The NEET-PG Error Log for Clinical Questions

Create one entry for every wrong answer, unanswered item, lucky correct answer, or correct answer that took much longer than planned. Record a short stem fingerprint: age group if relevant, setting, time course, two or three discriminating findings, and the actual task. Then record your choice, the keyed choice, and your confidence before reading the explanation. Confidence reveals dangerous high-certainty errors that raw accuracy hides.

Do not paste copyrighted question text or reproduce every option. Paraphrase the decisive structure in your own words. The ideal entry can be reviewed in under a minute and still recreate the reasoning fork. If the entry requires the original screen to make sense, it is incomplete. If it contains an entire page of background, it is too large to become a frequent retrieval cue.

Use a small stable taxonomy. A knowledge gap means you lacked a fact or mechanism. A discrimination error means you knew the candidates but missed the feature separating them. An interpretation error means you mishandled wording, timeline, data, or the requested next step. A premature-closure error means you stopped after the first plausible diagnosis. An execution error includes rushing, changing an answer without evidence, or spending too long.

Choose one primary cause even when several contributed. The primary cause is the earliest point at which a better action would have produced a better decision. For example, forgetting one criterion may look like a diagnosis error, but if the stem explicitly supplied the criterion and you skipped it, the primary failure is reading. Consistent classification lets you see whether more content study or better question behaviour will produce the larger gain.

A repair must be observable. For a knowledge gap, retrieve a concise concept from your trusted source, write a one-line rule, and answer fresh related items. For a discrimination error, build a two- or three-column comparison using only decisive features. For an interpretation error, write a reading trigger such as 'state whether the question asks diagnosis, investigation, mechanism, or management before looking at options.'

Avoid vague repairs such as 'revise medicine' or 'be careful.' They have no completion point. A better task is 'compare the two competing diagnoses by onset, hallmark clue, and contradicting finding; then solve five unseen vignettes.' When content concerns management, document the conceptual error but verify the current recommendation in an authoritative educational source rather than turning the error log into an unsourced treatment manual.

For each clinical vignette, identify the strongest supporting clue, the strongest conflict against your leading answer, and the exact task. This three-part check counters pattern matching based on a single familiar word. The conflict is especially useful: if a proposed answer requires ignoring the time course, a normal result, or a defining absent feature, it should lose priority even if several generic symptoms fit.

Log only clues that discriminate among the options, not every detail. Ask what each finding does to the relative plausibility of the closest two choices. Then separate diagnosis from the question's requested action. A learner may identify the condition correctly yet answer the wrong layer—mechanism instead of investigation, or definitive diagnosis instead of the immediate question. The task label prevents that category error.

Review a new entry after a short delay without exposing the answer. Recreate the stem fingerprint, state the discriminator, and explain why the tempting alternative fails. If retrieval succeeds, expand the interval. If it fails, simplify the rule or repair the underlying concept, then schedule an earlier return. Passive recognition of your own highlighted answer does not count as a successful review.

Retest with unseen questions because memory for the original item can imitate understanding. Mix repaired entries with other subjects so the cue is not supplied by the folder name. Retire an entry after repeated successful retrieval and transfer only genuinely volatile information to a compact final-revision list. Keeping every historical mistake active makes the log noisy and hides the few errors that still recur.

Once a week, count entries by primary cause, subject, confidence, and recurrence. Do not interpret tiny differences as precise statistics; look for obvious clusters. Several high-confidence discrimination errors in one system justify a comparison session. Repeated execution errors across subjects suggest a mock pacing or reading intervention. A single rare factual miss may need only one card.

End the audit with at most three changes for the next week. Examples include adding a task-label pause to every vignette, scheduling one targeted concept block, or enforcing a time cap before marking an item for review. Check the following week whether the relevant error type declined on new questions. The log earns its place only when it changes practice; otherwise simplify it until it does.