Everything, on one sheet

The card

WORDS IN CONTEXT — reference card
Predict before you look. Read the whole sentence, not just the blank's clause.
CONTRAST (although/despite/yet) -> opposite of the stated idea.
CONTINUATION (moreover/and/similarly) -> same direction as the stated idea.
CAUSE-EFFECT (because/as a result) -> word explained by, or explaining, the cause.
Wrong answers are usually right-direction, wrong-degree or wrong-register — not backwards.
"As used in the text" items: pick the specific sense the sentence defines through context, not the word's most common meaning.

Why it works — Why prediction beats recognition

Working memory can hold roughly four to seven items, and four near-synonymous answer choices are specifically designed to overload that capacity if you try to evaluate all four against the sentence simultaneously. Predicting first collapses the problem from "compare four words against a sentence" to "match one word I already have in mind against four labels" — a much smaller, much more reliable comparison. This is the same reason a multiple-choice test is easier when you cover the choices with your hand before reading them: you remove the option to be talked into a wrong answer by a plausible-sounding distractor.

Traps — 5

Right direction, wrong degree

The choice points the correct way but is too mild or too extreme for what the sentence establishes — picking "annoyed" when the sentence's own logic demands "furious." This is the single most common wrong answer on this question type, because the traps are built to be directionally correct.

Familiar-word bias

Choosing the most common, most comfortable word among the four regardless of fit, because it's the easiest one to imagine in a sentence — not because the sentence's logic actually calls for it. Precision, not familiarity, is what's rewarded.

Half-context reading

Deriving a prediction from only the clause next to the blank and missing a signal word later in the sentence that reverses the direction — reading only the first half of a sentence that has a "but" in the second half.

Register mismatch

A choice that means roughly the right thing but doesn't fit the sentence's formality — a casual word dropped into a clearly academic or journalistic register, or vice versa. The Digital SAT's passages have a consistent register, and the correct choice always matches it.

Secondary-meaning trap (on "as used in the text" items)

For items that ask what a word already in the passage most nearly means, the trap is picking the word's most common dictionary sense instead of the specific secondary or tertiary sense the passage is actually using — e.g. "sound" meaning valid/well-founded rather than sound meaning noise.

The card

TEXT STRUCTURE & PURPOSE — reference card
Three stems: main purpose of the text / function of the underlined portion / overall structure.
PURPOSE = an act, always "to + verb". MAIN IDEA = a proposition. Different questions.
FUNCTION is relational: read the sentence before and the sentence after, then name the job.
Label as you read: claim / evidence / concession / counter / qualification / revision / upshot.
Find the pivot (but, however, yet, "the find does not show"). The purpose lives after it.
Predict in your own words BEFORE any choice is visible. Then screen choices, don't select one.
Kill a choice on a specific word: verb too strong, move never made, wrong sentence, wrong order.
Compound choice ("X and then Y") fails if either half fails. Check the second verb separately.
CONCEDE / QUALIFY / COMPLICATE / REFUTE are four different answers. Never merge them.
Whose stance is it — the author's, or the researchers/critics/character being reported on?
Literary texts: purpose is what the portrayal establishes, not what happens in the plot.
Elimination narrows; it does not verify. Confirm the survivor positively, on a named phrase.

Why it works — Why the choice that says the most about the topic is usually the wrong one

Purpose and function are relationships between parts of a text, and relationships are not stored the way facts are. Ordinary reading encodes content readily — what the passage was about, which you can still recite a minute later — and encodes structure poorly, because structure is what your comprehension consumed rather than what it produced. That asymmetry is the whole reason these items work as tests. By the time you reach the choices, you still hold the content in working memory and you no longer hold the shape, so the choice that echoes content feels checkable and the choice that names the move feels abstract and unsupported. Labelling each sentence's job while you read is not a study ritual and not note-taking; it forces the structural representation to be built and kept, so that when you read the choices you have something to compare them against other than a memory of the topic.

Traps — 6

Content echo

The choice accurately restates what the sentence or text says, so it feels confirmed by the passage — but it names a topic rather than a job. This is the most-chosen wrong answer on function items, and there is a one-question test for it: could this choice have been written by someone who read only the underlined sentence and nothing around it? If yes, it is a paraphrase, and a paraphrase is never a function.

Scope inflation

The choice's verb claims more than the text delivers — "proves," "establishes," "demonstrates," "refutes," "overturns" — over a text that reports, suggests, describes, or complicates. On purpose items the wrong answers live in the verb, not the noun, so read the verb first. The mirror error also exists: a deflated verb ("mentions," "notes") applied to a text that genuinely argues.

Pre-pivot purpose

Building the purpose out of the setup. Most of these texts turn exactly once, and a choice describing only what came before the turn will be accurate about the text and wrong about its purpose. On literary excerpts this is the same trap wearing a different coat: the choice faithfully summarises the routine on the first half of the page and misses the letter that arrives on the second.

Part-for-whole swap

Answering "main purpose of the text" with the job of one sentence, or answering "function of the underlined portion in the text as a whole" with the purpose of the whole text. The stem names the size of the thing being asked about. This one is lost to momentum rather than to misunderstanding, which is why it survives so far up the score range.

Order swap

On structure items, the choice names moves the text really makes but in a sequence it does not follow. On function items, the same error appears as a job that belongs to a neighbouring sentence — the objection really is in the text, just not in the sentence you were asked about. Both are checkable in about ten seconds against the passage's actual order, and almost nobody checks.

Attributed stance

The choice's verb assigns an attitude the author never takes — "criticises," "laments," "champions," "dismisses," "warns" — over a text that is neutral or merely corrective. Correcting an explanation is not criticising the people who held it, and noting a limitation is not warning about it. The related version: attributing to the author a view the text explicitly attributes to the researchers, critics, or character it is reporting on.

The card

CROSS-TEXT CONNECTIONS — reference card
One sentence for Text 1's claim. One sentence for Text 2's claim. Verb of position in both.
Then name the relationship, with its strength, BEFORE reading any choice.
Six shapes: contradicts / alternative explanation / narrows the scope / says the evidence is insufficient / supports and extends / reframes the question.
Alternative explanation is the most common on hard pairs. Contradiction is the rarest.
The credited answer must be underwritten by the responding text alone. Plausible is not the standard; stated is.
Point at the clause. If you cannot, eliminate — however sensible the choice sounds.
Re-read the stem after predicting: whose response, to which claim, in which direction.
Match the hedge. "May / suggests / does not establish" never licenses "proves / refutes / disproves."
Both-agree items: the answer is the shared premise, not either conclusion. Verify against each text separately.
Silence is not a position. A text that never mentions X neither endorses nor denies X.
Text 1 often reports someone else's view. Answer about the view the stem names, not about Text 1's author.
Longest item in the module. Budget the time; the two claim sentences are what pay it back.

Why it works — Why naming the relationship before reading the choices is the whole method

Distractors on this item type are not written from scratch. They are written by taking the credited relationship and bending exactly one dimension of it: the target (the response is correct but aimed at a claim Text 1 never made), the strength (correct direction, overshot certainty — "refutes" where the author wrote "does not establish"), the direction (Text 1 responding to Text 2 rather than the reverse), or the source (a fact borrowed from the other text, or from the world, rather than from the responding author). That construction is why prediction works here even better than it does on Words in Context. If you walk into the choices with a fixed relationship — "Text 2 grants the observation and denies the cause, aimed at Reyes's innateness conclusion, at the strength of does-not-establish" — then each wrong choice announces which dimension it bent, and you are running four cheap one-dimensional checks instead of four open-ended judgements. Walk in without a prediction and you are doing the opposite: reconstructing the relationship four separate times from four fluent, confident paraphrases, each written to be persuasive on its own. That is not a reading task any more; it is four consecutive opportunities to be talked out of a correct reading you had not yet committed to.

Traps — 6

Assumed disagreement

Treating every pair as a fight, so the most adversarial choice wins by default. A substantial share of pairs are complementary — Text 2 supplies a mechanism, a case, or a boundary for what Text 1 observed — and on those the confrontational choices are the trap. Name the relationship honestly, including "supports," before the choices get a vote.

Binary collapse

Flattening a qualified response into a total one: Text 2 writes "does not establish," "cannot discriminate," "remains one candidate," and the chosen answer says "refutes," "disproves," "shows to be false." Not established and false are different claims. On the hardest pairs, two choices point the same direction and differ only in strength, and the weaker one wins whenever the author hedged.

Right relationship, wrong target

Text 2 genuinely does disagree, and the choice describes that disagreement accurately — but attaches it to a claim Text 1 never made, or to a subsidiary detail rather than the claim the stem named. This is why step 4 is re-reading the stem after predicting, not before: the prediction is what makes a mis-targeted choice visible.

Attribution swap

Putting one text's evidence in the other author's mouth. Text 2's author has not read Text 1 and cannot cite its data; a response that depends on a fact appearing only in Text 1 is disqualified on that ground alone, however well it fits the argument.

Outside-knowledge import

The choice is a reasonable, sometimes even correct, thing a specialist in the field would say — and nothing in the responding text licenses it. This trap gets stronger the more you know about the subject, which is why it catches high scorers specifically. Domain knowledge on this item type is a liability unless it is disciplined by the point-at-the-clause test.

Silence read as position

Text 2 never mentions X, so the choice has Text 2's author denying X — or endorsing it. Not addressing a claim is not taking a position on it. This distinction decides most "both authors would most likely agree" items, where the tempting wrong answer is something one author asserts and the other simply never raises.

The card

CENTRAL IDEAS & DETAILS — reference card
Two stems: "Which choice best states the main idea of the text?" (claim) and "According to the text, ...?" (one detail).
Predict in one clause before reading choices: [subject] + [what the text claims about it].
Three of four choices are usually true. Truth eliminates nothing. Sort on altitude first.
Altitude test: if this choice were all I knew, would I know why the text was written?
Read the last sentence twice — on non-fiction the point usually lives there.
"Although X, Y" / "While X, Y" -> the main idea is on the Y side, always.
Match the hedge: hypothesize -> think/propose. Never proved, confirmed, demonstrated.
Detail items: hold the exact interrogative (what challenge / one reason / which piece), then find the single sentence answering it. Return to the sentence; don't answer from memory.
Two survivors: compare each choice's SUBJECT to the passage's subject, not to each other.
Before committing, reread your choice for quantifiers only: only, first, all, most, always, never.
R&W module: 32 minutes, 27 questions (~71s average). These passages are short — bank time here.

Why it works — Why the wrong answer is almost always true

Two separate things make the true-but-subordinate distractor the default wrong answer on this skill, and both change what you look for. The first is item construction. College Board states that each Reading & Writing module contains a small number of unscored pretest questions used to trial items for future forms — that part is official. What follows is inference rather than a published rule, and should be held as such: a distractor that is flatly false gets eliminated by anyone who read the passage at all, so it separates nobody and adds nothing to the item's ability to discriminate. A distractor that is true but subordinate separates readers who tracked the passage's structure from readers who only tracked its content. Screening items on how they actually perform therefore selects for true distractors and quietly retires false ones. Expect the wrong answers on a live item to be things the passage says. The second mechanism is in your head, and it's the reason knowing the first isn't enough. Reading a choice that matches a sentence you read forty seconds ago produces a strong signal of recognition — and that signal is indistinguishable from the one produced by the correct answer. Recognition encodes "I have seen this" and carries no information whatsoever about altitude. There is no way to feel the difference from the inside. The only defence is to commit to your own one-clause summary before the choices are on screen, so you're comparing four labels against a claim you already hold rather than shopping for whichever choice feels most familiar.

Traps — 6

True-but-subordinate

The choice restates a real sentence — usually the striking statistic, the example, or the mechanism — and is wrong only because the passage exists to say something else. This is the single most common wrong answer on this question type and the reason accuracy-checking fails as a strategy. Diagnostic: if the choice were the only thing you knew about the passage, would you know why it was written? A detail leaves that question unanswered.

Setup-only summary

The choice captures the passage's opening framing — the background, the definition, the standard account — and stops before the turn. Passages that open with a definition or with "it has long been assumed" are built for this trap. Its 800-level variant is subtler: a choice that states the passage's premise rather than its conclusion, which is genuinely supported, genuinely important, and still not the point.

Altitude inflation

The choice is a broader claim than the passage supports: one study becomes a statement about the field, a described effect becomes a recommendation, a bounded finding becomes a general law. Broader is not better. The credited answer's scope matches the passage's scope, which means a choice can be wrong for being too large just as easily as for being too small.

Modality upgrade

The passage hedges — hypothesize, suggest, may, could, on this evidence — and the choice does not: proved, confirmed, demonstrated, established. The substance can be a perfect summary and the choice still fails on its verb. The mirror image also appears: a passage that reports a clear finding paired with a choice saying questions remain or that researchers are unsure.

Recombination

The choice is assembled from real words in the passage arranged into a relationship the passage never asserts. Every noun is familiar, so it reads as verified; the verb between them was invented by the item writer. This one survives an accuracy check better than any other trap, because the parts are all genuine and only the joins are fabricated.

Right question, wrong sentence

On detail items: the choice paraphrases a true sentence that answers a different interrogative than the one asked — the stem asks what challenge and the choice reports what technique, or the stem asks how the bird knows and the choice reports what the bird does. The fix is mechanical: reread the stem's interrogative after choosing, before committing.

The card

COMMAND OF EVIDENCE (TEXTUAL) — reference card
Evidence is a relation, not a property. Never ask "is this good evidence?" — ask "evidence for WHICH claim?"
1. Find the claim as a sentence. On two-view texts, check whose.
2. Underline the spine: rather than / not merely / not because...but because / more X than Y / chiefly / only when / without.
3. Write in one sentence what the credited choice must do — BEFORE reading the four.
4. Test each choice against that sentence, not against the topic. Read to the final clause.
5. Tie-break by counting assumptions. The credited answer needs none.
ILLUSTRATE = an instance, shown. SUPPORT = makes it more likely. WEAKEN = makes it less likely.
"If true" is an instruction, not an invitation: grant the finding, then judge the relation.
Strongest weakener of a causal claim usually contradicts nothing — it supplies a competing cause, or shows the groups differed beforehand.
Strongest support usually kills the most obvious competing explanation. Ask what a skeptic would say first.
Two views: evidence for a premise both sides concede discriminates nothing. Find what they predict DIFFERENTLY.
Belief claims need a quotation showing the belief, not one settling the fact.
Final check: "Because [choice], it is more likely that [claim]." If it needs an extra assumption, it isn't the answer.

Why it works — Why every wrong answer is on topic

Distractors on this question type are not invented from scratch — a choice that is visibly irrelevant is a wasted slot, because nobody selects it and the item’s difficulty collapses. They are built by taking the credited relation and deleting one component of it: keep the subject, drop the contrast; keep the topic, drop the causal link; keep the claim’s content, drop the requirement that it be shown rather than stated. The result is four choices that are all about the same thing, of which exactly one preserves the full relation. That construction is why fluency is actively dangerous here: the choice that most sounds like the passage is the one that was written to sound like the passage. Topic-matching is a fast, automatic process and relation-checking is a slow, deliberate one, so under time pressure an untrained reader will match topic and feel confident doing it — high confidence and a wrong answer produced by the same mechanism. Writing down what the credited choice must do, before the choices are visible, is the intervention that works, because it converts a four-way comparison into four independent yes/no checks against a fixed standard.

Traps — 6

Topic match, claim mismatch

The choice is unmistakably about the same subject as the claim and does nothing for the claim itself. The most common wrong answer on this question type, and the one that feels best while you are selecting it, because recognizing the subject is exactly the sensation of understanding. Test: state the choice and the claim as one sentence — “Because [choice], it is more likely that [claim]” — and listen for the point where it stops making sense.

Half-claim

The claim has two load-bearing parts and the choice satisfies one. Triggered by *rather than*, *not merely*, *not because … but because*, *more X than Y*, *only when*, *without*, *chiefly*. A quotation that lands the affirmed half while ignoring the rejected half is the standard build; a quotation that affirms the rejected half outright is the same trap with the sign flipped.

Circular restatement

The choice says the claim again in different words, or reports the observation the claim was invented to explain, instead of giving an instance of it or a reason to believe it. A claim cannot be evidence for itself, and neither can the data already sitting in the passage. On “support” items this distractor is often the one that most closely echoes the passage’s wording — which is why it gets picked.

Plausibility override

Eliminating a choice on a “which finding, if true” item because the finding seems unlikely. The stem has already granted it. This trap also runs in reverse: accepting a choice because it matches something you happen to know about the real world, on an item that is answerable from the text alone.

Wrong target

The choice supports or weakens something adjacent to the claim: a premise the question did not ask about, the study’s methods rather than its conclusion, the motive clause rather than the mechanism, or — on a two-view text — the other scholar’s position. Before committing, re-read whose claim is on the table and which sentence of it you were asked about.

