THE TELL INDEX
Every Third CPA Exam Question Has a Tell.
Wrong answers give themselves away. We measured the patterns, audited them against 5,278 questions we did not write, and kept only what survived.
Already know them? Free 400-question sample →
1 in 3exam questions show a tell
+2right answers per core exam sectionThe math
The core exams run 50 to 78 MCQs (FAR 50, REG 72, AUD 78), average 67.
67 × 1-in-3 flagged ≈ 22 questions carry a tell.
On those, the audited odds lift your hit rate by roughly 9 points vs. a blind pick (derived from the card rates; conversion in the footnotes).
22 × 9% ≈ 2 extra correct from the odds alone. The moves add on top; on AUD’s 78 it runs closer to 3.
Only pays where you would otherwise guess. If you knew it cold, the tell adds nothing.
The core exams run 50 to 78 MCQs (FAR 50, REG 72, AUD 78), average 67.
67 × 1-in-3 flagged ≈ 22 questions carry a tell.
On those, the audited odds lift your hit rate by roughly 9 points vs. a blind pick (derived from the card rates; conversion in the footnotes).
22 × 9% ≈ 2 extra correct from the odds alone. The moves add on top; on AUD’s 78 it runs closer to 3.
Only pays where you would otherwise guess. If you knew it cold, the tell adds nothing.
70–74half of all fails land here. +2 flips them
20xthe per-hour payoff of memorizing more contentThe math
FAR holds ~5,000 testable facts for ~50 MCQs. One more memorized fact ≈ a 1-in-100 shot of being tested. Once.
The 12 rules on this page fire ~16 times per 50-question section, on every section, forever.
Per item learned: a rule returns ≈ 0.15–0.2 questions; a fact ≈ 0.01.
That ratio is the 20x.
Then go learn the facts anyway. A tell only finishes a question your knowledge started.
FAR holds ~5,000 testable facts for ~50 MCQs. One more memorized fact ≈ a 1-in-100 shot of being tested. Once.
The 12 rules on this page fire ~16 times per 50-question section, on every section, forever.
Per item learned: a rule returns ≈ 0.15–0.2 questions; a fact ≈ 0.01.
That ratio is the 20x.
Then go learn the facts anyway. A tell only finishes a question your knowledge started.
5,278outside questions these claims survived
Cross the liars off. Shapes that mark an option wrong on sight. Run these first, on every question.
Wrong answers love absolutes: always, never, regardless, void, entirely, under no circumstances. Real rules of accounting and law are full of exceptions, so an option that allows none is usually describing a world that does not exist.
The move: an absolute-claim option starts guilty until proven innocent. Kill it unless you can name the exception-free rule it states.
Backfires when: conditions are not claims. "Proceed only if" and "Items 1 and 2 only" are how real standards talk. Never auto-kill the word "only." The tell is the shape of the claim, not the word.
Measured: options containing a hard absolute word are wrong 82% of the time, against a 75% baseline for any option. A real edge, but a modest one, which is exactly why the shape matters more than the word. Sample: 748 flagged options. Audited outside: in every bank we tested, including AICPA-released questions, flagged absolute-claim options were wrong 76-100% of the time.
See it on a real question
Lena, age 17, received emergency dental treatment and now refuses to pay solely because she is a minor. Which statement is most accurate?
ALena may avoid the agreement entirely because contracts made by minors are void.Killed on vocabulary: void is an absolute state. Minors' contracts are voidable, not void.
BLena may disaffirm most contracts, but she must pay the reasonable value of necessary medical services.Correct. Note what it sounds like: hedged, carved with exceptions.
CLena is bound to pay the full contract price because medical-service contracts with minors are always fully enforceable.Killed on vocabulary: there is the word.
DLena is liable only if Bright Clinic can prove she misrepresented her age before treatment.Parked, not killed. Only-if is a condition, not an absolute claim. It dies on substance: necessaries create liability without misrepresentation.
The payoff: A and C die before you finish reading them, on the words void and always. You are at a coin flip before applying any contract law. And note the discipline on D: the word only does not kill an option, its substance does.
Try it yourself on the free 400-question sample →The roundest number in the ladder is usually planted. Real answers come out of real arithmetic, and real arithmetic is rarely tidy.
The move: flag the suspiciously tidy option, then work out which shortcut error would produce it. That error is usually the trap the writer set.
