Inferences System

The correct inference is the boring one — the most cautious conclusion that the text forces to be true. Inference questions are the most-missed R&W type (consensus: PrepScholar, UWorld, PrepMaven) because students pick the answer that is interesting or plausible in the real world rather than the one the passage actually proves. Reframing the task from "what's a good conclusion?" to "what must be true given only these sentences?" is the whole fix.

Sources: Meltzer; UWorld 800 guide; PrepMaven. Examples below are MERIDIAN-original. Point convention: 1 of ~54 R&W questions; Module-1 misses are ceiling-risk.

---

#WHAT IT TESTS

A short passage states 2–3 facts and ends with a blank, usually after a logical cue ("Therefore ___" / "This suggests that ___" / "It can reasonably be concluded that ___"). You pick the choice that logically follows. The reasoning is deductive: the answer must be entailed by the text, using no outside knowledge.

---

#THE METHOD — WHAT MUST BE TRUE

1. READ the facts. Note exactly what is stated — and what is NOT.
2. DO NOT pre-write a clever conclusion. Stay close to the text.
3. For each choice ask: "Does the passage GUARANTEE this?"
   - Guaranteed by the text → keep.
   - Requires one extra assumption / outside fact → eliminate.
4. Pick the most CAUTIOUS choice that is fully supported.

Why this works (mechanism): every wrong answer is wrong for a findable reason — it adds an idea, overgeneralises, predicts the future, or asserts a cause the text never establishes. The right answer often feels underwhelming precisely because it claims only what is proven. Train yourself to prefer the modest choice. [This single reframe is the largest single-type gain available in R&W — UWorld.]

---

#TRAP TAXONOMY

• One step too far: true-ish but goes beyond the evidence (a stronger claim, a prediction, a generalisation to all cases).
• Causation from correlation: the text shows two things co-occur; the trap says one causes the other.
• Outside knowledge: true in the real world, but not stated or implied here.
• Reversed logic: flips the direction of the relationship the passage describes.
• Half-right: correct about one fact, wrong about a second clause (read the whole choice).

---

#SAME ITEM, TWO LEVELS (MERIDIAN-original)

Passage: "In the trial, the drug reduced symptoms in patients who also exercised daily. Patients who took the drug without exercising showed no improvement. Researchers noted that exercise levels were self-reported. Therefore ___" Choices: (A) the drug is ineffective without exercise. (B) exercise cures the condition on its own. (C) the observed improvement may depend on factors the study could not fully verify. (D) daily exercise should be prescribed to all patients.

~700 reasoning (trap): "No improvement without exercise → the drug needs exercise." → (A). Trap: "one step too far" + causation — the study only observed an association, and exercise was self-reported (unverified), so it can't guarantee the drug "is ineffective without exercise." [−1; ceiling risk in Module 1]

780+ reasoning (precise): What's guaranteed? Improvement appeared with self-reported exercise; the text flags that exercise data is unreliable. The only claim the text forces is caution about the finding. Eliminate (A) too strong/causal; (B) outside-knowledge + reversed; (D) a recommendation, not an entailment. → (C) — the cautious, fully-supported conclusion. The move: chose the modest claim and used the "self-reported" detail the trap ignored. [secures the question; Module-1 safe]

---

#DIAGNOSE YOUR ANSWER

• Did you pick the strongest/most interesting choice? That's the "one step too far" trap. Fix: prefer the most cautious supported claim.
• Did your answer use a real-world fact not in the text? Outside-knowledge trap. Fix: cover the choice; can you point to the sentence that proves it? If not, eliminate.
• Did you turn "linked" into "caused"? Correlation→causation trap. Fix: the text must state the mechanism, not just co-occurrence.
• Right idea, wrong second half? You didn't read the whole choice. Fix: validate every clause.

---

#DRILLS (timed · self-marked · MERIDIAN-original)

Drill 1 — Must-be-true filter (target: 5/5, ≤4 min). For each mini-passage, mark each candidate conclusion T (guaranteed) or X (overreach/outside/causal):

1. Facts: "All three samples that were heated expanded. Sample four was not heated." → (a) "Sample four did not expand" [X — not guaranteed] (b) "Heating expanded every sample that was heated" [T]
2. Facts: "Sales rose after the ad campaign; the firm also cut prices that month." → (a) "The campaign caused the rise" [X — confound] (b) "At least one change coincided with rising sales" [T]
3. Facts: "The author praises the novel's structure but never discusses its characters." → (a) "The author dislikes the characters" [X — silence ≠ dislike] (b) "The author's praise focuses on structure" [T] PASS: all tags correct (5/5 across the marked items). FAIL: re-read the trap taxonomy; redo tomorrow.

Repetition protocol: fail → repeat next day. Two clean passes → the "must be true" reflex is forming.

---

#TRANSFER

Abstract rule: pick the choice the text entails; eliminate anything needing an extra assumption — works on any topic. Second context: a humanities passage ("the poet revised the line twice; the final version is shorter → ___") and a science passage ("the reaction stopped when the catalyst was removed → ___") are the same deductive task. Stays constant: "what must be true?" + the trap list. Changes: the subject only.

---

#REFERENCE CARD — INFERENCES

Ask of every choice: "Does the passage GUARANTEE this?"  Keep only what's entailed.
Pick the MOST CAUTIOUS supported choice — the boring one.
TRAPS: one step too far · correlation→causation · outside knowledge · reversed · half-right
Silence ≠ proof. Plausible ≠ provable.
Each item = 1 R&W question; Module-1 miss = ceiling risk.

---

MERIDIAN™ · © 2026 · For Neo's personal study use. Not affiliated with or endorsed by the College Board. SAT® is a trademark of the College Board.

MERIDIAN · FORGE · INFERENCES SYSTEM | v1.0