Intensity mistaken for probative force

Choosing the most dramatic, most quotable, or most emotionally legible option. Vividness is a property of a sentence; evidentiary weight is a relation between a sentence and a claim, and the two are uncorrelated. On illustration items the credited quotation is very often the plainest one on the page.

The card

COMMAND OF EVIDENCE - QUANTITATIVE - reference card
Furniture first, once: title, axis labels, UNITS, scale, legend. Then never re-read them.
Then the text: name the claim in one clause, and note the connective on the blank.
Predict the shape of the needed data before reading a single choice.
Filter 1 (cheap, mechanical): is it TRUE of the figure? Wrong units = wrong answer.
Filter 2 (costly, decisive): does the true statement do the job the sentence asks?
however / by contrast -> a difference.   for example -> one concrete case.
because / since -> the reason.   "the data bear this out" -> that claim, not a neighbour.
Count vs rate: most cases and highest prevalence are usually different rows.
24% to 42% = +18 percentage points = +75 percent. Two units, two numbers.
"thousands of trips" on the axis: a bar at 305 is 305,000.
A figure shows WHAT. Never WHY, and never a year it did not plot.
Unlabeled bar between gridlines supports "between 160 and 180", not "exactly 175".
Rival hypotheses: cite the comparison the losing one CANNOT explain, not the biggest gap.
All four choices true? Stop reading the figure. Decide on the claim's scope words.
Before locking: re-read the claim once, after the arithmetic. Right answer, wrong quantity is the 1600 leak.

Why it works — Why the wrong answers are usually true

The item is built backwards from the figure, and that construction order determines what the distractors look like. A writer holding one graphic and needing three hard wrong answers has an almost free supply: every other comparison the same graphic supports is accurate, on topic, in the right register, and useless for the specific claim in the text. Manufacturing a false distractor takes deliberate work. Manufacturing a true-but-irrelevant one takes a second glance at a figure already drawn. That asymmetry is why "is it accurate?" is the weaker of the two filters even though it feels like the whole task — and it is why the choice you are drawn to is a liability. Your eye goes to the largest visual difference on the figure, and visual salience is a property of the drawing, not of the argument. The biggest bar is in the wrong answer for the same reason it is on the graph: because it is easy to see.

Traps — 6

True but irrelevant

An accurate reading of the figure that supports a different claim from the one the sentence makes. This is the single most common wrong answer on the question type, and it is common by construction rather than by accident: every figure supports many comparisons, and only one of them is the one the text asked for.

Superlative reflex

Reaching for the largest or smallest value on the figure because it is the most visible thing on it, when the claim asked for a comparison between two named groups. A maximum is not a comparison. If the sentence does not contain a superlative, the answer almost certainly should not either.

Count for rate

Answering with a raw total when the claim is about prevalence, share, per capita, or per unit area — or with a percentage when the claim is about how many. The orchard with the most infected trees and the orchard with the highest infection rate are routinely different orchards, and both numbers are printed in the table so that both answers can be written.

The printed number is not the quantity

Quoting an axis value as if it were the amount. "Thousands of trips" makes a bar at 305 into 305,000. A rise from 24 percent to 42 percent is 18 percentage points and a 75 percent increase, and calling it "an 18 percent rise" misstates the data even though the subtraction was right. A choice that misstates the data cannot support anything, however well it fits the argument.

Off-figure claim

Asserting something the figure cannot contain: a cause, a motive, a reason, a value for a year that was not plotted, a group that is not in the legend, or a value more precise than the scale can carry. A figure reports what; it never reports why, and it never covers what it did not measure.

Connective mismatch

Data that continues where the sentence said "however," a generalisation where the sentence said "for example," or a contrast drawn on a different axis from the one the text raised — comparing two settings when the sentence was comparing two treatments. The connective is part of the question; it is not decoration on the sentence.

The card

INFERENCES — reference card
The standard: could this be false while every sentence in the passage stays true? If yes, it is out.
Read the connective in front of the blank first. therefore/this suggests = consequence. however/but = limit. in other words = restatement. because = cause.
Strip the passage to 2-3 bare claims: what was assumed, what was found, what complicates it. Hard keys need two combined.
Predict in your own words, then test all four choices. Do not stop at the first that sounds right.
Two survivors? Take the weaker one, if it still does the connective's job. Hedged claims are cheap to guarantee.
Read every quantifier before every verb: all / none / only / never / always / proves is where close items are decided.
Outside knowledge is a liability. If the defence of a choice needs a fact not on the screen, it is not the answer.
"Found no effect" licenses "did not detect it", never "there is none".
A reported claim ("researchers proposed", "if this is sound") licenses conclusions about the claim, not about the world.
Direction check: rate is not total, share is not count, slower growth is not decline.

Why it works — Why the bar is necessity and not plausibility

A keyed answer has to survive every student who challenges it, every teacher who writes in, and the item's own statistics. Under a plausibility standard, two or three of four choices are defensible at once — and an item with two defensible answers is a broken item. It shows a low or negative point-biserial correlation, because the strongest readers split between the two good choices instead of converging, and items that behave that way get pulled from the pool. The only standard that leaves exactly one defensible choice is entailment: the answer the passage forces rather than the answer it invites. So when the key is the modest claim rather than the interesting one, the test is not being pedantic; it is doing the only thing that produces a scorable question. The same pressure explains the hedging. A narrow claim is cheap to guarantee, so item writers reach for it, because a narrow claim is what survives review. Reading a hedge as "too weak to be the answer" inverts the actual signal.

Traps — 6

The plausible outsider

The choice is probably true in the world, or true given what you happen to know, but the passage never gives it. This is the most common wrong answer on the type, and the tell is that justifying it requires a sentence you supply yourself. If your defence of a choice begins "well, obviously," the choice is out.

Quantifier inflation

The passage's claim with an absolute bolted onto it — all, none, only, never, always, any, proves, cannot. The passage supports "in this trial"; the choice says "in general." Read the quantifiers in every choice before you read the verbs; on a close item, the quantifier is usually where the decision is.

Half-passage completion

The choice is genuinely entailed by the setup but ignores the complication the blank sits after — it would have been the right answer if the passage had ended two sentences earlier. Produced almost entirely by not reading the connective in front of the blank and so not noticing that the completion's job is to limit the earlier idea, not extend it.

Reversed dependency

The choice swaps what depends on what, or flips the direction of a comparison: the passage establishes that A varies with B and the choice says B is caused by A; the passage says one quantity grew more slowly and the choice says it fell. The content is all from the passage, which is what makes it convincing at a glance.

Wrong quantity

Correct reasoning about a real quantity that is not the one the passage measured — a share answered with a count, a rate answered with a total, a change within one group answered as a difference between two groups. The arithmetic in these choices frequently checks out, which is precisely why they are dangerous.

Recommendation creep

The passage reports what is; the choice says what someone should do, will do, or must now believe. Prescriptions and predictions are almost never entailed by a description, and a completion that tells researchers what to investigate next is doing something the passage did not license.

The card

BOUNDARIES — reference card
One test, run twice: can the words LEFT of the mark stand alone? Can the words RIGHT of it?
Independent clause = subject + main verb, with no subordinator or relative pronoun at the front.
-ing alone, to + verb, and a lone which/who/that clause are never main verbs. No main verb, no clause.
IND | IND -> period, semicolon, or comma + FANBOYS (for and nor but or yet so). Never a bare comma.
DEP first | IND -> comma. IND | DEP last -> usually no comma. Order decides.
Semicolon = period. Independent on both sides. Sole exception: separating series items that already contain commas.
Colon: independent clause on the LEFT, anything on the right — and the right side must specify, explain, define, or list.
however / therefore / moreover / nevertheless / consequently are adverbs, not conjunctions. They cannot repair a splice.
Supplement test: delete it. If the sentence still works, bracket it with a MATCHING pair — comma+comma, dash+dash, or parentheses.
Never a single comma between a subject and its verb, or between a verb and its complement, however long the subject.
Never punctuate by pause. Punctuation marks clause rank, not breath.
If two choices would both be grammatical, neither is the answer — recheck where the clause boundary actually falls.

Why it works — Why a comma cannot do a semicolon's job

English punctuation encodes rank, not pause length. Commas are within-clause marks: they separate items in a series, attach a dependent clause, bracket a supplement — all operations inside a single predication. Periods, semicolons, and comma-plus-FANBOYS are between-clause marks: they close one complete predication and open another. A comma splice asks a within-clause mark to do a between-clause job, and the cost is concrete rather than aesthetic. The reader parses forward expecting the first clause to continue, meets a second subject and a second verb, and has to reparse the sentence from the comma. That forced reparse is what the prohibition exists to prevent. It is also why the pause heuristic fails exactly where it matters: you pause where you run out of breath and where you want emphasis, and neither has any relationship to where one predication ends. Two people reading the same sentence aloud will pause in different places; they will not disagree about where the clause boundaries fall. That objectivity is why this domain can be scored at all — and why it is the fastest part of the test to take to near-perfect.

Traps — 6

Splice by adverb

However, therefore, moreover, nevertheless, consequently, thus, instead, indeed — these announce a logical relationship and have no power to join clauses. A comma in front of one at a clause boundary is a splice however strongly the sentence seems to want it. The list of words that do earn a comma at a clause boundary is closed: for, and, nor, but, or, yet, so.

Pause punctuation

Choosing the mark that matches where you would take a breath. Breath tracks length and emphasis; punctuation tracks clause rank. The two coincide often enough to feel reliable on easy items and diverge precisely on the hard ones, which is what makes this the most expensive habit in the domain.

Subject-verb amputation

A single comma dropped between a long subject and its verb, or between a verb and its complement, because the reader wanted somewhere to rest. "The claim that the eruption destroyed the palaces outright, has been undercut" — no length of subject ever licenses that comma.

The unmatched pair

Opening a supplement with a dash and closing it with a comma, or opening with a comma and never closing it at all. Almost always caused by distance: the two marks sit far enough apart that the second is chosen without reference to the first.

Colon after a fragment

"The kit includes: a compass, a whistle, and a mirror." The presence of a list is not a licence for a colon; the completeness of the clause before it is. Watch for a colon parked immediately after a verb (includes, are, were) or after "such as" and "including" — in every one of those cases the left side has not finished its own predicate.

The -ing sentence

Treating a participial phrase as an independent clause and giving it a period or a semicolon: "The committee met for six hours. Producing no agreement." An -ing word is a main verb only with a helper in front of it, and with no main verb there is no clause for a strong mark to separate.

The card

FORM, STRUCTURE & SENSE — reference card

STEP 0: read the four choices DOWN the page. They differ in ONE dimension. Name it, then run only that check.

AGREEMENT -> strike every prepositional phrase, comma-fenced appositive, and relative clause between subject and verb. Re-read what is left.
  Present-tense verbs invert the noun rule: -s = SINGULAR verb. (the dog runs / the dogs run)
  Inverted order (There is/are ..., Among the X is/are Y): subject comes AFTER the verb. Find it first.
  "The number of" = singular. "A number of" = plural. Gerund or clause subject = singular.
  as well as / along with / in addition to / together with do NOT make a subject plural.
  A and B = plural. A or B / neither A nor B = agrees with whichever is NEARER.
  one of the X who [plural verb] -- but the only one of the X who [singular verb].

TENSE -> never by ear. Find the date, the time phrase, or the order of the two events.
  had + participle = the EARLIER of two past events. Not "a long time ago."
  has/have + participle = began earlier, still running or still relevant (since 2011, over the past decade).
  If X had ..., Y would have ... -- keep "had" in the if-clause, "would have" in the result.

PRONOUN -> say the antecedent out loud, then count it. its / their / theirs / whose take NO apostrophe.
  it's = it is. who's = who is. If the expansion does not fit, the apostrophe is wrong.

MODIFIER -> the noun immediately after the comma must BE the thing the opening phrase describes.
  A possessive in that slot is a dangler (Maria's finish, the map's survival).
  Passive voice is a legitimate fix when it puts the right noun in the subject slot.

PARALLEL -> items joined by and/or, and both halves of not only...but also / either...or / both...and, share a form.
  Comparisons compare like with like: "the rainfall in Lagos exceeds THAT in Nairobi."

APOSTROPHE -> three questions, in this order:
  1. Is anything owned? If no, no apostrophe anywhere. Plural takes a bare s.
  2. Is the owner singular or plural? Decide from the text, not from the choices.
  3. Apostrophe after the COMPLETE owner word; add s only if it does not already end in one.
     geologist's / geologists' / children's -- never geologists's.

Why it works — Why "sounds right" fails worse here than anywhere else on the test

Speech and edited writing diverge most sharply on exactly the features this skill point tests. In live sentence production, a plural noun sitting close to the verb pulls the verb into agreement with it even when it is not the subject — psycholinguists call this agreement attraction, and it is a documented error pattern in fluent native speakers, not a sign of weak English. That is why Digital SAT conventions sentences are built with long interrupters between the head noun and the verb: the distance is the weapon, because distance is what makes attraction fire. The feeling of correctness you get from reading a sentence aloud is being generated by the same system that produces the error, so listening harder cannot help. Deleting the attractor can. The apostrophe items exploit an even cleaner version of the same gap: possessive 's and plural s are homophones, identical in speech and distinguished only in writing, so reading an apostrophe item aloud carries literally zero information about the answer.

Traps — 6

Proximity agreement

Matching the verb to the nearest noun rather than to the head of the subject. This is the single most exploited pattern in the domain: every long interrupter you see between a subject and its verb was placed there to trigger it. The tell is that the sentence sounds fine — it is supposed to, because the same mechanism produces the error in ordinary speech.

Nearest-verb copying

Choosing a tense because it matches the verb closest to the blank instead of the tense the text's time evidence requires. Two verbs in one sentence often should differ, and that difference is exactly what encodes which event happened first — copying the neighbour deletes the information the sentence was built to carry.

Apostrophe by ear

Deciding a plural-versus-possessive item by reading the sentence aloud. Possessive 's and plural s are homophones; the spoken sentence is identical in all four cases, so any decision reached by ear here is a coin flip wearing the costume of a judgment. Run the three questions instead: is anything owned, is the owner singular or plural, where does the apostrophe fall.

The floating modifier

Accepting an opening phrase whose subject never actually appears, because the intended meaning is obvious anyway. Comprehensibility is not the standard being tested — attachment is, and the test will happily hand you a completely understandable sentence that is still wrong. Its 800-level variant is a possessive in the subject slot: after "Having trained for months," the word "Maria's" does not satisfy the modifier, because the subject is "finish."

Half-parallel series

Matching the form of the first two items in a list and letting the third drift, or matching the wrong halves of a correlative pair — "not only to reduce costs but also improving safety." With correlatives, check what immediately follows each half of the pair, not the general shape of the sentence.

Apostrophes on possessive pronouns

Writing it's, their's, or who's where the possessive its, theirs, or whose is required. Pronouns carry possession in their own form and never take an apostrophe; an apostrophe on one always makes a contraction instead. If "it is" or "who is" cannot be substituted into the sentence, the apostrophe version is wrong by definition.

The card

TRANSITIONS — reference card
Stem is always: which choice completes the text with the MOST LOGICAL transition. All four choices are grammatical by design; grammar and register cannot decide these.
Method: cover the choices. Compress idea A. Read idea B to the FULL STOP. Name the relationship in plain English. Then uncover, sort by family, split on the constraint, and verify by rereading the pair joined.
ADDITION (B adds to A): also, in addition, additionally, moreover, furthermore, besides. Moreover/furthermore = a further and heavier point.
PARALLEL (two distinct cases match): similarly, likewise, in the same way, by the same token. Needs TWO cases.
CONTRAST (A and B differ, nothing more): however, in contrast, by contrast, on the other hand, alternatively, whereas.
CONCESSIVE (B holds DESPITE A): nevertheless, nonetheless, even so, still, all the same, that said. Needs A to predict not-B. Name the defeated expectation or drop the family.
CORRECTION (A denied, B replaces it): on the contrary, rather, instead. Needs a negation or rejection in A. "Instead" adds substitution.
MIRROR (B is A with terms exchanged): conversely. Needs a real inversion, not a difference.
CAUSE -> EFFECT (B produced by A): therefore, thus, consequently, as a result, hence, accordingly, so. Directional — the connective belongs in the sentence holding the OUTCOME.
EFFECT -> CAUSE (B explains A): after all, because, since, for.
EXAMPLE (B is one member of A's class): for example, for instance, to illustrate. Test membership, not specificity.
SPECIFICATION (B narrows A): specifically, in particular, notably, more precisely. Sharpens without needing to be an instance.
INTENSIFICATION (B is A, stronger): in fact, indeed, more than that, above all. NOT the same as for example.
RESTATEMENT (B is A reworded): in other words, that is, put differently. One new fact in B and the family is out.
CONCESSION (granting before turning): admittedly, granted, to be sure, of course, certainly. Decided by the sentence AFTER the blank.
SEQUENCE/TIME: first, next, then, subsequently, meanwhile, previously, eventually, finally. Available almost always; correct almost never.
CONDITION/ALTERNATIVE: otherwise, if so, in that case, alternatively. SUMMARY: in short, in sum, overall, on the whole, ultimately.
Two survivors look interchangeable? The test cannot credit two — one carries a constraint you have not spotted. Go find it.
Blank's sentence already has although / but / though / despite? A contrast in the blank is probably double-marking; the real relation is often addition or result.
A contrast, comparison, or "so...that" INSIDE idea B is not the relationship between A and B.
Elimination narrows; it does not confirm. Reread A + connective + B and check it asserts what you named.