Backfires when: plenty of correct answers are round because the inputs were round. This demotes an option, it does not execute it.
Corroborated on our bank: when one option is clearly the roundest number, it is correct only 15% of the time, against a 25% baseline, across 747 questions. The outside audit agrees: 17% across 84 flagged questions in banks we did not write. The mechanism is human: round numbers get planted as bait for magnitude guessers.
See it on a real question
Lark shows $480,000 in trade receivables, which includes $12,000 of customer credit balances from overpayments. Expected credit losses are 4% of debit balances. What is net trade accounts receivable?
A$460,8004% taken off the wrong base, the raw 480,000.
B$468,000The tidy one: 480 minus 12, full stop. A clean subtraction that skips the allowance entirely. Suspiciously round numbers are usually shortcuts, and shortcuts are usually planted.
C$472,320Correct, and pleasingly ugly: credit balances reclassify to liabilities, so debit balances are 492,000, minus 4% (19,680).
D$460,320Another wrong-base construction.
The payoff: The roundest number in the ladder is the one produced by the laziest math. Real answers fall out of real arithmetic, and real arithmetic is rarely tidy. Flag the tidy option, name the shortcut that produces it, and you have usually found the planted trap.
Try it yourself on the free 400-question sample →An option that violates a hard definition is dead no matter what its number says. IRA distributions are never net investment income. Exempt is exempt.
The move: check each option against the definitions in play before doing any arithmetic. On monster calcs this routinely kills half the options for free.
Backfires when: it is only as strong as the definition. Use the real never-rules, not the ones you half remember.
Demonstrated, not statistically measured. The NIIT question in the video loses two options to definitions before any addition happens.
See it on a real question
Lee has wages, $14,000 of taxable interest, $10,000 of dividends, $36,000 of passive partnership income, a $40,000 IRA distribution, $20,000 of tax-exempt municipal interest, and $6,000 of investment interest expense. What is the NIIT treatment?
ANIIT is imposed on $94,000, because the taxable IRA distribution is included in net investment income...Killed by definition: retirement plan distributions are never net investment income. The other numbers are irrelevant.
BNIIT is imposed on $70,000, because... once modified AGI exceeds the threshold, the excess over the threshold is the NIIT base.Survives the definitions. This is the real decision: which number is the base?
CNIIT is imposed on $54,000: taxable interest, dividends, and passive income, reduced by the investment interest expense. Wages, the IRA distribution, and muni interest are excluded.Correct: the tax is the lesser of net investment income (54) or the MAGI excess (70).
DNIIT is imposed on $74,000, because municipal bond interest is considered in determining the NIIT base...Killed by definition: exempt is exempt.
The payoff: This stem looks like a nightmare and half of it dies without any addition. Two options each violate one hard definition. What is left is the single real decision the question was testing, the lesser-of rule, and that is where your knowledge belongs.
Try it yourself on the free 400-question sample →Not a tell. Math. Before computing anything, run three ten-second checks: direction, ballpark, ceiling. Which way must the answer point? Roughly how big must it be? What bound can it not cross? Options failing any check are dead, and they are dead by arithmetic, not psychology.
The move: direction first (the stem says fair value fell, so it is a loss, kill the gains). Then ballpark (round the inputs, five seconds of napkin math, kill the wrong sizes). Then ceiling (you cannot deduct more than you spent, kill anything over the bound).
Backfires when: it does not. These are properties of arithmetic, which is why this is the strongest tool on this page for calc questions. The only cost is the ten seconds, and the discipline to spend them before reaching for the calculator.
No stat needed. This transfers to the real exam at full strength, because the mechanism is math itself. If a page like this teaches you one habit for calc questions, make it this one.
See it on a real question
Park bought exchange-traded shares, a 2% stake with no significant influence, for $50,000. At year end, fair value is $46,000. How should Park report the investment?
AReport the investment at $46,000 and recognize a $4,000 unrealized loss in other comprehensive income.Survives the scan: right direction, right size.
BReport the investment at $46,000 and recognize a $4,000 unrealized loss in net income.Correct: equity securities run fair value through net income.
CReport the investment at $50,000 and recognize no gain or loss until the shares are sold.Killed by direction: the value fell. An option that pretends it did not is dead by arithmetic.
DReport the investment at $50,000 unless Park elected a fair value option at acquisition.Killed by direction: same refusal to mark the loss.
