest way to say that any observed score is an estimate of an underlying level, with some error around it.

The Standards for Educational and Psychological Testing (AERA, APA and NCME, 2014) explain why a single piece of a score is less precise than the whole. Standard 2.3 asks that estimates of reliability and precision be reported for each total score, subscore or combination of scores that is to be interpreted, and its comment warns that the consistency of a total score may be acceptably high while subscores have unacceptably low reliability. Standard 2.4 asks that when an interpretation emphasizes the difference between two observed scores, reliability data for that difference be provided, because a difference between two noisy scores is noisier than either. These are general psychometric principles and say nothing about any particular test, but they tell a reader what to trust: the broad composite before the narrow subtest, and the gap between two indices only when it is large and the technical documentation supports it.

At the top, a second source of uncertainty applies. The section on norming showed that the tail of a norm table is estimated from few observations, and Standard 2.14 asks that conditional standard errors be reported at several score levels because the error need not be the same at every level. Whenever a test reports a very high number, ask three questions. Who was in the comparison group, and how many of them were near that score? How was the raw score converted to the reported scale? And what is the uncertainty in the vicinity of the score? A test that cannot answer all three is describing a position, not measuring one.

Profiles carry information that a single number hides. The page on index score differences shows how often adults have gaps between indices and why a gap alone is not a flaw. The General Ability Index page describes a composite that gives less weight to working memory and processing speed than the Full Scale IQ does. The page on reliability and validity explains the two ideas behind every interval in this section, and the technical manual documents how the report builds its composites.

11 What conditions matter when you take a test on your own?

An unsupervised session measures what your device, your room and your attention allow, so fix those before you start any timed or audio task and do not retake a subtest after seeing the score. The same applies to any take home test, including the ones a society requires, with one difference: a society test has no clock and no sound, while a timed battery has both. In a normed online battery the main practical points are a screen large enough to see the whole task, a stable connection, and a quiet room without interruptions. Three ACIS subtests, Digit Span, Alphanumeric Sequencing and Arithmetic, present their items as audio, so they need working sound, headphones are sensible, and a notification chime in the middle of a sequence can cost a point that has nothing to do with ability.

The Standards describe the principle in general terms. In their treatment of administration, Standard 6.3 says that testing environments that are not monitored, such as those on the Internet, should meet the standardized administration conditions, and that otherwise the report on scores should note that standardized conditions were not guaranteed. Standard 6.4 adds that the environment should furnish reasonable comfort with minimal distractions. ACIS allows breaks and saves progress between subtests, with 30 days to complete a purchased form, so the sensible plan is to take a few subtests in each sitting when you are alert and to stop between subtests, never inside one.

For an untimed take home test the main condition is different. Redvaldsen's paper notes that, as it stands, the library resources available to test takers are a factor in the score generated. That is a design feature, since the Mega Test allows reference materials, but it means that two people with the same raw score may have used very different amounts of outside help and time. It is also the reason a score from such a test is not comparable with a supervised score, even when both are written as an IQ.

Answer leakage is the other risk. The society's journal gives compromise of answers as the reason older tests were retired, and Hoeflin suggested leakage as a possible cause of a spate of very high scores. Any test whose items circulate stops measuring what it set out to measure, which is why this page reproduces none.

12 Where ACIS sits, and what a buyer should do

ACIS is a normed, online, unsupervised battery sold by the publisher of this page, it is not accepted by any high IQ society, and it cannot say whether you would pass a society's test; what it can offer is an adult normed picture of where you stand, by index, before you pay a fee or apply. The limits come first. ACIS is online and unsupervised, it is not a clinical or diagnostic instrument, it is available in English only, and its own research page states that it "is not intended for clinical diagnosis, hiring decisions, accommodation determinations or admission to high IQ societies." Its verbal and knowledge tasks are affected by English proficiency and educational exposure, and the report is not a certificate for any institution.