Why it works — Why “it sounds right” is not evidence here

Reading is not passive intake. Comprehension works by building a connected model of the text, and when two sentences arrive without an explicit link the reader supplies one automatically — the bridging inference, one of the better-established findings in psycholinguistics, and ordinarily the thing that makes prose readable at all. On this question type it is the hazard. Insert “However” between two sentences and comprehension goes looking for a difference and finds one, because almost any two statements differ in some respect. Insert “Therefore” and it constructs a causal path. So the procedure of reading each choice into the passage and asking whether it sounds right is a test the mind is built to pass four times out of four, which is exactly why all four choices sound acceptable and why the confidence that follows carries no information. Naming the relationship before the choices are visible is the only point in the process at which the work is being done from the passage alone, before coherence-building starts helping toward whichever option happens to be under the eye. It is also why these items can be written at all: nobody is testing whether a student knows what “moreover” means. They are testing whether a relationship was committed to before four were offered.

Traps — 6

Half-sentence direction flip

Predicting from idea B only as far as the first comma, when the clause after the comma reverses it. The digital SAT puts the turn in the second clause often enough that reading idea B to the full stop should be a standing check rather than a piece of advice. This trap is worse than having no prediction at all, because a prediction built from half a sentence gets held with the same confidence as one built from the whole of it.

Family right, member wrong

The direction is correct and the word is not: “however” where the logic requires the concessive “nevertheless,” “for example” where it requires the intensifier “in fact,” “similarly” where there is only one subject to be similar to. This is where nearly all the remaining loss sits once prediction is habitual, because sorting by family cannot separate two choices that share a family.

Backwards arrow

Putting a cause → effect connective in the sentence that states the cause, or a reason-giving connective in the sentence that states the result. The two sentences really are causally related, so the family feels confirmed and the direction never gets checked. The test takes one second: does the blank’s sentence hold the reason, or the outcome? “Therefore” goes with the outcome.

Concession without an obstacle

Choosing nevertheless, nonetheless, admittedly, or to be sure because the passage’s tone sounds like it is granting something. A concessive asserts that B holds in spite of A, which requires A to give a reason to expect that B would not. If the defeated expectation cannot be said out loud, there is no obstacle and the word is wrong — and it will still be the most sophisticated-sounding option on the screen.

Chronology for logic

A date, a decade, or a narrative order in the passage pulls meanwhile, subsequently, or previously into the blank. Time words are almost always available, because things do happen in an order, and almost always wrong, because asserting order means declining to assert the logical link the passage has just built. Availability is not fit.

Inside-the-sentence relationship

Idea B contains its own contrast (“unlike humans”), comparison (“much as a person does”), or cause (“so crowded that”), and that internal relationship gets read off as the one the blank must name. The connective governs A to B, whole sentence to whole sentence. A contrast inside B is not a contrast between A and B, and this is the trap that most reliably catches a reader who is otherwise doing everything right.

The card

RHETORICAL SYNTHESIS — reference card
Read the GOAL first, then the notes. The bullets are a parts bin, not a passage.
Decompose the goal: verb (the act) + objects (which notes) + audience clause + concession.
Count the requirements before reading a choice. Two is normal. The second is the one dropped.
Every choice will be accurate. Accuracy is not the criterion — performing the act is.
EMPHASIZE X: X must be the main clause's assertion. Negate the sentence; what survives was never the point.
COMPARE A and B: both present AND the relationship stated. Two facts side by side is a list.
EXPLAIN WHY: a cause or mechanism, not a restatement of the thing being explained.
METHODOLOGY = what was done. FINDINGS = what came out. CONCLUSION = what it meant. Three answers.
"Unfamiliar with X" -> define X in the sentence. "Already familiar with X" -> use X as the anchor, don't explain it.
"Uses relevant information from the notes" is a requirement: a perfect sentence built on an outside fact is wrong.
Keep the notes' hedges. "Suggests" must not become "proves"; "some argue" must not become an assertion.
Whatever the goal concedes (despite / although / even though) must appear AND must be subordinate.
Two survivors? Pick the one that names the specific bullets, not the one that reads well in general.

Why it works — Why emphasis is a fact about grammar, not about tone

A sentence asserts what its main clause says. Everything packaged as a modifier — a relative clause, an appositive, a participial phrase, a subordinate clause — is handed to the reader as already settled, something to accept in passing on the way to the point. There is a three-second test that makes this visible: negate the sentence and see what survives. Take "The modular design, which cut construction time in half, was chosen mainly for its seismic performance." Negate it — "was NOT chosen mainly for its seismic performance" — and the halved construction time is still standing there, completely untouched, because it was never what the sentence claimed in the first place. Whatever survives negation is background, and background cannot be emphasis. This is the mechanism that lets the test build a distractor containing every single word the goal named and still have it be, unambiguously, the wrong answer: the fact is in the sentence, and the sentence is about something else. It is also why "but I checked, and that fact IS in the choice" is not a defence. Checking that a fact is present verifies the wrong property.

Traps — 6

True but off-goal

The choice is accurate against the notes and performs a different act — it describes where the goal said explain, states a similarity where the goal said difference, gives the findings where the goal said methodology. This is the default wrong answer on this question type and it accounts for more losses than the other five combined. The reason it works: accuracy is checkable against a bullet, and checking produces the physical sensation of being right. The fix is to name the act the choice performs before deciding whether it is the act you were asked for.

Buried emphasis

The required element is in the sentence, and it is in a relative clause, an appositive, or a subordinate clause, so the sentence asserts something else. Run the negation test: negate the main clause and see whether the required fact survives untouched. If it does, it was background, and background is not emphasis. This is the trap that catches students who have learned to check that the right material is present — because presence is exactly what it satisfies.

Half-comparison

On a compare or contrast goal: only one of the two things appears, or both appear with no relationship stated between them. Two facts joined by a comma or an "and" are a list. The variant that survives furthest up the score range is the WRONG-DIMENSION comparison — both things present, an explicit relationship stated, on a property the goal did not name. A goal asking you to compare two fibres in terms of performance is not satisfied by a flawless contrast of how they are manufactured.

Plausible outsider

A sentence that performs the goal cleanly using a fact that is not in the notes. It is often the best-written choice on the item, because it was constructed by someone who knew the subject, and it frequently matches the goal verb better than the credited answer does. The stem's second clause — "uses relevant information from the notes" — is the requirement it fails. Before accepting any choice, put a finger on the bullet behind each of its claims; the one you cannot find is the answer to the question of why it is there.

Attribution collapse

A note that reads "suggesting that," "may indicate," "has been proposed," or "some historians argue" turns up in a choice as a flat assertion. The hedge is information, and deleting it changes the claim into a stronger one the notes do not license. The mirror error is equally wrong and much rarer: a note that states something outright, softened by a choice into a "may." Both are settled by rereading one bullet.

Conceded-half emphasis

The goal itself grants something — "despite being the weaker fibre," "although the sample was small," "even though the technique is slower" — and the choice makes that granted material its main clause. The result is accurate about both facts and has the emphasis exactly backwards. Whenever a goal contains "despite," "although," or "even though," the conceded material must appear in the credited sentence and must sit in a subordinate position, and this trap is the most-chosen wrong answer on those items.

The card

LINEAR EQUATIONS IN ONE VARIABLE — reference card
Solve order: clear fractions -> distribute -> combine like terms -> x's one side, numbers the other -> divide by the coefficient.
Push to Ax = B. A != 0: one solution, x = B/A. A = 0 and B != 0: no solution. A = 0 and B = 0: infinitely many.
ax + b = cx + d: one solution iff a != c; NO solution iff a = c AND b != d; INFINITELY many iff a = c AND b = d.
Same slope, different intercept -> never meet. Same slope, same intercept -> meet everywhere.
Simplify both sides FULLY before comparing: in 4(kx - 3) the coefficient is 4k; in 9x + 7 - x it is 8.
A minus in front of parentheses belongs to the multiplier: 7 - 2(x - 4) = 15 - 2x.
Clearing fractions reaches every term on both sides, including the lone number.
Vanishing x-terms do NOT mean x = 0 — read the numeric statement they left behind.
Before the algebra, write down the exact quantity asked. After the algebra, read it again.
Check by substituting into the ORIGINAL equation, not into your rearranged version.

Why it works — Why one comparison decides all three outcomes

A linear equation in one variable is the question "where do these two lines meet?" asked without the graph: ax + b on the left is one line, cx + d on the right is another, and a solution is an x at which their heights agree. Two lines in a plane cross once, never, or coincide — there is no fourth possibility, which is precisely why a linear equation cannot have two solutions or three. Which case you are in is settled by whether the slopes are equal, because equal slopes are exactly the condition under which the vertical gap between the two lines never changes: if that fixed gap is anything other than zero they never touch, and if it is zero they were the same line all along. The coefficient comparison is not an extra rule bolted on top of the algebra — it is the algebra, read one step before it finishes.

Traps — 6

Unsigned distribution

Distributing the number in front of parentheses while leaving its minus sign behind: reading 7 − 2(x − 4) as 7 − 2x − 8 instead of 15 − 2x, or −3(x − 4) as −3x − 12 instead of −3x + 12. The multiplier is the whole signed quantity. This one error accounts for more lost points on this skill than every conceptual misunderstanding combined.

Half the rule

Checking the x-coefficients and stopping. Matching coefficients are shared by both degenerate cases, so they narrow the answer to "no solution or infinitely many" and decide nothing further — the constants pick the winner. The mirror-image version is checking only the constants: identical constants with different coefficients still give exactly one solution, usually x = 0.

Cancelled variable read as zero

The x-terms vanish and the student writes x = 0. A vanished variable means the equation has collapsed into a numeric statement, and the only remaining job is to judge whether that statement is true (infinitely many) or false (none). x = 0 is a different claim entirely — a specific value that can be substituted and checked.

Comparing unsimplified sides

Pulling k straight out of 4(kx − 3) as though the coefficient were k rather than 4k, or matching against the 9 in 9x + 7 − x when the side's actual coefficient is 8. The comparison rule is exactly right and applied to the wrong numbers, which is why the resulting answer feels earned.

Partial clearing of fractions

Multiplying through by the common denominator and missing something: the lone constant on the far side, or the surviving factor when the denominator divides the LCD unevenly — multiplying (x − 3)/2 by 4 gives 2(x − 3), not (x − 3). Every term on both sides, every time.

Answering the unasked quantity

Solving correctly for x when the question asked for 2x − 3, or for adult tickets when it asked for student tickets, or for the discounted price when it asked for the original. The intermediate value is always one of the four choices. This is the most reliably placed distractor on the Math section, and it is not a knowledge failure.

The card

LINEAR EQUATIONS IN TWO VARIABLES — reference card
y = mx + b: m = rate of change (y-units per x-unit); b = value of y when x = 0.
From two points: m = (y2 - y1)/(x2 - x1) -- same order in numerator and denominator.
Point-slope: y - y1 = m(x - x1). Use when you have a slope and any one point.
Standard form Ax + By = C: slope = -A/B, NOT A. Don't read it off directly.
Before writing symbols, name what m and b mean in the actual scenario, out loud.
Table linear? Check that consecutive differences in y over differences in x are constant.

Why it works — Why slope is a genuine constant, not just a formula

A relationship is linear precisely when its rate of change is the same between any two points you pick on it — that's the definition, not a property you have to separately verify once you know it's linear. That's why you can compute slope from any two points on a line and get the same number every time, and it's why a table is linear exactly when consecutive differences in y, divided by consecutive differences in x, come out constant. If they don't, the relationship isn't linear, and the whole slope-intercept toolkit stops applying.

Traps — 5

Slope/intercept swap

Assigning the fixed value to m and the rate to b, or vice versa — most common when the fixed fee is stated first in the sentence and gets grabbed as "the first number in the equation" without checking which role it actually plays.

Sign error from mismatched point order

Computing slope as (y₂ − y₁)/(x₁ − x₂) — flipping the order in the denominator but not the numerator. The rule isn't "subtract in order"; it's "use the same order in both places."

Unit mismatch

Using a rate given in one unit (per year, per dozen, per six miles) directly as the coefficient on an input variable measured in a different unit (per month, per item, per mile) without converting first.

Standard-form slope trap

In Ax + By = C, the slope is −A/B, not A — pulling the coefficient straight off the equation without solving for y first is a reliable way to get the sign or the reciprocal wrong.

Assumed linearity

Writing a slope-intercept equation for a table without first checking that consecutive differences in y over differences in x are actually constant — some Digital SAT tables are deliberately not linear.

The card

LINEAR FUNCTIONS — reference card
f(x) is NOT f times x. f names the rule; the parentheses hold the input.
f(a) = b means input a gives output b, and means the point (a, b) is on the graph.
Number INSIDE the parentheses -> evaluate (substitute). Number across the = -> solve.
f(0) = the y-intercept. A zero of f = the input where f(x) = 0. Opposite ends of the line.
Model from a description: f(x) = (signed rate)(x) + (value at x = 0).
Name the input and output units in writing FIRST, then find the rate, then the start.
Convert every rate to output-units per ONE input-unit before it enters the coefficient slot.
Two points: f(a)=p, f(b)=q -> m = (q - p)/(b - a), same order top and bottom, then solve for b.
Change over an interval = slope x interval length. Not the slope, not a level.
f(x)+k: intercept +k. f(x+k): intercept +mk, graph moves LEFT. k*f(x): both scale. f(kx): slope only.
Constant AMOUNT per step = linear. Constant PERCENT per step = not linear, toolkit off.
Substitute negative inputs inside their own parentheses: -3(-3) = +9.
Before answering: reread the last line and confirm which quantity was asked for.

Why it works — Why the coefficient slot means "per one"

A slope is not a number. It is a ratio carrying units — output-units per one input-unit — and the position it occupies in f(x) = mx + b is what makes the "per one" mandatory rather than stylistic. Multiplying m by x means "take one unit's worth of change, m, and take x of them." If m is secretly worth four minutes rather than one, then mx is counting four-minute blocks while x counts minutes, and the two halves of the expression are measuring different things. Nothing in the arithmetic will complain. You will get a clean, confident number that is wrong by a factor of four. This is also why the unit check is the cheapest error-detector available in the Math section: write the units underneath the symbols and read across — (liters per hour)(hours) + (liters) = liters. If the multiplication does not cancel down to the output's units, the model is wrong before a single value is substituted into it. Students who lose points to unit slips are almost never students who cannot convert; they are students who never wrote the units down and so had nothing to check the model against. And the deeper reason the two-point method is legal at all: a function is linear exactly when that rate is the same between every pair of points on it. That is the definition of linear, not a separate property to verify afterwards. It is why any two points give the same slope, why "the average rate of change of f over the interval" is simply the slope when f is linear, and why a description saying a quantity falls by 12% each year is not describing a linear function at all — a constant percentage is a constant multiplier, not a constant amount, and none of the toolkit on this page applies to it.

Traps — 6

Evaluate/solve swap

The number was given as an output and got substituted as an input, or the reverse. f(−6) and f(x) = −6 use the same two symbols to ask opposite questions. The rule is positional: a number inside the parentheses is an input, so substitute; a number across the equals sign is an output, so solve. This is the highest-frequency wrong answer on the skill, and the test writes a distractor for it on almost every item.

Raw-rate transplant

Dropping a rate stated per four minutes, per six items, or per half hour straight into the coefficient slot, which is defined as per one. Any rate whose "per" does not match the input variable's unit has to be converted first — and the conversion has a direction, so 90 liters per 4 minutes becomes 1,350 per hour, not 5,400 and not 22.5.

Slope/intercept swap

Assigning the fixed amount to the coefficient and the rate to the constant. It happens most when the sentence states the fixed amount first — "a $75 setup fee plus $54 for every 4 posters" — because the first number in the sentence gets grabbed as the first number in the equation. Naming both roles explicitly before writing a symbol is the whole fix.

Intercept/zero confusion

f(0) is the output when the input is zero — the y-intercept, the initial value. A zero of f is the input that makes the output zero — the x-intercept, found by solving f(x) = 0. Items phrase these as "the initial amount," "where the graph crosses the axis," "the zero of the function," "when will it reach zero," and the two answers are almost never the same number.

The wrong quantity answered

The algebra is finished and every line of it is correct, and the number reported is b, or the slope, or the input, when the question asked for f(2), or the year, or the total. This is the characteristic error at the top of the score range: nothing was misunderstood, the last line of the question was simply not reread. Underline the quantity the question names before starting the work.