The payoff: One number fell from 50,000 to 46,000. Direction alone, before quoting any standard, kills the two options that refuse to show a loss. You are at a coin flip from reading one comparison, and the survivors happen to be twins whose one-phrase difference is the whole question.
Try it yourself on the free 400-question sample →Cut it to two. Shapes that trap the answer inside a pair. Half the field gone before you reason.
When two options say exactly opposite things, one of them is usually right. The writer built the question around that fork.
The move: kill the two non-opposite options first, then find the rule that picks a side.
Backfires when: occasionally both sides of the fork are wrong. Confirm the fork is actually the tested issue.
Audited. Our strongest tell: across 98 flagged questions in review-course banks we tested, the answer sat inside the fork 73% of the time. On official AICPA-released questions: 7 out of 7. It never dropped below 66% in any bank or section we measured. Baseline: 50%.
See it on a real question
Larch Co. reported a loss from continuing operations of $1,200,000, a gain on discontinued operations of $1,500,000, and net income of $300,000. Stock options would add 60,000 shares if included. Which statement best describes the EPS presentation?
ABasic and diluted EPS should be presented for income from continuing operations and for net income, but the options should be included in diluted EPS for net income only because net income is positive.Opposite 1 of 2: INCLUDED.
BBasic and diluted EPS should be presented only for net income, and the options should be included in diluted EPS...Not part of the fork, killed. Sanity check confirms: EPS must be presented for continuing operations, not just net income.
CBasic EPS should be presented for income from continuing operations and net income, but diluted EPS need not be presented...Not part of the fork, killed. Sanity check confirms: diluted EPS presentation does not just disappear.
DBasic and diluted EPS should be presented for both; the options should be excluded from diluted EPS because income from continuing operations is a loss and controls the diluted-share test.Opposite 2 of 2: EXCLUDED. Correct: the control number is a loss, so added shares are antidilutive.
The payoff: A and D stare each other down: included versus excluded. The question was built on that fork, so one of them is almost certainly right. Kill the two non-opposites, then find the rule that picks the side. Here it is the control number, and it is a loss.
Try it yourself on the free 400-question sample →When two options are near-identical, the answer usually lives in that pair, and the difference between them is the exact concept being tested.
The move: scan option shapes before solving. Found a twin pair? Kill the two outsiders, after a one-sentence sanity check each.
Backfires when: the pair holds the answer 73% of the time, not 100%. Never skip the sanity check on the outsiders.
Measured: on our bank the answer sits inside the pair 73% of the time (1,253 pairs, all sections). Audited outside: across 88 flagged pairs in review-course banks and AICPA-released sets, 61%. Never below the coin flip anywhere. The edge is real; ours runs hot, and we say so.
See it on a real question
The IRS mailed a valid notice of deficiency to Patel's last known U.S. address. Patel disagrees and wants judicial review without first paying the tax. Which response is required?
APay the asserted tax, file a claim for refund, and, if necessary, sue for refund in U.S. district court or the Court of Federal Claims.Outsider to the pair, killed. Sanity check confirms: the stem says WITHOUT paying first, and this is the pay-first refund route.
BRequest an administrative conference with the IRS Office of Appeals within 30 days after the notice of deficiency was mailed.Outsider to the pair, killed. Sanity check confirms: an Appeals conference is not judicial review.
CFile a petition with the U.S. Tax Court within 90 days after the notice of deficiency was mailed.Twin 1 of 2. Correct: 90 days for a U.S. address.
DFile a petition with the U.S. Tax Court within 150 days after the notice of deficiency was mailed.Twin 2 of 2. A real rule, for the wrong situation: 150 days applies to a notice mailed to a person outside the U.S.
The payoff: C and D are the same sentence with a different number. Before you know any tax procedure, the twins have told you what the question is about, and the two outsiders each die on one sentence. You are at a coin flip on structure alone. The concept, 90 days for a U.S. address, flips the coin.
Try it yourself on the free 400-question sample →When two surviving numeric answers differ by a small amount, the difference is the trap. The wrong one forgot exactly one step.
The move: compute the difference between the finalists and name what was forgotten. Discount amortization, the wrong base, a missed accrual. Naming the forgotten step picks the winner.
Backfires when: it is a finisher, not an eliminator. It only works after other tells have narrowed the field.
