VS cheat sheet

Weight-sum rule: for a complete blueprint the domain weights total 100 percent; any other total means at least one weight is missing, duplicated, or mis-transcribed and must be re-checked against the source document.

Expected item count per domain = domain weight × total item count; observed counts that diverge far from expectation indicate either blueprint drift or mis-mapping of items.

Two-source rule: a blueprint claim is verified only when two independent primary sources agree; forums, secondary summaries, and third-party study guides are leads, not evidence.

Every internal topic is crosswalked to exactly one primary domain per the official outline's wording; secondary placements are recorded in a separate column rather than merged.

Scope boundary test: topical similarity to a domain is not sufficient evidence of placement; the outline must actually name the topic under that domain.

Unconfirmed facts are stored as null with a needs_review flag rather than guessed, so downstream consumers can see the gap explicitly.

Each derived artifact carries the blueprint version identifier and retrieval date, so stale content can be detected and retired when the outline changes.

Placeholder labels are provisional: keep an alias map from provisional to official domain names so existing items re-map cleanly once the real outline arrives.

Difficulty index p = number answering correctly ÷ number attempting the item, bounded between 0 and 1; higher p means an easier item, and p near 0.5 usually permits maximum discrimination.

Discrimination index D = p(upper group) − p(lower group), where groups are typically the top and bottom 27 percent of total scorers; rules of thumb: D ≥ 0.30 good, 0.20–0.29 marginal, below 0.20 revise.

Distractor diagnosis: an option chosen by fewer than about 5 percent of examinees is non-functioning, and an option chosen more often by high scorers than low scorers suggests a miskeyed item.

Standard error of measurement: SEM = SD × √(1 − reliability); it estimates the typical spread of observed scores around a true score and shrinks as reliability rises.

Guessing correction for k-option multiple choice: corrected score = R − W ÷ (k − 1), where R is rights and W is wrongs; omitted items are not penalized.

Item-total point-biserial correlation near zero or negative flags an item measuring something other than the intended construct, or an item whose key is wrong.

Internal-consistency coefficients such as KR-20 and Cronbach's alpha rise with more items and higher average inter-item covariance; they are not measures of validity.

Miskey check: if high scorers converge on one distractor while the designated key is chosen mainly by low scorers, treat the key as suspect before treating the item as flawed.

Retrieval practice produces larger delayed retention than rereading the same material for an equal amount of time, at the cost of feeling less fluent during study.

Expanding review intervals - for example 1 day, 3 days, 7 days, and 30 days - counteract the forgetting curve more efficiently than fixed or massed repetition.

Interleaving problem types within a session improves discrimination among types and transfer, although it depresses performance during practice relative to blocked study.

Working memory holds roughly four novel chunks at once; new material should therefore be introduced in small, linked units rather than as long unstructured blocks.

Retention of unrehearsed material decays approximately exponentially, with the steepest loss in the first day after initial learning.

Deliberate practice is defined by a targeted weakness, immediate feedback, and repeated correction - not by the total number of hours spent on already-easy tasks.

A mastery criterion (for example, several consecutive correct attempts) ends study of a topic and schedules a later review rather than stopping the schedule entirely.

Extraneous cognitive load from split sources, redundant text, or decorative detail competes with intrinsic load; reducing it frees capacity for the material's actual difficulty.

Average pace = total testing minutes ÷ total item count; track cumulative pace at checkpoints and adjust early rather than discovering the deficit at the end.

Two-pass allocation: the first pass answers only items solved with confidence and flags the rest; the second pass returns to flagged items so no easy item is lost to time pressure.

No penalty for wrong answers: answer every item, since the expected value of a blind guess is positive whenever p > 0.

Penalty of 1 ÷ (k − 1) per wrong answer: a guess has positive expected value only when eliminated options raise the correct probability above that penalty threshold.

Expected value of guessing = p(correct) × credit − p(wrong) × penalty; compare this value against the zero from omitting the item.

Change an answer only when a concrete reason supports the change, such as a misread qualifier or a recalled fact; undirected second-guessing tends to reduce accuracy.

Dimensional check: confirm that units match on both sides of a calculation, since a mismatch reveals an algebra or conversion error before any arithmetic is completed.

Read the stem and predict an answer before reading the options, and carry full precision through intermediate steps, rounding only the final result to the requested precision.