Sign slip on a negative input

Substituting a negative input into a rule with a negative coefficient and losing exactly one of the two signs: for f(x) = −4x + 7, f(−5) = −4(−5) + 7 = 20 + 7 = 27, not −20 + 7 = −13. Two negatives, one chance to drop one. Writing the input inside its own parentheses is the mechanical prevention, and a distractor built on precisely this slip appears on most hard items in this skill.

The card

SYSTEMS OF TWO LINEAR EQUATIONS — reference card
A solution is an ordered pair (x, y) making BOTH equations true at once. Only three outcomes: one, none, infinitely many.
Substitution: use when a variable is isolated or has coefficient 1 or -1.
Elimination: scale until one variable's coefficients are opposites, then ADD. Prefer multiplying an equation by -1 and adding over subtracting.
Eliminate the variable you were NOT asked about.
Mirrored coefficients? Adding gives x + y, subtracting gives x - y -- often the whole answer without solving.
Put BOTH equations into Ax + By = C before comparing anything. 8y = 3x + 5 has A = -3, not 3.
Parallel test (safe form): A1*B2 = A2*B1. If that fails, exactly one solution, whatever the constants are.
Then the constants decide: whole triple proportional -> infinitely many. Coefficients proportional, constants not -> none.
One-motion recipe: find the multiplier taking eq1's coefficients to eq2's, apply it to eq1's constant. Matches -> infinitely many. Doesn't -> none.
The multiplier is the tool, not the answer. Apply it to the specific coefficient the stem asks about.
Desmos: type both equations exactly as printed, click the crossing. Zoom out before concluding 'no intersection'.
Desmos cannot tell one line drawn twice from two parallel lines, and cannot answer 'for what value of k'. Use the coefficient rule.
Check in the equation you did NOT use to find the second variable.
Reread the stem before answering: x, y, x + y, x - y, y - x and the wrong unit are all in the choice list already.

Why it works — Why elimination is legal, and why the no-solution and infinite-solution rules are not extra rules

Everything in this lesson follows from one fact: if a pair (x, y) makes both equations true, then it makes any multiple of either equation true, and it makes their sum true. That is just adding equals to equals. If 4x + 3y really is 6 and 5x − 2y really is 19 for your pair, then 8x + 6y really is 12 and 15x − 6y really is 57, and adding those gives 23x = 69, which your pair must also satisfy. Elimination is not a trick for making variables vanish; it is the observation that a solution to a system is automatically a solution to every equation you can build by scaling and adding the originals — so you are free to build the most convenient one, the one with a variable missing. And because every operation involved is reversible — multiplying by a nonzero constant, adding an equation you can subtract back — nothing is gained or lost along the way. Linear systems have no extraneous solutions, unlike the squaring steps that create them elsewhere in algebra; a pair that survives your elimination genuinely solves the original system. Now push the same machinery one step further and the two degenerate cases fall out for free. Suppose the second equation is a constant multiple of the first. Then the combination that eliminates x eliminates y at the same moment, because both coefficients were scaled by the same factor, and what is left is a statement containing no variables at all. If the constants were scaled by that same factor too, that statement is 0 = 0 — true for every pair on the line, which is what "infinitely many solutions" means. If the constants were not, the statement is 0 = some nonzero number — true for no pair whatsoever, which is what "no solution" means. Mechanism: the ratio test is not a separate fact to memorise alongside elimination. It is a description of what elimination returns when the second equation carries no information the first did not already contain, and the constants are the only place that missing information could have been hiding.

Traps — 6

Proportional coefficients, forgotten constant

Seeing that one equation's x- and y-coefficients are a clean multiple of the other's and concluding "infinitely many solutions" without touching the constants. Proportional coefficients establish only that the lines are parallel; the constants decide whether that means one line or two. This is the most common wrong answer on the whole skill, and the fix is mechanical: apply the multiplier to the constant too, every single time, before naming the case.

Ratio test on mismatched forms

Comparing coefficients from equations that are not written the same way — one solved for y, one in Ax + By = C, one with the constant on the left. "8y = 3x + 5" has an x-coefficient of −3, not 3, and a system where that sign is misread produces a fully coherent chain of correct steps to a wrong number. Rearrange both equations into identical form first; it costs one line and removes the trap entirely.

Subtraction sign collapse

Subtracting one equation from another and negating only the first term of the second equation, or only the left-hand side. Subtracting 20x − 8y = 76 from 20x + 15y = 30 means every term changes sign, including the 76. The countermeasure is structural rather than attentional: multiply the whole equation by −1 as its own visible step, then add.

Scale factor written as the coefficient

On parameter questions, putting the multiplier itself into the answer slot rather than the number it produces. If the multiplier from the first equation to the second is 3, then a system 2x + 5y = 8 and ax + 15y = b has a = 6, not a = 3. The scale factor is the tool; the answer is what the tool produces when applied to the specific coefficient the stem asked about.

Answered the wrong quantity

Solving the system perfectly and then reporting x when the stem asked for x + y, the number of tickets when it asked for the dollars they raised, or y − x when it asked for x − y. This costs more points at the top of the scale than any conceptual error, because the distractor list on a live item is deliberately built from x, from y, from the reversed difference, and from the other combination — the wrong quantity is always on the screen, already computed, waiting to be selected.

Desmos decimal drift and the off-screen crossing

Two failures with the same source — trusting the picture past what the picture can show. An intersection readout of 0.6667 is a rounded 2/3, and typing 0.667 into a grid-in is not always accepted as that value. And the default window spans roughly −10 to 10, so a system whose solution is (−40, 120) shows two lines that appear never to meet. "I graphed it and there was no intersection" is a claim about the window, not about the system.

The card

LINEAR INEQUALITIES — reference card
Every equation step is legal. ONE exception: multiply or divide by a negative -> reverse the symbol.
Adding or subtracting a negative never flips anything. Only multiplying/dividing by one does.
Strong path: move variables to the side that keeps the coefficient positive, and never divide by a negative.
Check in 10 seconds: one number inside your answer must satisfy the original; one outside must fail it.
Two variables = a region. Boundary solid for ≤ and ≥, dashed for < and >.
Which side? Test a point off the line — (0, 0) when the line misses the origin. True -> shade that side.
Never read shading off the symbol unless y is alone with a positive coefficient. 4x - 2y > 10 shades BELOW.
System = intersection of the regions. Every constraint, simultaneously. Overlap only.
Language: at least / no fewer than = ≥. At most / no more than / cannot exceed = ≤. More than / fewer than = strict.
Whole-number answers: round from the constraint (greatest that fits -> down; least that suffices -> up), not from the decimal.
Max or min over a bounded region sits at a corner: drop the symbols and solve the two boundaries as equations.
Never multiply or divide both sides by an expression whose sign you do not know.

Why it works — Why the direction reverses, and when the step is illegal

An inequality is a claim about position on the number line: a < b says nothing more than "a sits to the left of b." Adding the same amount to both sides slides both numbers the same distance in the same direction, so their left-right order is untouched — which is exactly why adding and subtracting never flip anything, not even when the thing being added is negative. Multiplying both sides by a negative number does something categorically different: it reflects the whole number line through zero. Everything that was to the left of something is now to the right of it. 2 < 5, but multiply both by −1 and −2 > −5. Dividing by a negative is multiplying by a negative reciprocal, so it reflects too. The symbol reverses because the reflection reversed the order the symbol was describing — the flip is not a rule bolted onto the algebra, it is the algebra reporting what just happened. The same fact tells you when the step is not allowed at all: if you multiply or divide both sides by an expression whose sign you do not know, anything containing a variable, then you do not know whether you slid the line or reflected it, so you cannot know which symbol to write. That step is illegal rather than merely risky, and it is the failure mode that costs strong students this question type.

Traps — 6

The unflipped division

Dividing or multiplying both sides by a negative number and leaving the symbol as it was — turning −14x ≤ −28 into x ≤ 2 rather than x ≥ 2. The result is a range of exactly the right size pointing the wrong way, which is far harder to notice than a wrong number.

Flipping the number instead of the symbol

Remembering that something has to flip and applying it to the constant: negating the right-hand side to get x ≥ −2 instead of reversing the direction to get x ≥ 2. The rule reverses the relationship between the two sides, never the value on either side.

The phantom flip

Reversing the symbol when adding or subtracting a negative number, because a minus sign appeared. Adding −7 to both sides slides both by the same amount and preserves the order; only multiplication and division by a negative reflect it.

Solid/dashed mismatch

Drawing a dashed boundary for ≤ or ≥, or a solid one for < or >. In point-testing form the same error is counting a point that lies exactly on the boundary as a solution to a strict inequality — and SAT items place a distractor precisely on that line often enough that it should be the first thing checked, not the last.

Wrong half-plane from an unsolved form

Reading the shading straight off the symbol when the inequality has not been solved for y, or when y's coefficient is negative. In 4x − 2y > 10 the > shades below, because isolating y divides by −2 and reverses it. A test point never makes this error.

Union instead of overlap

Accepting a point that satisfies one inequality of a system and never checking the rest, or shading everything either region covers. A system's solution set is the intersection — every condition, simultaneously — and the distractors are built from points that clear one constraint convincingly and fail another.

The card

EQUIVALENT EXPRESSIONS — reference card
Equivalent = same value at every input where both are defined. Prove it or break it by substituting x = 2 (never 0, never 1, never an excluded value).
Exponents: x^a·x^b = x^(a+b); x^a/x^b = x^(a-b); (x^a)^b = x^(ab); (xy)^a = x^a·y^a.
x^0 = 1; x^(-a) = 1/x^a; x^(1/n) = nth root of x; x^(m/n) = (nth root of x)^m.
The outer exponent hits the coefficient too: (3x^4)^3 = 27x^12, not 3x^12.
x^a + x^b never simplifies. Exponent rules govern multiplication and division only.
Factor in this order: common factor -> difference of squares -> trinomial -> ac-method grouping.
a^2 - b^2 = (a-b)(a+b). a^2 + b^2 does NOT factor over the reals. (x+4)^2 = x^2+8x+16, not x^2+16.
Cancel FACTORS, never terms. If a plus or minus is still loose on either side, you have not finished factoring.
Three forms: standard ax^2+bx+c shows the y-intercept c; factored a(x-r)(x-s) shows the zeros; vertex a(x-h)^2+k shows the vertex (h,k) and the min/max k.
Vertex x-coordinate straight off standard form: x = -b/(2a). Sign of h: a(x+4)^2 has h = -4.
Completing the square with a != 1: factor a out of the x-terms FIRST; the number subtracted inside leaves multiplied by a.
'Equivalent for all x except c' names a zero of the denominator — and usually the factor that cancels. Free information.
Remainder form: multiply both sides by the divisor, then substitute the x that zeroes it.
Radicals to fractional exponents on sight. sqrt(x^2) = |x| — check whether the stem stipulates x > 0.
Answer the letter asked: a+b is not a; k is not h; the minimum VALUE is not where it occurs.

Why it works — Why every legal move is either multiplying by 1 or adding 0

Rewriting an expression is constrained by exactly one requirement — the value at every input has to stay the same — and there are only two ways to change how something looks without changing what it is worth: multiply it by something that equals 1, or add something that equals 0. Every technique in this lesson is one of those two wearing a disguise. Building a common denominator multiplies a fraction by (x + 2)/(x + 2), which is 1. Rationalising a denominator multiplies by conjugate-over-conjugate, which is 1. Cancelling a common factor f divides top and bottom by f, which is multiplying the fraction by (1/f)/(1/f), which is 1. Completing the square adds and subtracts the same number — adding 0, written as 9 − 9 so that part of it can be folded into a square. Once you can see which of the two a move is, you can tell instantly whether it is legal. "Cancelling the x in (x + 3)/(x + 7)" is neither: it deletes a term, which is not multiplying by 1 and not adding 0, and it changes the value. This also explains the domain restriction that trails every cancellation. The disguised 1 you multiplied by, (1/f)/(1/f), is only equal to 1 where f ≠ 0. So the equivalence holds everywhere except there — and the restriction is not an afterthought bolted onto the answer, it is a record of which disguised 1 you used.

Traps — 6

Cancelling across a sum

Striking a symbol that appears on both sides of a fraction bar without checking that it is a factor of the whole side rather than a term inside a sum. (x + 3)/(x + 7) is finished; there is nothing to cancel. The test writes whole answer choices out of this error, because it produces a shorter, cleaner-looking expression, and shorter feels like progress. Rule: if the top or the bottom still has a plus or minus sign that is not inside a bracket, you have not factored yet, and until you have factored you cannot cancel anything.

The forgotten coefficient

Raising the variable to the outer power and leaving the number alone: (3x⁴)³ read as 3x¹² instead of 27x¹². An exponent on a product reaches every factor inside the parentheses, coefficients included. The same slip in reverse gives (2x)² = 2x² instead of 4x².

Add/multiply exponent swap

Adding exponents where the rule multiplies, or the reverse: (x⁴)³ read as x⁷, or x⁴ · x³ read as x¹². The distinction is worth saying in words rather than symbols — multiplying like bases piles copies up, so the counts add; a power of a power makes several identical groups of copies, so the counts multiply. The same confusion produces adding on division, which turns x¹²/x⁵ into x¹⁷.

The half-completed square

Completing the square correctly and then failing to multiply the subtracted number by the leading coefficient on its way out of the parentheses — writing 3(x + 4)² + 41 − 16 instead of 3(x + 4)² + 41 − 48. Occurs only when a ≠ 1, which is exactly when the test uses it, and it produces a wrong constant that sits in the answer choices looking entirely reasonable.

Phantom factorisation

Applying a pattern that does not exist, or stopping halfway through one that does. x² + 25 does not factor over the reals. (x + 4)² is x² + 8x + 16, not x² + 16. And x⁴ − 16 is a difference of squares whose first factor is another difference of squares: (x² − 4)(x² + 4) = (x − 2)(x + 2)(x² + 4). Stopping at the first step is a real answer choice on "completely factored" items.

Answering the wrong letter

Correct algebra, wrong quantity handed in. The item asks for a + b and gets a; asks for k and gets h; asks for the minimum value and gets the x where it occurs; asks for the value of the constant and gets the value of x. This trap does not test algebra at all, which is precisely why it survives at the top of the score range — it catches people whose work was right.

The card

NONLINEAR EQUATIONS AND SYSTEMS — reference card
Factoring concludes nothing unless one side is 0. AB = 0 -> A = 0 or B = 0; AB = 8 -> nothing.
Route 1 square root: no plain x term, or a squared binomial. Write the +/- . Two answers.
Route 2 factoring: small integers. a != 1 -> find the pair with product a*c, sum b; split; group.
Route 3 formula: x = (-b +/- sqrt(b^2 - 4ac)) / (2a). Always works. Write a, b, c down first.
Route 4 complete the square: when the question wants the vertex. x^2 + bx -> add and subtract (b/2)^2.
Route 0 Desmos: graph both sides, read intersections. Weakest when a letter parameter is involved.
D = b^2 - 4ac. D > 0 two real; D = 0 one (double root at x = -b/2a); D < 0 none real.
Translations of D: exactly one solution / no x-intercepts / tangent to / system has one solution.
Sum of roots = -b/a. Product = c/a. Only in standard form, only with the real a.
Line + parabola: substitute, collect to ONE side, then solve or take D. Collected b is a new number.
Squaring and clearing denominators are one-way. Every candidate goes back into the ORIGINAL.
Radical sign filter: sqrt(A) >= 0 always, so sqrt(A) = B forces B >= 0. Kills candidates before algebra.
Rational: write the excluded values BEFORE solving. A root equal to one of them means no solution.
Never divide both sides by a variable expression - you delete a root. Factor instead.
(A + B)^2 = A^2 + 2AB + B^2. The cross term is where the points are.
Parameter questions almost always have TWO branches. Produce both, then apply the constraint.
Answer the target: positive solution / sum / y / x + y / the parameter - not 'the roots'.

Why it works — Why "make one side zero" is not a ritual, and what the discriminant actually is

Factoring is not a way of solving an equation. Factoring is a way of rewriting one side as a product — and a product only tells you something when it equals zero, because zero is the one number that cannot be built from two nonzero real factors. That is the whole content of the zero-product property, and it is why (x − 5)(x + 2) = 0 collapses into two easy equations while (x − 5)(x + 2) = 8 collapses into nothing. The instruction "set it equal to zero first" is not a step in a procedure; it is the precondition that makes the next step legal. The discriminant has an equally concrete origin. Complete the square on the general equation ax² + bx + c = 0 and you get (x + b/(2a))² = (b² − 4ac)/(4a²); taking the square root of both sides produces the quadratic formula, and the only thing in that derivation that can go wrong is the number under the radical. If b² − 4ac is positive, its square root is a real number, the ± gives you two different values, and the parabola crosses the x-axis twice. If it is zero, the ± adds and subtracts nothing, so the two roots collapse into one at x = −b/(2a) — which is the x-coordinate of the vertex, meaning the parabola touches the axis exactly at its turning point. If it is negative, no real number squares to it, so there is no real solution and the parabola misses the axis entirely. The discriminant is not a memorised rule about how many answers there are. It is the answer to a single question — what is under the root — and every "exactly one solution," "no real solutions," and "tangent to" question on this skill is that same question in a costume.