Demonstrated, not statistically measured. Watch the $6,000 versus $2,000 decision in the video.
See it on a real question
Hall bought bonds for $480,000, classified AFS. During the year Hall recognized $3,000 of discount amortization. Year-end fair value is $490,000. What is reported, and what is the consequence of the fair value adjustment?
AReport at $490,000 and recognize a $10,000 unrealized holding gain in OCI.Finalist 1.
BReport at $490,000 and recognize a $7,000 unrealized holding gain in OCI.Finalist 2, correct: the base moved to 483 (480 plus 3 of amortization), and 490 minus 483 is 7.
CReport at $483,000 and recognize no gain until sold.Killed instantly: AFS sits at fair value.
DReport at $490,000 with the $7,000 gain in net income.Killed instantly: AFS gains go to OCI.
The payoff: Two options die in five seconds on the standard AFS kills. The finalists differ by exactly $3,000, and the stem contains exactly one $3,000 item. The gap between the survivors IS the trap: option A forgot that amortization moved the base. Name what the wrong one forgot and you are done.
Try it yourself on the free 400-question sample →On a pure-numbers ladder with nothing killable in the stem, the extremes are rarely right. Here is why the mechanism is real: distractors on calc questions are built from plausible mistakes, and plausible mistakes bracket the true answer. Forget a step and you land low. Double-count and you land high. The correct number ends up in the middle of its own error family.
The move: last resort only. Sort the four numbers mentally, drop the highest and the lowest, and decide between the middle two. Any stem-based kill outranks this.
Backfires when: exam reviewers know about this cue and edit against it, so treat the measured number as a ceiling, not a promise. And it only applies when you truly have nothing else. A stem you half understand beats this rule every time.
Corroborated on our bank: on our bank, extremes are right only 22% of the time against a 50% baseline. Then we audited it on our largest outside sample: 1,216 all-numeric questions from review-course banks and AICPA-released sets. Result: 41%. Either way the lesson holds, the middle two are where answers live. Context: a FAR/REG/TCP tool; real AUD has almost no pure-number questions.
See it on a real question
Lane bought a machine for $200,000 and also paid $12,000 sales tax, $5,000 freight, $8,000 installation, $4,000 employee training, and $3,000 for a one-year maintenance agreement. At what amount should Lane initially record the machine?
A$217,000Killed as the low edge: the forgot-something pole.
B$225,000Middle pair. Correct: 200 + 12 tax + 5 freight + 8 installation. Training and maintenance are expensed.
C$229,000Middle pair, the coin flip partner: 225 plus the training that should have been expensed.
D$232,000Killed as the high edge: capitalize absolutely everything, the kitchen-sink pole.
The payoff: Nothing in the options to read, so read the construction. The low edge is built by forgetting, the high edge by including everything in sight, and the truth usually sits between the error poles. On our bank the middle two hold the answer 78% of the time. Last resort only, and only after the stem gives you nothing.
Try it yourself on the free 400-question sample →Pick the answer. The strongest tells. They point at it, or at the one sentence that decides it.
When three options are variations of one theme, the outlier is usually the answer. Strongest on NOT and EXCEPT questions, where three options are true statements and the stranger is your point.
The move: group the options by species out loud. Three siblings and a stranger? Bet the stranger.
Backfires when: you are in FAR, where the stranger pays only about 38% of the time. Strong bet in AUD and REG, a whisper in FAR.
Audited, and it depends on your section: in AUD and REG, the stranger is the answer 70-75% of the time in outside banks, and 7 of 7 on AICPA-released questions. In FAR it drops to 38%, because FAR strangers are usually filler, not the point. Same shape, different meaning by section. Baseline: 25%.
See it on a real question
Intangible assets are generally amortized over their useful lives. Which asset is the exception and therefore should not be amortized?
AA patent with 9 years of remaining legal life and expected usefulness for 7 yearsSibling: a finite life.
BA customer list expected to generate benefits for 6 yearsSibling: a finite life.
CA copyright with 12 years of expected remaining usefulnessSibling: a finite life.
DA franchise right that can be renewed indefinitely at nominal costThe stranger: no finite life anywhere in the sentence. Correct: indefinite-life intangibles are not amortized.
The payoff: Three options are the same species, an asset with a countable number of years. One option has no years in it at all. In an exception question, the stranger is your point, and you found it by sorting, not by reciting ASC 350.