The Standards for Educational and Psychological Testing (AERA, APA and NCME, 2014) set the bar that applies to ACIS as much as to any society test. Standard 1.1 asks the developer to set out clearly how scores are intended to be interpreted and to delimit the population and the construct, and its comment adds that no test permits interpretations that are valid for all purposes or in all situations. Standard 1.3 asks that, where an interpretation has not been evaluated, that fact be made clear and users cautioned. Applied here, an ACIS score has not been evaluated as a predictor of a society's decision, so it should not be read as one. The APA Guidelines for Psychological Assessment and Evaluation (2020) point the same way. Guideline 6 asks psychologists to select tools with sufficient validity evidence for their uses and sufficient score reliability, and Guideline 8 asks them to reflect the appropriate normative comparison, situational influences, effort and standardized administration. Those are duties of psychologists, and they are also a useful checklist for a lay reader choosing among tests.

Here is what ACIS does offer, in its own terms. Every form reports a Full Scale IQ with its 95 percent confidence interval, plus indices and scaled scores, against adult norms for ages 16 to 90. Full Scale covers all 20 subtests in six domains in about 175 minutes for 59 dollars; Optimized covers 13 subtests in about 110 minutes for 35 dollars; Quick covers six subtests in about 45 minutes for 25 dollars; and the Custom Test lets you choose subtests at 10 dollars each. Prices were read on November 4, 2026, and they can change. The refund wording is literal: "Full refund within 5 days if the report does not deliver the features described at checkout or a technical issue prevents you from accessing your results." That is not an unconditional guarantee.

What should a buyer do? If your question is whether to apply to the Mega Society, start with the society's own pages, which state the tests, the cutoffs and the fee. Read the dispute in Noesis 206 and the Langdon article before deciding what a raw score means for you, and treat a fee as the price of a raw score checked against a cutoff, not as a measurement of your abilities. If you want to know where you stand against adult norms first, with no promise about any society, the free trial of the Full Scale form needs no card and opens five subtests, and a purchased form returns the full index picture. If you want a rigorous account of one person's profile, a supervised battery from a psychologist is the right route. And if your interest is instruments for people at the far top of the range, read the page on high range testing, and keep in mind that none of the sources above shows that a norm sample can directly confirm a rarity of one in a million.

13 Sources Behind This Page

Every figure above is traceable to one of the following sources, each linked in the body where it is first used. The normal curve figures in the section on rarity, the sample size probabilities and the 1 in 33,000 to 1 in 53,000 reading of Redvaldsen's table are our own arithmetic on published numbers and are labeled as such. The society pages were read on November 4, 2026 and can change.

• American Educational Research Association, American Psychological Association and National Council on Measurement in Education. Standards for Educational and Psychological Testing. Washington, DC: AERA, 2014 (Standards 1.1, 1.3, 2.3, 2.4, 2.14, 5.8, 5.9, 6.3 and 6.4, and the discussion of extreme groups under criterion evidence).
• American Psychological Association, Task Force on Psychological Assessment and Evaluation Guidelines. APA Guidelines for Psychological Assessment and Evaluation. Approved by the APA Council of Representatives, March 2020 (Guidelines 6 and 8).
• Kell H, Lubinski D and Benbow C. Who Rises to the Top? Early Indicators. Psychological Science, 2013, volume 24, issue 5, pages 648 to 659.
• Langdon K. Concerning Mega Admission Standards. Reprinted in Noesis, the journal of the Mega Society (from Noesis 125; the page shows no issue date). Read November 4, 2026.
• Lenhard A, Lenhard W, Suggate S and Segerer R. A Continuous Solution to the Norming Problem. Assessment, 2018, volume 25, issue 1, pages 112 to 125.
• Mega Society. Home page. Read November 4, 2026.
• Mega Society. Tests Accepted for Admission to the Mega Society. Read November 4, 2026.
• Mega Society. Ultra Test admission page. Read November 4, 2026.
• Mega Society. About the Mega Society. Noesis 149. Read November 4, 2026.
• Mega Society. Bylaws of the Pre-merger Mega Society. Noesis 150. Read November 4, 2026.
• Noesis, the journal of the Mega Society. Issue 206, September 2020, containing Redvaldsen's investigation and Hoeflin's response. Read November 4, 2026.
• Pearson Assessments. Wechsler Adult Intelligence Scale, Fifth Edition (WAIS-5) product page. Read November 4, 2026.
• Redvaldsen D. Do the Mega and Titan Tests Yield Accurate Results? An Investigation into Two Experimental Intelligence Tests. Psych, 2020, volume 2, issue 2, pages 97 to 113.

14 Frequently Asked Questions