Traps — 6

Zero-product on a nonzero product

Factoring a side and then setting each factor equal to whatever is on the right: from (x − 5)(x + 2) = 8, writing x − 5 = 8. Only zero licenses that move, because only zero cannot be written as a product of two nonzero numbers. Expand, subtract, get a zero, then factor. The distractor for this trap is usually the root you would get from the untouched factors, which makes it look reassuringly like an intended answer.

The root nobody checked

Squaring both sides or clearing a denominator produces candidates, not solutions. Both operations are one-way: they can create roots the original equation never had. Any radical equation and any equation with a variable in a denominator is unfinished until every candidate has been substituted back into the ORIGINAL equation — and on rational equations, until every root has been compared against the excluded values written down at the start.

The root you deleted

Dividing both sides by an expression containing a variable, which silently assumes that expression is nonzero. x² = 5x divided by x gives x = 5 and loses x = 0. Same for cancelling (x − 2) from both sides of an equation. Move everything to one side and factor; the factor you were about to divide away is a solution.

One branch of two

Taking a square root and writing only the positive case. x² = 49 has two solutions; (2x − 5)² = 49 has two; (6 − m)² = 36 has two; k − 2 = ±6 has two. The ± is not decoration, and parameter questions in particular are built so that the second branch is the one a hurried solver never writes down. If a constraint in the problem (x < 0, b < 0, k > 0) is going to eliminate one branch, you still have to produce both first in order to eliminate anything.

Half a square

Squaring a two-term expression term by term and losing the cross term: writing (x − 5)² as x² + 25, or (√x + 2)² as x + 4. The identity is (A + B)² = A² + 2AB + B². This is the reason the method says isolate the radical before squaring — a one-term side has no cross term to lose.

Right root, wrong question

Solving the equation perfectly and then writing down a number the question did not ask for: both roots when it asked for the positive one, x when it asked for y, the roots when it asked for their sum, a root when it asked for the parameter k. The answer choices are populated with these on purpose, so recognising your number in the list is not confirmation. Name the target before you start and check your final line against the target, not against the choices.

The card

NONLINEAR FUNCTIONS — reference card
Three forms, one parabola. Pick the form that already shows what the question asks.
Standard ax² + bx + c -> y-intercept is c; opens up if a > 0; axis at x = -b/(2a).
Factored a(x - r)(x - s) -> zeros are r and s; axis at x = (r + s)/2, their midpoint.
Vertex a(x - h)² + k -> vertex (h, k); the max/min VALUE is k; axis at x = h.
Every sign inside a parenthesis is the opposite of the number it names: (x + 5) -> -5.
Fast vertex without completing the square: x = -b/(2a), then substitute back for k.
Completing the square: factor a out of the x-terms only, then distribute a back over BOTH pieces.
Discriminant b² - 4ac: > 0 two x-intercepts; = 0 exactly one (vertex on the axis); < 0 none.
"Exactly one solution", "no real solutions", "meets at one point" -> set up the discriminant.
Exponential a·b^x: a = output at x = 0; b = multiplier per one unit of x; b = 1 + r.
1.08 -> +8%. 0.82 -> -18%. 2 -> doubles. 0.5 -> halves. Never read b as the percentage.
Multiplier applies once per p units -> exponent is t/p. The exponent counts PERIODS.
Constant AMOUNT per step = linear. Constant PERCENT or factor per step = exponential.
Table test: constant first differences linear; constant second differences quadratic; constant ratios exponential.
Transformations: f(x)+k up; f(x-h) RIGHT h; -f(x) flips over the x-axis; inside runs backwards.
a·b^x + d has a horizontal asymptote at y = d; a positive exponential never reaches zero.
End behaviour = leading term only. Even degree ends match; odd degree ends oppose; sign of a sets the right end.
Zeros in factored form: even exponent touches and turns, odd exponent crosses.
Sum of the solutions of ax² + bx + c = 0 is -b/a; their product is c/a.
Before answering: is the question asking WHERE the extreme happens (an x) or WHAT it is (a y)?

Why it works — Why the vertex is free once you can see the square

Squaring destroys sign: (−3)² and 3² are the same number. So in f(x) = a(x − h)² + k, two inputs the same distance either side of h hand the squaring the same magnitude and produce the same output. That symmetry is not a property of parabolas to be memorised separately — it is the squaring operation, made visible, and three things the exam tests fall straight out of it. First, the axis of symmetry is x = h, always. Second, the extreme value is at x = h, because (x − h)² is never negative: it contributes nothing at x = h and something strictly positive everywhere else, so k is the smallest output the function can produce when a > 0 and the largest when a < 0. Third, if the parabola has two x-intercepts, they are mirror images across that axis, which is why the vertex sits exactly at their midpoint — and why you never need to complete the square on a factorable quadratic just to find where its maximum happens. Even the formula is the same fact in different clothes: expanding a(x − h)² + k gives ax² − 2ahx + (ah² + k), so b = −2ah, and x = −b/(2a) is that identity solved for h. One equation, not two. The exponential half has an equally short reason, and it explains the most expensive error on this skill. Writing a·b^t says: start at a, then multiply by b once for every unit of t. The exponent is therefore not a duration, it is a COUNT of multiplications. If the multiplier applies once every 5 years and t is measured in years, then t years is t/5 multiplications and the exponent has to be t/5 — that division is not a formula, it is the answer to "how many periods have gone by?". The same reasoning is why a percentage cannot be spread across a period the way a dollar amount can. Losing 12% every 5 years is not losing 2.4% a year, because each year's loss would come out of a base the previous year already shrank. Four successive 12% losses leave 0.88^4 ≈ 0.60, about 60% of the original — not 100% − 48% = 52%. Every wrong answer built on "it falls 18% a year, so after 5 years it has fallen 90%" is correct arithmetic performed in the additive system while the model is running in the multiplicative one.

Traps — 6

Sign flip inside the parentheses

Vertex form is a(x − h)² + k and factored form is a(x − r)(x − s), so both forms SUBTRACT the number they name. A plus sign inside means the value is negative: 2(x + 5)² − 3 has vertex (−5, −3), and (x − 3)(x + 7) has roots +3 and −7. The rule is mechanical — every sign inside a parenthesis is the opposite of the number it names — and a distractor built on the un-flipped reading appears on almost every item that hands you one of these forms.

Multiplier read as a percentage

Reading the base of a·b^x as the rate of change. b = 1 + r, so 1.35 is +35% and not +135%, and 0.82 is −18% and not −82%. The direction is easy to get right and the size is easy to get wrong, which is why the exam writes both the "82%" and the "18%" versions into the same answer set. Say the base out loud as "keeps 100% and adds r" or "keeps 100% minus r" before choosing.

Period/exponent mismatch

A multiplier stated per 5 years, per 8 hours, or per 30 days dropped into the exponent as though it applied per one unit of the input. The exponent counts PERIODS, so a per-p-unit multiplier with input t needs exponent t/p — and the conversion has a direction, so tripling every 8 hours with the input in DAYS gives 3^(3d), not 3^(d/8) and not 3^(8d). Ask how many multiplications have happened by time t, and the exponent falls out.

Linear reasoning on a multiplicative model

Computing a percentage of the ORIGINAL amount and then adding or subtracting it repeatedly. A machine losing 18% a year from $18,000 does not lose $3,240 every year — it loses $3,240 in year one and less every year after, because each year's percentage comes off a base the previous year already reduced. The same error in the other direction turns "grows 10% a year" into "grows 30% over three years" when the true figure is 1.1³ = 1.331, a 33.1% rise.

Half-completed square

Completing the square correctly inside the parentheses and then failing to multiply the subtracted piece by a on the way out. For 3x² + 30x + 68, the correct conversion is 3(x + 5)² − 75 + 68 = 3(x + 5)² − 7; writing 3(x + 5)² + 68 − 25 = 3(x + 5)² + 43 forgets that the 25 was inside a bracket being multiplied by 3. Every wrong answer this produces is off by exactly (a − 1) times the correction, and it is the single most common arithmetic failure on this conversion.

The wrong coordinate answered

The algebra is finished and every line of it is correct, and the number reported is where the maximum happens when the question asked what the maximum is — or the height when it asked for the time, or the axis of symmetry when it asked for the vertex's y-value. This is the characteristic error at the top of the score range, and it is why a well-built item puts the x-coordinate, the y-coordinate, and often a zero into the same four choices. Before starting, underline the quantity the last line names.

The card

RATIOS, RATES, PROPORTIONS, UNITS — reference card
"per" = a fraction bar. "of" = multiply. Write every rate as a fraction with units.
Every conversion gives TWO fractions, both equal to 1. Pick the one that cancels.
Write units at every step. The surviving unit is the check — and it is free.
Areas square the factor, volumes cube it:
   1 m = 100 cm  ->  1 sq m = 10,000 sq cm  ->  1 cu m = 1,000,000 cu cm
Ratio a:b has a+b parts. Share of whole = a/(a+b). Comparison = a/b. Not the same.
Proportion: same quantity in the same position on both sides, THEN cross-multiply.
Proportional means y = kx, through the origin. A flat fee makes scaling illegal.
Inverse: xy = k. More workers, fewer hours, same product.
Rates that pool ADD. Never average two rates; never add two solo times.
Average speed = total distance / total time. Always.
Rate given per 100 km or per 4 items: reduce to a unit rate first, or carry it whole.
Off by 1,000 or 1,000,000 from your estimate = a unit error, not an arithmetic one.
Last check, every time: does the surviving unit match the unit the question asked for,
   and did it ask for the amount ADDED or the new TOTAL?

Why it works — Why a conversion factor is allowed to change the number

A statement like "1 meter = 100 centimeters" is not a recipe; it is an equation between two names for the same length. Divide both sides by 1 meter and you get 1 = 100 cm / 1 m — a fraction that is literally the number one, wearing units. Multiplying any quantity by it therefore cannot change the quantity. It can only change how the quantity is described, which is why the number in front is free to move while the physical amount is not, and why the orientation is a free choice: 1 m / 100 cm is the same number one, flipped over. The squaring rule then falls out in a single line and never has to be memorized separately. A square meter means (1 m)(1 m), and each of those meters is 100 cm, so 1 m² = (100 cm)(100 cm) = 10,000 cm². The factor gets used once per dimension — once for a length, twice for an area, three times for a volume — because an area is a product of two lengths and a volume is a product of three.

Traps — 6

Unsquared conversion factor

Converting an area with the length factor, or a volume with the area factor. One square meter is 10,000 square centimeters, not 100. Anything carrying "square" or "cubic" takes the factor to that power — including a squared unit hidden inside a rate, like grams per cubic centimeter or people per square mile, where the word is easy to read straight past at speed.

Upside-down factor

The conversion fraction used in the orientation that does not cancel. It is the cheapest trap to catch and the most expensive to miss, because the tell is immediate — the wrong unit survives — and the damage is not a small slip: a mis-oriented factor puts you off by the square of the conversion number, so a flipped minutes-to-hours factor lands you a factor of 3,600 away rather than 60.

Inverted rate

Using seconds-per-item where items-per-second was needed, or dollars-per-kilogram where kilograms-per-dollar was needed. The two are the same relationship and only one of them cancels the unit you are holding. This is the trap a magnitude check cannot catch when the conversion number is close to 1, which is why the cancellation has to be written rather than felt.

Part-to-part read as part-to-whole

Reading 5:3 as "5 out of 8" when the question wants the comparison, or as "5 out of 3" when it wants the share. Count parts first: 5:3 is eight parts, so 5/8 and 3/8 are the shares of the total and 5/3 is the ratio of one part to the other. The exam offers both numbers as choices on the same item.

Assumed proportionality

Scaling from a single data point in a scenario that contains a flat fee, deposit, base charge or starting amount. A "cost per unit" computed by dividing a total that includes a fixed charge is not a rate, and multiplying by it is not valid at any other quantity than the one it came from.

Averaged rates

Combining two rates by taking their mean, or combining two jobs by adding the times each takes alone. Rates add only when the things they measure genuinely pool — two pumps into one tank — and an average speed for a whole trip is total distance over total time, never the average of the leg speeds.

The card

PERCENTAGES — reference card
Percent means per hundred. 37% = 0.37. Convert once, at the start, then work in decimals.
The number after "of" (or after "than") is the base — the denominator, the 100%.
Percent change = (new − old) ÷ old. The denominator is ALWAYS the value you started from.
Work in multipliers: +p% -> ×(1 + p/100); −p% -> ×(1 − p/100). +8% -> ×1.08. −8% -> ×0.92.
"p% of x" = x(p/100). "p% more than x" = x(1 + p/100). "p% less than x" = "p% off x" = x(1 − p/100).
"150% of x" -> ×1.50. "150% more than x" and "increased by 150%" -> ×2.50. Different sentences, different multipliers.
Chains MULTIPLY: +25% then −30% is 1.25 × 0.70 = 0.875, a 12.5% fall. Never −5%.
Net change = combined multiplier − 1. 0.875 -> −12.5%. 1.19 -> +19%. Above 1 = increase, below 1 = decrease.
Order never matters for pure percent multipliers (ab = ba). One flat fee in the chain breaks that.
Up p% then down p% always loses: net multiplier = 1 − (p/100)². 30% up then 30% down = 0.91.
REVERSE: before = after ÷ multiplier. After a +8%, divide by 1.08 — never multiply by 0.92.
Percentage points ≠ percent. 24% to 30% is 6 percentage points AND a 25% increase.
Never average two percents unless the groups are the same size. Convert to counts, combine, convert back.
A percent of a percent multiplies: 18% of the 45% who qualified = 0.18 × 0.45 = 8.1% of the whole.
Increases are unbounded; a decrease can never exceed 100%. "Decreased by 125%" is eliminable on sight.
Before selecting: is the answer the change, the new total, the original, or the percent? All four are in your work.

Why it works — Why percent changes multiply instead of adding

Write the two changes as multipliers (1 + a) and (1 + b), where a and b are the decimal versions of the percents, negative for a decrease. Multiplying them out gives (1 + a)(1 + b) = 1 + a + b + ab. Adding the percents produces 1 + a + b. The entire discrepancy is the cross-term ab, and it has a physical meaning: it is the second change applied to the amount the first change added. A 25% markup on $80 adds $20; the following 30% discount then comes off $100 rather than $80, so it removes an extra 30% of that $20 — the ab term, worth $6. That is why 80 × 1.25 × 0.70 = $70 while "25% up then 30% down, net 5% down" would predict $76. The same term tells you when adding is nearly harmless and when it is catastrophic: for two small changes of 2% and 3%, ab = 0.0006, so the true net of 5.06% is within a rounding error of the naive 5%; for 40% and 50%, ab = 0.20, so the true net of +110% is nowhere near the naive +90%. Two consequences fall straight out of the algebra. First, ab = ba, so the order of pure percent changes never affects the result. Second, an increase of p% followed by a decrease of p% gives (1 + a)(1 − a) = 1 − a², which is less than 1 for every nonzero a — up-then-down by the same percent is always a net loss, of exactly a² as a fraction: 30% up then 30% down is 1 − 0.09 = 0.91, a 9% loss, every time.

Traps — 6

The wrong base

Dividing the change by the ending value instead of the starting one. It has a predictable signature: using the new value understates every increase and overstates every decrease. A rise from 40 to 50 is 10/40 = 25%, not 10/50 = 20%; a fall from 50 to 40 is 10/50 = 20%, not 25%. If two answer choices are the same change measured against the two different bases, the item is testing this and nothing else.

Percents added, not multiplied

Treating a chain of changes as a sum: +25% then −30% read as −5%. The percents are taken of different numbers, so they cannot be combined by addition. Convert each to a multiplier and multiply — 1.25 × 0.70 = 0.875, a 12.5% decrease. The error is largest when the percents are large, and it survives on small ones precisely because it is nearly right there.

The "of" / "more than" swap

"30% of x" is 0.30x. "30% more than x" is 1.30x. "30% off x" is 0.70x. "130% of x" is 1.30x, but "130% more than x" is 2.30x. These are four different multipliers hiding behind nearly identical sentences, and the test writes them deliberately. The base is whatever follows "of" or "than"; the phrasing decides whether you keep the percent, add it to 100%, or subtract it from 100%.

Reverse by subtraction

Undoing a p% increase by taking p% off the result, or undoing a p% decrease by adding p% back. It is never right and it is always close: after a 25% increase, taking 25% back off lands at 1.25 × 0.75 = 0.9375 of the original, 6.25% low. Reversals are divisions. Given the after value and the multiplier, before = after ÷ multiplier.

Percentage points read as percent

When the quantity itself is a percentage, the difference between two values is in percentage points, not percent. A rate moving from 8% to 10% has risen 2 percentage points and 25%. Saying it "increased by 2%" claims something different and false — that would be 8 × 1.02 = 8.16%. Items about rates, shares, and proportions put both numbers in the choices.