Try it yourself on the free 400-question sample →The correct answer tends to share the most elements in common with the other options. Distractors are usually built by varying the correct answer one piece at a time, so the answer sits at the center of the family resemblance. This cue has a name in the research literature (Smith, 1982) and has been documented on professionally built standardized tests.
The move: a selection aid between survivors. Ask which remaining option the others look like they were carved from. That option is the better bet.
Backfires when: the lift is moderate, not decisive. Use it to break ties, never to skip reading.
Corroborated on our bank: the option most similar to the other three is correct 36% of the time, against a 25% baseline, across 11,105 questions. Consistent in every section, from 32% to 41%. Audited outside at strict thresholds: the family center was right 68% of 25 flagged questions in review-course banks. Thin sample, same direction.
See it on a real question
Four year-end events: $51,000 of sales taxes collected and owed to the state, $180,000 of bonuses required by written contracts for Year 1 services, a $120,000 dividend declared January 10 of Year 2, and a $420,000 probable litigation loss. Which amounts are liabilities on the December 31, Year 1 balance sheet?
ASales taxes collected ($51,000) and litigation loss accrual ($420,000) only.Option C minus the bonuses.
BEmployee bonuses ($180,000), declared dividend ($120,000), and litigation loss accrual ($420,000).Option C with the sales taxes swapped for a dividend that was declared after year end.
CSales taxes collected ($51,000), employee bonuses ($180,000), and litigation loss accrual ($420,000).Correct, and notice: it shares the most pieces with every other option.
DSales taxes collected ($51,000) and employee bonuses ($180,000) only.Option C minus the litigation accrual.
The payoff: Count the appearances. The $51,000 shows up in three options, the $180,000 in three, the $420,000 in three, and the dividend in exactly one. Every wrong option is the correct one with a piece missing or one bad piece added. The answer sits at the center of the family resemblance. That is convergence.
Try it yourself on the free 400-question sample →Where the number goes kills options faster than what the number is. Fair value or cost, OCI or net income, operating or investing. One classification sentence in the stem usually decides half the options.
The move: classify first, then kill every option that puts the number in the wrong place, then compute only to pick between the finalists.
Backfires when: your placement rules have to be cold memory. A shaky rule makes this a guess wearing a costume.
Demonstrated, not statistically measured. The two AFS questions in the video show the pattern paying twice.
See it on a real question
Purl bought bonds at $480,000. Management may sell the bonds before maturity if cash is needed. At year end, amortized cost is $484,000 and fair value is $478,000. What is the correct reporting?
AReport the investment at $478,000 and recognize a $6,000 unrealized holding loss in net income.Killed by placement: AFS unrealized losses go to OCI, not net income.
BReport the investment at $478,000 and recognize a $6,000 unrealized holding loss in other comprehensive income.Correct: fair value on the balance sheet, loss in OCI, and 484 minus 478 is 6.
CReport the investment at $484,000 and do not recognize any unrealized holding loss until the bonds are sold.Killed by placement: AFS sits at fair value, not amortized cost.
DReport the investment at $478,000 and recognize a $2,000 unrealized holding loss in other comprehensive income.The wrong-base trap: 480 minus 478 uses the purchase price instead of amortized cost.
The payoff: One sentence in the stem, may sell before maturity, classified the whole question as available-for-sale. Placement then killed two options before any math. The only arithmetic in the entire question is one subtraction, and it only decides between the last two.
Try it yourself on the free 400-question sample →On many calc questions the numbers are decoration. Every option is really an inclusion decision or a concept decision wearing a dollar sign.
The move: before computing anything, ask whether the dollar amounts even matter. If they do not, solve it as a concept question and never touch the calculator.
Backfires when: some questions really are computations. The check costs ten seconds. Do not force it.
Demonstrated, not statistically measured. Watch the video: four real calc questions solved with almost no arithmetic.
See it on a real question
Maya received a $6,000 employer bonus, a $4,000 birthday gift, $10,000 of bank loan proceeds, and $1,200 of municipal bond interest. Which is the most appropriate reporting action?
AInclude the $6,000 cash performance bonus from her employer in gross income.Correct: compensation through payroll is gross income.
BInclude the $4,000 cash birthday gift from her aunt in gross income.Killed: gifts are not gross income. True at $4,000, true at $4 million.
CInclude the $10,000 bank loan proceeds in gross income when received.Killed: loan proceeds are not income. You owe them back.