Averaged percents over unequal groups

Taking the plain average of two group percentages as though the groups were the same size. If 70% of 40 students and 20% of 160 students did something, the overall figure is not 45% — it is (28 + 32)/200 = 30%, pulled toward the larger group. The reliable move is to abandon percents entirely for a moment, convert everything to counts, combine the counts, and convert back at the end.

The card

ONE-VARIABLE DATA — reference card
Mean = total ÷ count. Median = middle value AFTER SORTING. Mode = most frequent. Range = max − min.
n sorted values: median is the ((n+1)/2)-th if n is odd; average the (n/2)-th and (n/2 + 1)-th if n is even.
Frequency table: mean = sum of (value × frequency) ÷ total frequency. Median: walk the cumulative count to the middle position — never sort the rows.
Outliers move the mean, the range and the standard deviation. They leave the median and the mode alone.
A LOW outlier drags the mean DOWN; removing it pushes the mean UP. Check which end before choosing a direction.
Skew is named for the TAIL. Right tail -> mean > median. Left tail -> mean < median. The mean chases the tail.
Skewed or outlier-heavy data: the MEDIAN is the better summary of a typical value. That is why house prices are quoted as medians.
Standard deviation = typical distance from the mean. Equal range does NOT mean equal SD. SD = 0 only if every value is identical.
Add c to every value: mean and median +c; range and SD unchanged. Multiply every value by k: mean, median, range and SD all × k.
Add one value equal to the mean: mean unchanged, SD DECREASES.
Combining two groups' means: recover the totals first. (20 × 80 + 30 × 90)/50 = 86, not 85.
Box plot: min · Q1 · median · Q3 · max. IQR = Q3 − Q1. Each of the four sections holds ~25% of the data — width is spread, not count.
A box plot shows no mean, no mode and no sample size. A histogram gives the median's interval, not its value. Only a dot plot shows every value.
Before selecting: name the statistic asked for, and say whether the answer is a count or a percent.

Why it works — Why the mean moves, the median cannot, and a shift leaves the spread alone

The mean is the balance point of the data — the single number for which the distances above it and the distances below it cancel exactly, so that the deviations sum to zero. That definition makes it a function of how far every value is, which is why dragging one value out to 900 forces the balance point to slide toward it: the enormous positive deviation has to be offset, and the only way to offset it is to move the pivot. The median is a rank statistic. It knows only the ordering, not the distances. Push the largest value from 40 to 4,000,000 and it is still, simply, the largest value, so whichever value occupied the middle position occupies it still — the median is not resistant to outliers by luck or by convention, it is resistant because outliers change a quantity it does not read. Standard deviation is a root-mean-square distance from the mean: each deviation is squared, the squares are averaged, and the square root is taken at the end. The squaring is why one far-flung value affects it more sharply than it affects the mean — a point three times as far away contributes nine times as much to the total. And the same algebra settles the shift rule in one line: add a constant c to every value and the mean increases by exactly c as well, so every (value − mean) is unchanged, and standard deviation is built from nothing but those differences. Multiply every value by k and the mean is multiplied by k too, so every deviation is multiplied by k, and the spread scales with it. Shift, and the distances survive. Scale, and they don't.

Traps — 6

Median taken from the list as printed

Finding the middle of the list in the order it was given instead of sorting first. For 12, 3, 40, 7, 5 the printed middle is 40 and the median is 7. The same error in table form is taking the middle row of a frequency table rather than the middle data value — the rows are already sorted by value, but there are five of them and there may be two hundred data points.

Frequency read as a value

In a frequency table, computing a statistic from the wrong column: averaging the distinct values while ignoring how many times each occurs, or averaging the frequencies themselves. The values are the data; the frequencies say how many copies of each there are. A table with five rows and 24 students has 24 data points, and the mean divides by 24.

Skew named for the bulk

Calling a distribution left-skewed because most of the bars sit on the left. Skew is named for the tail, and a distribution with its bulk on the left has its tail on the right, which makes it right-skewed. Get this backwards and the mean-versus-median conclusion inverts with it, which is why answer choices in this family pair a skew label with a mean/median claim.

Range read as spread

Concluding that two data sets have the same standard deviation because they have the same range, or that the larger range must have the larger standard deviation. The range is set by two values; the standard deviation is built from all of them. Two sets can share a minimum, a maximum, a mean and a sample size and still differ in standard deviation by 50%.

Box width read as a count

Treating a wide section of a box plot as containing more data. Every one of the four sections holds about a quarter of the values by construction — a wide section means those values are spread out, a narrow one means they are packed together. Width is spread, never count.

Shift mistaken for a change in spread

Assuming that adding a constant to every value, or subtracting one, changes the standard deviation or the range. It changes neither, because it moves every value and the mean by the same amount and leaves every internal distance intact. Only multiplying or dividing every value changes the spread.

The card

TWO-VARIABLE DATA: MODELS AND SCATTERPLOTS — reference card
One dot = one case measured twice. Read the axis labels before the dots.
Axis riders that cost points: "years since ____" and "in thousands / per 100,000".
The line = PREDICTED values. The dots = ACTUAL values. The exam signals which it wants.
Slope = y-units per ONE x-unit, with its sign. Take it from two points ON THE LINE, never two dots.
Intercept = predicted y at x = 0. Meaningful only if x = 0 is inside the data's range.
Residual = actual - predicted. Positive -> dot ABOVE the line -> model UNDERestimates.
Check every residual with: predicted + residual = actual.
"By how much does the model overestimate?" is a residual question without the word.
Change over an interval = slope x interval length. 2001 to 2011 is 10 steps, not 11.
Inside the data's x-range = interpolation, supported. Outside = extrapolation, not supported.
Table: constant DIFFERENCES -> linear. Constant RATIOS -> exponential.
Words: constant AMOUNT per step -> linear. Constant PERCENT per step -> exponential.
Exponential y = a*b^x: a = value at x = 0; b = multiplier per one x-unit; percent = (b - 1).
15% decline -> factor 0.85, not 0.15. 20% growth -> 1.2. Doubling every 5 hours -> 2^(h/5).
Residual plot: random scatter = linear is fine. Arch or U = curved relationship, model wrong.
Residuals summing to zero is an identity of the fitting method, never evidence of fit.
Before answering: reread the last sentence and confirm which of the five jobs it named.

Why it works — Why residuals are the currency, and why the line has no authority outside the data

The line of best fit is not drawn by eye and it is not the average of anything obvious. It is chosen to make the residuals collectively as small as possible — specifically, to minimize the sum of their squares. Squaring does two jobs at once: it stops a point 5 above the line from cancelling a point 5 below it, and it makes one large miss cost more than several small ones, which is why a single outlier can pull the whole line toward itself. Three consequences follow, and all three are tested. First, the residual is not an incidental quantity invented for exam questions; it is the exact thing the line was built to control, which is why "how far is this point from the model" is the natural question to ask about a fitted line. Second, the residuals of a least-squares line with an intercept sum to zero by construction — so being told that they sum to zero tells you nothing whatsoever about whether the model fits. What carries information is their PATTERN: scattered randomly around zero means the straight line captured the shape, while a systematic run of positives, then negatives, then positives means the relationship is curved and the line is wrong in a way more data will not fix. Third, and most costly: the line was fitted using only the dots that exist. Inside the window of x-values where data were collected, the line is constrained on both sides by real measurements, which is what makes a prediction there defensible — that is interpolation. Outside that window nothing constrained it. Extending the line to x = 40 when the data stop at x = 10 does not extend the evidence; it assumes that a shape observed over one interval continues over another where it was never observed, which is why a linear model of a child's height fitted from ages 2 to 10 confidently predicts a 3.3-metre adult. The arithmetic will not object. Nothing in the algebra knows where the data stopped, which is why noticing the range is your job and not the model's. The same reasoning explains why the choice between linear and exponential is not cosmetic: adding a constant amount and multiplying by a constant factor agree closely over a step or two and diverge without limit afterwards, so the wrong shape can match the first row of a table perfectly and still be worthless three rows later.

Traps — 6

Residual sign flip

Computing predicted − actual instead of actual − predicted, or computing it correctly and then reading the sign backwards. The definition runs one way only, and the three phrases that travel together are: positive residual, dot above the line, model underestimates. A distractor carrying the correct magnitude with the opposite sign appears on essentially every residual item, because it is the error the item was written to catch.

Data point used as a model point

Computing the slope of the line of best fit from two dots on the scatterplot rather than two points on the line. The line is a summary that generally passes through none of the data, so the segment joining two dots has its own slope and it is not the model's. Use a stated equation, or points explicitly described as lying on the line, or two clean grid crossings the drawn line passes through.

Blind extrapolation

Feeding the model an x far outside the range where data were collected and treating whatever comes back as a prediction the data support. The algebra never complains, which is the problem. This includes the quiet version: interpreting the y-intercept as a real quantity when x = 0 lies well outside the observed range, so the model has never been checked anywhere near it.

Rate, level, and interval confused

Reporting the slope when a value was asked for, a value when a change was asked for, or the one-unit change when the question named a multi-unit interval. Change over an interval is the slope times the interval's length, and the interval from 2001 to 2011 is ten steps, not eleven. On a well-built item all three numbers are printed as choices, and all three are correct answers to questions that were not asked.

Percent-versus-amount misfit

Forcing a straight line onto data that changes by a constant factor, or the reverse. The tell in a table is differences versus ratios; the tell in a sentence is an amount ("falls by 40 units a year", linear) versus a percentage ("falls by 8% a year", exponential). Its close relative: using the percentage itself as the base — writing 0.15 where a 15% annual decline requires a factor of 0.85, or 1.15 for a 15% rise.

Axis-label unit slip

Ignoring a rider on an axis or a variable definition and answering in the model's internal units instead of the question's. "In thousands" turns a slope of 2.4 into 2,400 per year; "years since 1995" means the answer 8 may need to be reported as 2003; "per 100,000 residents" means the model's output is a rate, not a count. Convert once, at the end, after underlining the units the question asked for.

The card

PROBABILITY AND CONDITIONAL PROBABILITY — reference card
Every answer is (how many count) / (how many you are drawing from). Find the second one FIRST.
Denominator words: given that / among / of the / from those who / if the selected X is Y.
No condition stated -> the denominator is the grand total.
Condition names a row -> that row's total. Names a column -> that column's total.
The numerator is counted INSIDE the denominator's pool. Nowhere else.
One cell, three answers: cell/row, cell/column, cell/grand total. All three will be offered.
P(A given B) and P(B given A) share a numerator; equal only if the two totals are equal.
Complement: P(not A) = 1 - P(A), taken inside the conditioned pool, never over the whole table.
"A or B" = A's total + B's total - the shared cell, over the grand total. Subtract the overlap once.
Pooled categories ("did not support", "not 18-64") span several cells — add every one, top and bottom.
A percent of a subgroup applies to that subgroup only. Convert it to a count and write it in the table.
Missing cells: every row and column sums to its total, and both directions reach the same grand total.
Relative-frequency table: check what it sums to (each row = 1, or the whole table = 1) before reading.
Compare groups of unequal size by rate, never by cell count.
Independent = the row's split matches the table's split. Association is never causation without random assignment.
Asked for a count, not a probability? Probability x the size of the population it was measured on.
Before answering: numerator <= denominator, value between 0 and 1, and reread the last line.

Why it works — Why the numerator survives the condition and the denominator does not

A probability is not a property of an event. It is a ratio between an event and a population, and "given that the employee takes transit" does not add information to the old ratio — it replaces the population. Before the condition there were 200 people to draw from; after it there are 50, and the other 150 have not become less likely, they have stopped existing for the purposes of this question. Meanwhile the people who are both transit riders and under 40 are the same 30 people either way. That single asymmetry explains everything else on this page. It is why P(A given B) and P(B given A) share a numerator and differ only in denominator: the overlap is one group of people, and the two questions divide it by two different populations, so they agree only when those two populations are the same size. It is why a conditional can sit far away from the unconditional in either direction — narrowing to transit riders raised the probability of being under 40 from 75/200 = 0.375 to 30/50 = 0.6, because transit riders skew young in this table. And it is why the complement has to be taken inside the conditioned pool: 1 − P(A) computed over 200 people is not the complement of a fraction whose denominator is 50, and subtracting it from 1 produces a number that is not a probability of anything. The case where the conditional does not move is worth naming, because the test asks about it in words rather than symbols: when the conditional equals the unconditional, the condition carried no information and the two traits are independent. In a table that shows up as every row having the same internal split as the grand total. When the splits differ, the traits are associated, and the size of the gap between one row's split and another's is the whole of the evidence for it.

Traps — 6

Grand-total denominator

The condition was read and then ignored: the right cell is found and divided by the table's grand total instead of the conditioned group's total. It is the highest-frequency wrong answer on this skill, the test writes a distractor for it on essentially every conditional item, and it is always smaller than the correct answer — which makes "that came out surprisingly low" a usable alarm.

Reversed conditional

Computing P(A given B) when the question asked for P(B given A). Same cell on top, the other margin underneath. The tell is grammatical, not mathematical: whatever noun follows "given that," "among," or "of the" is the denominator, and whatever noun the question then asks about supplies the numerator. Read those two nouns in order, out loud, before touching the numbers.

Joint mistaken for conditional

"The probability that a randomly selected student is a senior and opposes the change" is one cell over 300. "The probability that a senior opposes the change" is the same cell over the senior total. Nothing but the wording separates them, both wordings appear on the same test, and the word doing the work is a single "and" versus a single "a."

Complement taken in the wrong pool

Using 1 − P(A) computed over the grand total when the question needed 1 − P(A given B) computed inside the row. A complement is always taken inside whatever population the question has already narrowed to; taken outside it, the two fractions do not even share a denominator, so the subtraction is meaningless before it is wrong.

Pooled-category miss

"Did not support," "not in the 18-to-64 group," and "either biology or chemistry" each span more than one row or column, and the fraction needs every cell in the span — on the top and on the bottom, wherever the span applies. Taking a single cell where a category was pooled is the standard way a hard item is lost by a student who understood it perfectly.

Overlap double-count on "or"

Adding a row total to a column total to answer "A or B." The cell where they cross belongs to both groups and gets counted twice, so the answer comes out too large and can exceed 1. Subtract the shared cell exactly once: (row total) + (column total) − (shared cell), over the grand total.

The card

INFERENCE FROM SAMPLE STATISTICS & MARGIN OF ERROR — reference card
SAT only. Not assessed on the PSAT/NMSQT or PSAT 10.
Plausible interval = estimate - margin, up to estimate + margin. Write BOTH endpoints.
The margin is the HALF-width. Given an interval: estimate = midpoint, margin = (width)/2.
Given as a percentage? The margin is in PERCENTAGE POINTS. 38% +/- 3 pts = 35% to 41%.
Credited verb is always "it is plausible that" — never "is", never "certainly", never "proves".
The interval estimates a MEAN or a PROPORTION for a population. Never an individual.
Margin shrinks with 1/sqrt(n): quadruple the sample to halve the margin; x9 to cut it to a third.
Wider interval = smaller sample, OR more variable population, OR higher confidence level.
Higher confidence always means a wider interval from the same data. Never more precise.
Margin of error measures sampling variability ONLY — blind to bias, non-response, bad frames.
More data collected the same biased way = a more precise wrong answer.
Random SAMPLING licenses generalising to the sampled population. Nothing causal.
Random ASSIGNMENT licenses a causal claim. Different word, different conclusion.
Overlapping intervals: difference NOT established. That is not the same as "equal".
Over-correction is also wrong: a random sample DOES support a conclusion about its population.
Before selecting: name the population, name the quantity, check the verb.

Why it works — Why more data narrows the interval, and why it cannot rescue a biased one

The margin of error has the form (a number set by the confidence level) × (the spread of the data) ÷ √n. You are never asked to use that expression on the SAT, but its shape explains every comparison item on the topic. The √n in the denominator is there because the sample mean is an average of n independent draws, and averaging is a cancellation process: values above the true mean and values below it increasingly offset one another as n grows, so the average settles down even though the individual values are as spread out as they ever were. The cancellation accumulates with the square root of the count rather than the count itself, which is why quadrupling a sample halves its margin instead of quartering it, and why the fourth thousand respondents buy far less precision than the first hundred did. Now notice what that derivation assumed: that the draws are from the population you want to describe. Bias violates the assumption before the arithmetic starts. If your frame is a magazine's subscribers and your target is a country, every additional draw is another sample of subscribers, so the estimate converges — tightly, obediently, with a beautifully small margin of error — on the subscribers' opinion. The formula cannot notice, because n counts how many people you asked and contains no term for whether they were the right people. That is the precise sense in which precision and accuracy are independent quantities: √n governs one of them and the sampling frame governs the other, and only the first appears in the number the study reports.