DInclude the $1,200 interest from qualifying municipal bonds in gross income.Killed: qualifying muni interest is exempt. The $1,200 is bait.
The payoff: Four dollar amounts in the stem and not one of them matters. Every option is an inclusion decision, not a computation. Ten seconds spent asking whether the numbers matter saved you all the arithmetic the dollar signs were begging you to do.
Try it yourself on the free 400-question sample →We audited every card on this page against thousands of questions we did not write. These four did not hold a usable edge on outside material. We could have deleted them quietly. They are here instead, stats and all, because the difference between a method and a sales pitch is what happens to the claims that fail.
Retired
Absolutist logic hiding in polite vocabulary. No tell-word appears, but the claim allows no exceptions, like "information cannot be useful unless it is precise."
The move: test the shape, not the vocabulary. Does this option describe a world with no exceptions? Kill it.
Backfires when: it takes an extra beat of reading. Easy to miss at speed, so make the check a habit, not an event.
Not measurable by script. This is Bumper Stickers for grown-ups, and it is caught by reading, not scanning.
See it on a real question
Redwood's CFO proposes that general-purpose financial statements should focus on regulator requests, internal profit targets, and reduced discussion of uncertain estimates. What is the primary issue that should govern the content of the statements?
AWhether the statements provide information useful to existing and potential investors, lenders, and other creditors in making decisions about providing resources...Correct: the broad, balanced textbook objective.
BWhether the statements provide information useful mainly to regulators and management in evaluating compliance...Killed: a narrowing absolute. General purpose means general, not one audience.
CWhether the statements are tailored to the most influential external user group, such as the lending syndicate...Killed: same narrowing logic in politer clothes.
DWhether the statements avoid reporting highly uncertain estimates whenever possible, because information cannot be decision-useful unless it is nearly precise and fully verifiable.Killed: no tell-word, but the logic allows no exceptions. The framework says the opposite, estimates are everywhere.
The payoff: Not one always or never in sight, but three options make no-exceptions claims in polite vocabulary. The tell is the shape of the claim, not the word. Test the shape and three options fall.
Try it yourself on the free 400-question sample →Right answers hedge: generally, usually, may, most, ordinarily. Carved-out exceptions are what true rules sound like. The mirror image of Bumper Stickers.
The move: this is a selection aid, not a kill. Between two survivors, the hedged option is the better bet.
Backfires when: hedged wording alone never proves an answer right. Distractor writers can hedge too.
Measured: options containing hedge words are correct 30% of the time against a 25% baseline. The outside audit agrees: 33% in banks we did not write. A nudge is all it ever was, and a nudge is what it measures as. Sample: 6,034 flagged options.
See it pick a winner (same question, other side of the coin)
Same question as Bumper Stickers, on purpose. That tell killed A and C. This one picks the winner between the survivors.
A...contracts made by minors are void.Already dead (Bumper Stickers).
BLena may disaffirm most contracts, but she must pay the reasonable value of necessary medical services.Listen to it: most, but-must, reasonable, necessary. A rule with a general case, an exception, and a boundary. That is what true legal rules sound like. Correct.
C...always fully enforceable.Already dead (Bumper Stickers).
DLena is liable only if Bright Clinic can prove she misrepresented her age before treatment.The losing survivor: flat, single-condition, no texture. And wrong on substance.
The payoff: The right answer sounds like a lawyer. Wrong answers sound like bumper stickers. Between two survivors, bet on the one carrying carve-outs and qualifications, because real rules are full of exceptions.
Try it yourself on the free 400-question sample →The option that echoes the stem's key vocabulary is right more often than chance. Correct answers have to engage the exact thing the stem asked about, and item-writing guidelines have warned test authors about this word-repetition cue for decades.
The move: a selection aid between survivors. When one finalist speaks the stem's language and the other changes the subject, favor the one that echoes.
Backfires when: experienced item writers plant stem words in distractors as deliberate bait, so an echo alone proves nothing. It is a nudge on top of substance, never a substitute for it.
Audited, and downgraded, by us: on our bank the echo looked strong. When we ran the identical test on thousands of questions we did not write, it fell to statistical noise: 28% against a 25% baseline. The gap was our own question-writer's habit, and we are rewriting those questions. Keep the echo as a tiebreak whisper at most. The card stays up because pulling a claim quietly is how everyone else does it.