Traps — 6

Certainty upgrade

Taking the correct interval and asserting it: "the mean IS between 7.07 and 7.63" instead of "it is plausible that the mean is between 7.07 and 7.63." The numbers match the credited answer exactly, which is what makes this the most-missed distractor on the topic. An interval estimate is a claim about plausibility; if the sample could establish the parameter, no interval would be needed.

Interval applied to individuals

Reading a range built for a MEAN as a range that every member falls inside — "every bottle contains between 7.07 and 7.63 ounces," "95% of bottles are in that range." Individual values are always far more spread out than the estimate of their average. Ask what the interval is an interval FOR, and this collapses immediately.

Precision mistaken for accuracy

Treating a small margin of error as evidence that a study is trustworthy. The margin measures only how much the draw wobbles; it is blind to a wrong sampling frame, non-response, and self-selection. Collecting more data the same way narrows the interval around the same off-target number — a bigger biased sample buys a more precise wrong answer.

Half-width slip

Confusing the margin with the width of the interval. The margin is half the width: an interval of 61.4 to 68.2 has a width of 6.8 and a margin of 3.4. The error runs in both directions — doubling the margin when building an interval, or reporting the full width when asked for the margin — and both produce a number that is exactly right for the wrong quantity.

Overlap read as a verdict

Two errors that mirror each other. Comparing point estimates while ignoring the intervals entirely ("15.0 is bigger than 14.2, so it went up"), or concluding from overlapping intervals that the two population values are equal. Overlap means the difference has not been established; it never means the values are the same.

Scope drift

Extending a conclusion past the population the sample was drawn from — from one plant's employees to a whole company, from riders to residents. Its under-corrected twin costs just as much: refusing to generalise at all, and insisting the result applies only to the people surveyed. A random sample exists precisely to license the step out to the frame it was drawn from, and no further.

The card

EVALUATING STATISTICAL CLAIMS — reference card
Nothing is computed here. Two switches decide the answer; the finding is irrelevant.
SWITCH 1 — random SELECTION, before the study -> decides WHO the conclusion covers.
SWITCH 2 — random ASSIGNMENT, after you have the subjects -> decides whether you may say CAUSED.
Selection ON -> generalise to the list sampled FROM, and no further. OFF -> the people studied only.
Assignment ON (with a comparison group) -> "caused". OFF -> "is associated with".
Both ON: cause, generalisable. Selection only: association, generalisable.
Assignment only: cause, these subjects only. Neither: association, these subjects only.
The two are independent. Neither ever substitutes for the other.
Observational = the researcher only recorded. Experiment = the researcher imposed a treatment.
Two pre-existing groups compared is still observational. Comparison is not assignment.
Write the licensed sentence BEFORE reading the choices: [verb] ... among [ceiling population].
Eliminate on TWO axes: strength of claim, then breadth of population. Then cut anything under-claimed.
A recommendation ("they should ...") is a causal claim. Cut it from any observational study.
Bias is systematic; size only shrinks random error. A bigger biased sample is precisely wrong.
Read the noun after "randomly selected from" — registered voters are not adults.
Invited 2,000, heard from 240 -> a self-selected 240, not a random sample. Same for dropouts.
One group, before and after, no comparison -> no cause, however random the selection was.
Credited answers hedge: likely, evidence, tends to, association, about. Cut proves, guarantees, all, any, will.
A mean difference is not a promise about any individual.
The conclusion covers what was MEASURED, at the dose given, over the period run, against the treatment compared.
"No significant difference" is a failure to detect an effect, not proof there is none.

Why it works — Why each randomisation does exactly one job, and cannot do the other

Random assignment works by BALANCE. When a coin flip decides which condition each subject gets, every other characteristic those subjects carry — age, motivation, prior sleep quality, income, diet, genetics, things nobody thought of and nothing measured — lands in the two groups by chance rather than by any rule connected to the outcome, so the groups end up alike in everything except what the researcher controlled. When the outcome then differs by more than chance would produce, the treatment is the only systematic difference left standing. That is the whole argument, and it is why randomisation beats the alternative of listing confounders and adjusting for them: it neutralises the confounders you never thought of, which are exactly the ones that ruin observational studies. It is also why no observational study can be repaired by a bigger sample or a cleverer analysis — with nothing assigned, the subjects sorted themselves, and that sorting is itself a systematic difference carrying along everything that produced it. Random selection works by REPRESENTATIVENESS, which is a different property with a different reach. Drawing at random from a list gives every member of that list an equal chance of appearing, so the sample's composition tracks the list's composition on average — the same mix of ages, incomes and opinions, up to sampling variability — which is what makes a statistic computed from the sample an estimate of the parameter for the list. Note the word "list": the mechanism is a fact about the frame the draw was made from, and it says nothing whatsoever about anyone who was not on it. That is also the precise reason a large sample cannot rescue a biased one. Size governs random error, the scatter around the truth, and it shrinks as the sample grows; bias is systematic error, a fixed offset built into the procedure, and it does not shrink at all. Averaging more readings from a mis-calibrated instrument buys a more confident wrong number, and a voluntary-response survey of 9,400 is not a better version of one with 94 — it is the same error, measured more precisely. So the two mechanisms answer two different questions. Balance answers "is the treatment responsible?" Representativeness answers "responsible for whom?" Neither contains any part of the other, which is exactly why the exam can build — and does build — items where one is present and the other is conspicuously missing.

Traps — 6

Causation from observation

Reading a pattern in recorded data as a cause. The tell is a verb: causes, lowers, improves, leads to, results in. It also arrives disguised as advice — "students who want higher grades should work fewer hours" is a causal claim wearing an imperative, and it is wrong for exactly the same reason. The defence is mechanical: if nobody was assigned anything, name one confounder out loud and the choice dies.

Comparison mistaken for assignment

Two groups appear in the description, so the study reads like an experiment. "Researchers compared students who take music lessons with students who do not." Nobody assigned anybody — the students sorted themselves, and everything that made them sort that way came along with them. Assignment means the RESEARCHER decided who got what. Comparing groups that already existed is observation with two columns.

Frame overreach

Generalising past the list the sample was actually drawn from. Registered voters are not adults, gym members are not the public, one university's students are not university students, and people reachable by landline at 2pm are a narrow slice of anyone. Read the noun that follows "randomly selected from" and treat it as a hard ceiling; every choice naming a wider group is cut without further thought.

Reflexive correlation-is-not-causation

The trap built specifically for strong students. Having learned that observational data cannot support a cause, the reader applies the rule to a genuine randomised experiment and refuses a conclusion the design has already earned — usually reasoning that the subjects were volunteers, which is switch 1 doing switch 2's job. When "randomly assigned" appears and there are two conditions, the causal verb is available. The only question left is who it applies to.

Big-n laundering

Treating sample size as a remedy for a broken selection procedure. Size shrinks random error and leaves systematic error exactly where it was, so a voluntary-response survey of 9,400 is not a better version of one with 94 — it is the same bias, estimated more precisely. Any answer choice whose reasoning turns on how large or how small the sample was is almost always naming the wrong defect.

One-axis elimination

The characteristic execution error at the top of the score range. Each of the four choices varies on TWO axes — strength of claim and breadth of population — and a reader who checks only one will find two survivors and pick the wrong one. The choice that gets the population right and the verb wrong is written to catch exactly this. Check both axes on every choice, every time; it costs about eight seconds.

The card

AREA AND VOLUME — reference card
ON THE SHEET: circle A = πr² and C = 2πr; rectangle A = lw; triangle A = ½bh.
ON THE SHEET: c² = a² + b²; 30-60-90 is x, x√3, 2x; 45-45-90 is s, s, s√2.
ON THE SHEET: box lwh; cylinder πr²h; sphere (4/3)πr³; cone (1/3)πr²h; pyramid (1/3)lwh.
NOT ON THE SHEET: any surface area at all. Build it by adding faces.
NOT ON THE SHEET: trapezoid ½(b1 + b2)h; parallelogram bh; rhombus ½d1d2.
NOT ON THE SHEET: equilateral triangle (s²√3)/4; regular hexagon (3√3/2)s².
NOT ON THE SHEET: V = Bh for a general prism; sector and arc; circle equation; space diagonal.
Worth memorising: closed box SA = 2(lw + lh + wh); closed cylinder SA = 2πr² + 2πrh; sphere SA = 4πr².
Prism or cylinder: V = (base area) × height. Tapers to a point: multiply that by 1/3.
Hemisphere = half a sphere. No hemisphere formula is printed — halve it yourself.
Every printed circular formula wants r. Write r = d/2 on its own line before anything else.
The h in a cone, pyramid or triangle is PERPENDICULAR height, never the slant.
SCALING: multiply every length by k -> lengths ×k, areas and surface areas ×k², volumes ×k³.
Backwards: area ratio -> square root; volume ratio -> cube root. Volumes 8:27 means lengths 2:3.
Only for SIMILAR figures. One dimension tripled on a cylinder is ×9 (radius) or ×3 (height), not ×27.
Percents compound: +20% per edge is ×1.728 (+72.8%); -10% per edge is ×0.729 (-27.1%).
Units are scaling: 1 ft³ = 1,728 in³; 1 yd³ = 27 ft³; 1 m³ = 1,000,000 cm³.
Hollow solids: square first, subtract second. π(R² - r²)L, never π(R - r)²L.
Uniform cross-section: V = (cross-sectional area) × length. A net's area IS the surface area.
Rate items: volume ADDED ÷ rate = time. Ask whether the question is about a state or a change.
Before answering: is it a length, an area or a volume? The unit's exponent is the check.

Why it works — Why the exponents, and why the one-third

An area is a product of two lengths and a volume is a product of three. That is the entire reason for the exponents, and it is worth holding as an operation rather than a rule: multiply every length by k, and a product of two lengths picks up the factor k twice, while a product of three picks it up three times. Nothing about the shape enters the argument — it works identically for a circle, a trapezoid, a grain silo, or a blob with no formula at all, which is precisely why a scaling item can be answered without knowing what the figure is or what its dimensions are. The units say the same thing out loud: cm² has a 2 in it and cm³ has a 3, and if you can read the exponent off the unit you can read the exponent off the scale factor. This is also the whole reason one cubic foot is 1,728 cubic inches rather than 12 — a unit conversion is a scaling with k = 12, applied to three dimensions at once. The one-third on cones and pyramids looks arbitrary and is not. A cube can be dissected into exactly three congruent square pyramids: pick one vertex of the cube as the shared apex, and each of the three faces that does not touch that vertex becomes the base of one pyramid. The three fill the cube exactly, with nothing left over, so each holds one third of it — and each has base s² and height s, giving V = ⅓ × base × height. The same relationship survives changing the base shape and the height, which is why the reference sheet carries a third on the cone and a third on the pyramid and no third anywhere else. If a solid tapers to a point, it holds a third of the prism it fits inside. If it has two identical parallel ends, it holds all of it. And the reason the box and the cylinder are the same formula in two costumes: a prism is a stack of identical copies of its base, and a cylinder is a stack of identical circles. The volume is the area of one layer times how tall the stack is, V = Bh — with ℓwh and πr²h being that sentence with a rectangle and a circle substituted in. The sheet prints the two special cases and leaves the general one to you, which is exactly the gap that a prism with a trapezoidal cross-section is designed to find.

Traps — 6

Diameter dropped into a radius slot

Every circular formula on the reference sheet — area, circumference, cylinder, cone, sphere — is written in terms of r, and problems state diameters far more often than radii. Using d where r belongs multiplies an area by 4 and a volume by 8, and both of those wrong numbers are reliably present in the choices, because they are the two most predictable errors on the whole skill. The fix is mechanical: write r = d/2 as a separate line before any formula gets touched. The mirror image is just as costly — solving correctly for r and reporting it when the question asked for the diameter.

Slant used as height

The h in (1/3)πr²h and (1/3)ℓwh is the PERPENDICULAR height from the apex straight down to the base, not the slanted edge running up the outside. The same error appears one dimension down, in ½bh, where the height must be perpendicular to the base and not the length of a leaning side. Diagrams supply both numbers precisely because only one of them is correct, and the slant is always the larger of the two — so an answer built on it is always too big.

Hunting for a formula that was never printed

Spending thirty seconds scanning the sheet for a trapezoid, a hexagon, a surface area, a hemisphere, or a composite solid, all of which are absent. Nothing about surface area is printed anywhere on the sheet, and there is no general prism formula on it either. Knowing the boundary converts that lost time into an immediate decision: build the formula from faces, or cut the solid into printed pieces.

Linear factor applied to an area or a volume

Doubling a length and doubling the area, or converting cubic feet to cubic inches by multiplying by 12. Areas take k² and volumes take k³, always, and the unit itself is announcing the exponent: ft² takes 12², ft³ takes 12³. This is the same error whether it arrives dressed as similarity, as a unit conversion, or as a percentage — and as a percentage it is worst, because +20% on each edge is +72.8% on the volume, not +20% and not +60%.

Cube law applied to a partial scaling

k³ requires SIMILAR figures — every dimension multiplied by the same k. When a problem changes only the radius, or only the height, the rule does not apply and you go back to the formula. Tripling only the radius of a cylinder multiplies its volume by 9; tripling only the height multiplies it by 3; tripling both multiplies it by 27. The exam puts all three in the same answer set and lets the word "similar" — present or absent — decide which is right.

The wrong quantity answered

The geometry is finished and every line of it is correct, and the number reported is r when the question asked for d, or r² when it asked for r, or the surface area when it asked for the volume, or the total volume when it asked for the volume added. This is the characteristic error at the top of the score range, and the unit is the cheapest detector available: an answer in square centimetres cannot be a volume, and an answer in cubic centimetres cannot be a surface area. Before selecting, read the last clause of the question again and name the unit it wants.

The card

LINES, ANGLES & TRIANGLES — reference card
Straight line = 180 deg. Full turn = 360. Vertical angles are equal.
Complementary = sums to 90. Supplementary = sums to 180. Do not swap them.
Parallel + transversal: 8 angles, only TWO values, and the two sum to 180.
   corresponding / alternate interior / alternate exterior -> EQUAL
   same-side interior (co-interior) -> SUPPLEMENTARY
Triangle angles sum to 180. Exterior angle = sum of the two REMOTE interior angles.
Equal sides face equal angles. The longest side faces the largest angle.
Similar: AA is enough. Congruent: SSS, SAS, ASA, AAS, HL — never SSA, never AAA.
Correspondence is the letter order: ABC ~ DEF means AB/DE = BC/EF = AC/DF.
Scale factor k: lengths xk, perimeters xk, areas xk^2, volumes xk^3.
   Backwards: side ratio = sqrt(area ratio). Given an area, take a root first.
Parallel line inside a triangle: small side matches the WHOLE side, never the leftover.
Midsegment (joins two midpoints): parallel to the third side, half its length.
Triangle inequality: |a - b| < third side < a + b. BOTH bounds strict.
Integers strictly between L and U: U - L - 1. Write out the first and last to confirm.
45-45-90 -> 1 : 1 : sqrt2 (leg : leg : hyp).
30-60-90 -> 1 : sqrt3 : 2 (short leg : long leg : hyp); short leg is opposite the 30.
Equilateral, side s: height = s*sqrt3/2, area = s^2*sqrt3/4. Neither is on the sheet.
sqrt2 = 1.414, sqrt3 = 1.732 — use them to check the sides came out in the right order.
Pythagorean triples on sight: 3-4-5, 5-12-13, 8-15-17, 7-24-25, and all multiples.
Reference sheet HAS: Pythagoras, both special triangles, angle sum 180.
   It does NOT have: exterior angle, triangle inequality, midsegment, congruence criteria, the k^2 rule.
Drawn to scale lets you ESTIMATE. Only the givens let you CONCLUDE.
Last move, every time: re-read the final sentence and answer THAT quantity.

Why it works — Why two angles are enough, and why that makes the special triangles constants

Everything in this lesson beyond the angle bookkeeping falls out of one fact, so it is worth having the fact rather than the list. A triangle's three angles sum to 180°, so fixing two of them forces the third. Fix all three and you have fixed the triangle's SHAPE completely and left exactly one thing free: its size. Two triangles with the same angles therefore differ by a single number — a scale factor k applied to every length at once — and that is what similarity is. Three consequences follow immediately, and all three are tested. First, AA is a sufficient criterion for similarity, because the third angle is not independent information. Second, AAA can never prove congruence, because it fixes shape and leaves size free; this is not a technicality but the reason SSS, SAS, ASA, AAS and HL all contain at least one SIDE — a criterion with no side in it cannot pin down size. Third, and this is where the points are, every length in the figure scales by k, so corresponding sides are in ratio k and so are perimeters, altitudes and midsegments; but an area is built from two lengths multiplied together, so it scales by k × k = k², and a volume from three, so k³. The exponent is the number of dimensions, not a rule to memorise. It also runs backwards, which is the part strong students miss: given an area ratio, the side ratio is its square root. The same fact explains why the two special right triangles can be printed on a reference sheet at all. Every 30-60-90 triangle in existence has the same three angles, so by AA every one of them is similar to every other, so the ratios among its sides are a property of the ANGLES alone and not of any particular triangle. That is why 1 : √3 : 2 is a fact about the angle set rather than about one drawing, and you can confirm it in one line: cut an equilateral triangle of side 2 down its altitude and you get a right triangle with hypotenuse 2 and short leg 1, so the third side is √(4 − 1) = √3. Cut a unit square along its diagonal and you get 1 : 1 : √2. And this is the seed of the next skill point: sine, cosine and tangent are exactly "the side ratio determined by the angle," a definition that is only coherent because AA similarity guarantees the ratio does not depend on which triangle you drew.