See it on a real question
Lark is defending a lawsuit. Outside counsel concludes a loss is probable and provides a single best estimate of $450,000. Which reporting treatment is most appropriate?
ARecord a loss and liability of $450,000 in Year 1, with note disclosure as appropriate.The echo: the only option that speaks the stem's exact language, probable plus estimable plus the actual number. Correct.
BProvide note disclosure only, because a lawsuit is not recognized until it is settled or adjudicated.Killed: changes the subject to settlement, a test the standard does not impose.
CProvide note disclosure only, because the final amount is still uncertain.Killed: uncertainty alone does not defeat a best estimate.
DRecord nothing unless the loss becomes virtually certain.Killed: wrong threshold, and listen to the shape: nothing-unless is an absolute wearing a costume.
The payoff: The stem gave you two trigger words, probable and estimable, and one number. One option engages exactly those words and that number. The distractors each change the subject. The echo is a nudge, not a proof, but here the nudge and the substance point at the same letter.
Try it yourself on the free 400-question sample →When one numeric option is exactly half of another, the pair is spotlighting a convention fork. Half-year conventions, 50% limitations, mid-period math.
The move: a tiebreaker only. When a halving convention is plausibly in play, favor the half pair.
Backfires when: the signal is real but small. Never kill anything with it.
Corroborated on our bank: on our bank the pair holds 59%. The outside audit nearly killed this card: 53% across 151 flagged questions, a whisker over the coin flip. It stays because the pair reliably points at WHICH convention is being tested, even when it will not pick the side for you. We publish the weak numbers too. That is the point of this page.
See it on a real question
Taylor has AGI of $60,000 and paid $9,000 of unreimbursed qualifying medical expenses. Taylor itemizes. What is the medical expense deduction?
A$0Killed: the expenses clearly exceed any plausible floor.
B$4,500Half pair, part 1. Correct: 9,000 minus the 7.5% AGI floor (4,500).
C$3,000The stale-law trap: 9,000 minus a 10% floor (6,000), the old rule.
D$9,000Half pair, part 2: the forgot-the-floor-entirely trap.
The payoff: One option is exactly half of another, and that pair frames the real decision: does a floor apply, and how much survives it? The pair held the answer here, as it does 59% of the time on our bank. A tiebreaker, not a kill, but it pointed straight at the fork the writer built.
Try it yourself on the free 400-question sample →What this page will not tell you
Tells do not pass the CPA exam. Structure kills one or two options on a typical question. Knowledge does the rest. That is what you drill for. Drill enough questions and you would find every tell on this page yourself. We found them so you can start with them.
This page is free and stays free. The paid bank is this method built in: 17,651 questions with the tells flagged by name as you drill, every distractor decoded in the explanation, the same measured stats you just read attached to every flag, and an AI that grades you teaching it back. Reading this page is learning the method. The bank is 16,644 reps of it. That starts at $59, or $499 once, forever.
These numbers are tendencies, not promises. And when the outside data disagreed with ours, we changed the page: the audit made Opposites and Odd One Out stronger than we first claimed, shrank Twins and the Half Pair, and knocked the Echo down to a whisper. Every card above says which. If a prep course has never told you what they got wrong, ask what they do with claims that fail.
400 questions across the six sections, full explanations. Free, no credit card.
Try the free sample examHolding a failing score notice? The Retake Rate is half off the bank, forever. You've paid enough already. Claim the Retake Rate →
These are metal detectors, not landmine maps. A tell tells you where to look. It never replaces knowing the material. Most of these turn one-in-four into a coin flip, and the concept flips the coin.
Where the 9 points in the +2 math come from. Each card's rate converts to odds against a blind 25% pick. A pick tell pays its rate straight: pick the stranger at 72% and you gained 47 points. A pair tell splits its rate two ways: a 73% fork gives each finalist 36.5%, a gain of 12. A kill tell leaves three options standing at about 28% each, a gain of 3. Weight each gain by how often that tell fires and the average flagged question comes out about 9 points better than a blind pick. Every input is on a card above; the multiplication is yours to check.
Every tell here has to pass one test: a reason the real exam itself would produce the pattern. Content logic, or the way exam distractors get built. Quirks of our own bank do not qualify, and we cut rules that fail that test. In the latest audit that standard demoted two of our own cards. They wear it above.