Traps — 6

Figure trusted over the given

Concluding from the drawing that two lines are parallel, an angle is right, two segments are equal, a point is a midpoint, or a segment bisects an angle, when nothing said so. Digital SAT figures are drawn to scale unless the item states otherwise, and that is worth using — to estimate a size and eliminate an answer that is visibly wrong. It is never worth using to assert a fact, because an item that wants you to make the assumption is an item built to punish it. The most expensive version is the apparent angle bisector: a segment drawn from a vertex to the opposite side looks like it splits the angle in half in almost every figure, and it does so only when the item says it does.

Equal-or-supplementary, guessed

In a parallel-lines figure the eight angles take exactly two values, so the only decision is which of the two a given angle is — and reaching for the wrong one produces a clean, plausible, wrong number every time. Corresponding, alternate interior and alternate exterior pairs are EQUAL. Same-side interior (co-interior) pairs are SUPPLEMENTARY. If you cannot recall which family a pair belongs to, use the fact that the figure is drawn to scale: two angles that clearly look the same size are equal, and one obviously acute paired with one obviously obtuse are supplementary. That check takes two seconds and settles it without the vocabulary.

Part used as whole

When a segment parallel to one side cuts a triangle, the small triangle's side corresponds to the WHOLE side of the large triangle, not to the leftover piece. AD/AB, never AD/DB. The same error reappears wherever a figure nests one shape inside another — a transversal cutting parallel lines into proportional segments, or the altitude to the hypotenuse of a right triangle creating two smaller triangles similar to the original and to each other. Write out the correspondence as pairs before writing a ratio; the pairing is the answer, and the algebra is bookkeeping.

Linear factor applied to area

Multiplying an area by the side scale factor k instead of by k². Lengths, perimeters, altitudes and midsegments all take k; areas take k²; volumes take k³. It runs backwards too, and the backwards direction is where 1500-level students lose it: an area ratio of 9 : 16 means a side ratio of 3 : 4, not 9 : 16. On any similarity item involving area, both the k answer and the k² answer will be among the choices.

Wrong role in a special right triangle

The ratio 1 : √3 : 2 is useless until you decide whether the given length is the short leg (opposite 30°), the long leg (opposite 60°) or the hypotenuse. Get that assignment wrong and you multiply where you should divide, which produces an answer roughly √3 or 3 times too large or too small — and each of those is a printed choice. The companion error is applying the 45-45-90 ratio (√2) inside a 30-60-90, or the reverse. Name the given side's opposite angle before you touch the ratio.

Solved for x, answered x

The item defines an angle as (4x + 12)°, you correctly find x = 17, and 17 is sitting in the choices — as is the other angle in the figure, and as is its supplement. This is the highest-frequency loss on the whole skill point and it has nothing to do with geometry. Circle the quantity the final sentence names before you start solving, and read that sentence again before you select. It is the cheapest point on the Math section and it is bought with about four seconds.

The card

RIGHT TRIANGLES AND TRIGONOMETRY — reference card
Hypotenuse = the side opposite the right angle. Always the longest side. Never a leg.
"Opposite" and "adjacent" belong to an ANGLE, not to the triangle. Name the vertex, then label.
Pythagoras: leg² + leg² = hypotenuse². Missing hypotenuse -> add, then root. Missing leg -> SUBTRACT, then root.
Triples on sight: 3-4-5, 5-12-13, 8-15-17, 7-24-25, and every multiple (6-8-10, 9-12-15, 10-24-26).
45-45-90: x, x, x√2. Both legs equal; the √2 sits on the hypotenuse. Leg = hypotenuse ÷ √2.
30-60-90: x opposite 30°, x√3 opposite 60°, 2x hypotenuse. Hypotenuse = twice the SHORT leg.
SOH-CAH-TOA: sin = opp/hyp, cos = adj/hyp, tan = opp/adj. NOT on the reference sheet — memorise it.
Pick the ratio by roles: hyp + opp -> sine. hyp + adj -> cosine. two legs -> tangent. tan θ = sin θ / cos θ.
Unknown in the numerator -> multiply. Unknown in the denominator -> DIVIDE. Write the equation first.
Cofunction: sin(x°) = cos(90° − x°). If sin A = cos B for acute A and B, then A + B = 90°. Never 180°.
Pythagorean identity: sin²θ + cos²θ = 1. It is a² + b² = c² divided by c². NOT sin θ + cos θ = 1.
One ratio in, another out: cos θ = 7/25 -> adjacent 7, hypotenuse 25, opposite 24 -> tan θ = 24/7.
Ratios are scale-invariant. Similar triangles share every sine, cosine and tangent. Never scale a ratio.
Exact values, read off the two triangles: sin30 = 1/2, cos30 = √3/2, tan30 = √3/3; sin45 = cos45 = √2/2, tan45 = 1; sin60 = √3/2, cos60 = 1/2, tan60 = √3.
The SAT prints rationalised forms: 1/√3 = √3/3 and 1/√2 = √2/2. Check before deciding your answer is absent.
For an acute angle, 0 < sin < 1 and 0 < cos < 1. Any sine or cosine at or above 1 is eliminable on sight.
Above 45°, sin > cos. Below 45°, cos > sin. Equal only at 45° — which is what sin A = cos A tells you.
Angle of depression from the top = angle of elevation from the bottom. Put the angle at the far vertex.
Altitude to the hypotenuse makes three similar triangles: CD² = AD·DB. Geometric mean, not the average.
Degrees, not radians. Confirm sin(30) = 0.5, not −0.988, before the first Math module.
Before selecting: which vertex, and which quantity — a side, a ratio, an angle, or an area? All of them are in your work.

Why it works — Why a ratio can be a function of an angle at all — and where both identities come from

Fix an acute angle θ and draw any right triangle containing it. Now draw a second, different one that also contains θ. Both have a right angle and both have θ, so by angle-angle they are similar, and similar triangles have proportional corresponding sides. Multiply every side of the first by the same scale factor k and you get the second — which means opposite/hypotenuse comes out as (k · opposite)/(k · hypotenuse), and the k cancels. Every ratio of two sides is therefore the same number in every right triangle containing θ. That, and only that, is what licenses writing "sin θ" as a function of an angle with no triangle attached: the triangle was never doing any work beyond fixing the angle. It is also why enlarging a figure, or being handed a similar triangle with different numbers, changes no sine, cosine or tangent anywhere in it. Now label that triangle a for the leg opposite θ, b for the leg adjacent, c for the hypotenuse, and both identities fall out in one line each. Pythagorean identity: sin θ = a/c and cos θ = b/c, so sin²θ + cos²θ = a²/c² + b²/c² = (a² + b²)/c², and a² + b² = c², so the whole thing is c²/c² = 1. The identity is not a separate fact to memorise — it is the Pythagorean theorem divided through by c², which is exactly why it has squares in it and why dropping them destroys it. Complementary-angle identity: the two acute angles sum to 90°, so the other one is (90 − θ)°. The leg a is opposite θ; stand at the other vertex instead and that same leg a is now the adjacent one, while c is still the hypotenuse. So sin θ = a/c = cos(90° − θ). It is one fraction with two names, chosen by which corner you are standing in — no calculation, no approximation, exact for every θ. Both identities are consequences of the same picture, which is worth noticing: if you can draw the triangle and label it, you can re-derive either one in the margin faster than you can recall which is which.

Traps — 6

The wrong vertex

Computing a correct ratio of the correct triangle from the other acute angle. "Opposite" and "adjacent" are defined relative to a chosen angle, and moving to the other acute angle swaps them, so a tangent becomes its reciprocal and a sine becomes a cosine. This is the single most common wrong answer on the skill, and it is invisible in the arithmetic — every step after the mislabelling is flawless. The fix is procedural: name the vertex aloud before labelling any side.

Hypotenuse used as a leg

Putting the hypotenuse into a tangent, or applying a² + b² = c² with c set to a side that is not opposite the right angle. Two related slips live here: assuming the hypotenuse is whichever side is drawn longest or most vertical, and treating the hypotenuse as the sum of the legs — legs 1 and 2 give a hypotenuse of √5 ≈ 2.24, never 3. The hypotenuse is the side opposite the right angle, it is longer than either leg and shorter than their sum, and it is identified before any other labelling happens.

The identity without its squares

Using sin θ + cos θ = 1 in place of sin²θ + cos²θ = 1. The false version is used to "find" a cosine by subtracting a sine from 1, and it fails everywhere: at 30°, 0.5 + 0.866 = 1.366. The identity is the Pythagorean theorem divided by c², so the squares are load-bearing. If a value was obtained by subtracting from 1 rather than by subtracting from 1 and then taking a square root, it is wrong.

Complement swapped for supplement

Writing sin(x°) = cos(180° − x°) instead of cos(90° − x°), or assuming sin(x°) = cos(x°) outright. The 90° comes from the two acute angles of a right triangle summing to 90°, and nothing about the rule involves 180°. Sine and cosine of the same acute angle are equal only at 45°; above 45° the sine is larger, below it the cosine is.

Special triangle assembled backwards

In a 30-60-90, the hypotenuse is twice the SHORT leg — the one opposite the 30° — and the √3 belongs to the leg opposite the 60°. Doubling the long leg, or putting the √3 on the hypotenuse, produces a triangle that does not satisfy the Pythagorean theorem. The √2 belongs to the 45-45-90 and to nothing else; borrowing it for a 30-60-90 is the same error in the other direction. If you are unsure, sketch half an equilateral triangle in the margin and read the sides off.

The ratio scaled like a length

Multiplying a sine, cosine or tangent by the similarity scale factor when moving between similar triangles. Lengths scale; ratios do not. If a triangle is enlarged threefold, every side triples and every trigonometric ratio is unchanged, because both the numerator and the denominator tripled. A question that hands you a scale factor alongside a ratio is usually testing precisely this, and the scale factor is there to be ignored.

The card

CIRCLES — reference card
A circle is every point a fixed distance r from a fixed centre. Every fact below is that sentence.
Standard form: (x - h)² + (y - k)² = r². Centre (h, k), radius r.
Signs flip: (x + 4)² -> h = -4. The form SUBTRACTS the coordinates.
The right side is r², not r. Square-root it. Then double it if the question said diameter.
General form: x² + y² + Dx + Ey + F = 0. Same coefficient on x² and y², no xy term.
Recovery: divide out the leading coefficient FIRST, group, move the constant, complete both squares, add to BOTH sides.
Complete the square: halve the coefficient, square it, add it to both sides; the HALF goes inside the parenthesis.
Shortcut: centre = (-D/2, -E/2) and r² = D²/4 + E²/4 - F. Use it to check the long way.
Inside / on / outside: compare (x - h)² + (y - k)² with r². Smaller is inside.
From a centre and a point on the circle: r = the distance between them.
From a diameter's endpoints: centre = midpoint, radius = HALF the distance.
Reference sheet gives you A = πr² and C = 2πr, and 360° = 2π radians. Nothing else on this page.
180° = π radians. Degrees -> radians: × π/180. Radians -> degrees: × 180/π.
One method for everything: fraction of the circle = θ/360 (deg) or θ/(2π) (rad).
Arc length = fraction × circumference. Sector area = fraction × area. Same fraction, different whole.
Radian-only shortcuts: s = rθ, A = ½r²θ, A = ½rs. Never feed degrees into these.
Perimeter of a sector = arc + TWO radii. An arc alone is not a perimeter.
Central angle = arc measure. Inscribed angle = HALF the arc it intercepts.
Angle inscribed in a semicircle (arms on a diameter) = 90°, always.
Two radii make an isosceles triangle. This is how a central angle becomes a rim angle.
Tangent ⊥ radius at the point of contact. Tangent length from an external point = √(d² - r²).
Chord: r² = d² + (c/2)², so c = 2√(r² - d²). Distance to the y-axis is |h|; to the x-axis is |k|.
Tangent to a line = distance from centre to that line equals r. Tangent to a curve = discriminant zero.
r² = 0 is a single point; r² < 0 is no graph at all. "Defines a circle" means r² > 0.
Before answering: radius or diameter? Length or area? Degrees or radians? Which vertex?

Why it works — Why the equation is the distance formula, and why s = rθ needs radians

Take the definition literally and the equation writes itself. A circle is every point (x, y) whose distance from the centre (h, k) equals r. The distance between those two points is √((x − h)² + (y − k)²), so the condition is √((x − h)² + (y − k)²) = r. Square both sides and you have (x − h)² + (y − k)² = r². The equation of a circle is not a formula to memorise; it is the distance formula with the radicals cleared, and three things the exam tests fall straight out of that. First, the right-hand side holds r² because the whole equation is a SQUARED distance — the squaring was what removed the radical. That is why the recovery operation is a square root and not a halving. Second, the coordinates appear subtracted because (x − h) is a DISPLACEMENT, the horizontal gap between the point and the centre; it is a difference by construction, so a plus sign printed inside the parenthesis can only mean the centre's coordinate is negative. Third, completing the square works because the general form is simply this equation with its squares multiplied out. Expanding (x − h)² gives x² − 2hx + h², so the Dx term you see in the general form is −2hx in disguise: the coefficient of x is twice the centre's coordinate with the sign flipped, which is exactly why halving D and flipping its sign recovers the centre, and why the constant you have to add back is that half, squared. You are not applying a trick — you are reversing a multiplication that already happened. The radian half has an equally short reason, and it explains why a formula can be unit-dependent at all. A radian is defined as arc divided by radius, θ = s/r. Both are lengths, so the units cancel and a radian measure is a bare number. Rearranging gives s = rθ with no conversion factor anywhere, because there is nothing to convert — the angle already IS the arc expressed in units of the radius. A degree, by contrast, is an arbitrary unit: 360 is a historical convention, not a property of circles, so a degree measure has to be multiplied by π/180 before it will behave like a ratio. Feed 150 into s = rθ and you get 12 × 150 = 1800, a number that answers no question, because you multiplied a length by an angle measured in an invented unit. The same reasoning generates the sector formula rather than requiring it to be memorised: a sector of angle θ radians is the fraction θ/(2π) of the disc, so its area is (θ/2π)(πr²) = ½r²θ — one line, and it will still be there when the memorised version has faded.

Traps — 6

Sign flip inside the parentheses

Standard form SUBTRACTS the centre's coordinates, so every sign printed inside a parenthesis is the opposite of the number it names: (x + 4)² + (y − 1)² = 20 is centred at (−4, 1), not (4, −1). The rule is mechanical, it is the same rule that governs vertex form for a parabola, and an answer choice built on the un-flipped reading appears on essentially every item that hands you a standard-form equation.

r² read as r

The right-hand side of the equation is a squared length, because the whole equation is the distance formula with the radical squared away. A right side of 20 is a radius of 2√5 ≈ 4.47, not 20 and not 10. The cousin of this trap runs the other way — building an equation from a radius of 6 and writing = 6 instead of = 36 — and its sibling is the diameter, since the exam will ask for the diameter precisely when the equation displays the radius.

Completing the square without keeping the equation true

Two versions of one failure — a step that changes what the equation says. UNBALANCED ADDITION: adding 25 to the left to build (x + 5)² and not adding it to the right, which makes the radius too small by exactly the amount that never crossed over. LEADING COEFFICIENT LEFT IN PLACE: general form only reads correctly when the x² and y² coefficients are 1, so 2x² + 2y² − 12x + 20y − 6 = 0 has to be divided through by 2 before anything else, and completing the square on the undivided version corrupts the centre and the radius together. Both produce a clean-looking circle that is not the one on the page.

Degrees inside a radian formula

s = rθ and A = ½r²θ are derived from the definition of a radian and are meaningless with a degree measure in them: a radius of 12 and an angle of 150 gives 1800, a number that answers nothing. The fraction method — θ/360 of the circumference — is immune to this by construction, which is the argument for using it as the default and treating the shortcuts as checks.

Arc answered as sector, sector answered as arc

The fraction of the circle is identical for both, so a student who computes the fraction correctly can still hand back the wrong quantity — a length where an area was wanted, or the reverse. The exam builds both into the same answer set for exactly this reason. Units settle it in one glance: a rim is a length, a slice is an area.

Inscribed and central angles interchanged

The angle at the centre is twice the angle at the rim standing on the same arc, and the direction is easy to reverse under time pressure. Two guards: an inscribed angle is always the smaller of the two, and an inscribed angle whose arms end at a diameter is exactly 90° — if you have computed 45° or 180° for such an angle, the halving went the wrong way.