General Medicine Question Quality Review

Executive Summary

This review covers 200 validated non-gold General Medicine questions sampled across eight shards of 25 questions each. No benchmark or recent PYQ items were available as a gold standard for this subject, so quality judgments are grounded in the internal evidence of the sample and standard expectations for NEET-PG/INI-CET level assessment.

The Bloom's distribution is heavily skewed toward Level 4 (86 questions, 43%), with a reasonable Level 2–3 base (54 and 36 respectively) and a small but concerning Level 1 tail (21 questions, 10.5%). Only 3 questions reach Level 5. On paper this looks acceptable. In practice, the Level 4 classification is frequently aspirational: many questions tagged Bloom's 4 are either non-functional without a missing image or are thin recall items dressed in minimal clinical clothing.

The dominant finding across all eight shards is structural, not conceptual: a large fraction of the question set — conservatively estimated at 40–50% of the 200 questions — is wholly or partially non-functional because it references an image, ECG strip, clinical photograph, radiograph, or procedural visual that is not embedded in the question text. This is not a content quality problem that can be fixed by rewriting stems; it is a media-integrity problem that must be resolved before any other quality judgment is meaningful.

Beyond the image problem, five additional issue categories emerge consistently across shards: a cluster of factually incorrect or guideline-discordant answer keys; a pattern of low-yield recall items that fall below the PG entrance threshold; subject-boundary contamination from Surgery, Nursing, and allied specialties; a specific overuse of the "all except" format in ways that reduce rather than increase reasoning demand; and a small but important set of questions with structurally broken option sets or orphaned stems.

The sample contains genuinely strong material — particularly in cardiology clinical vignettes, tropical and infectious disease reasoning, rheumatology, and toxicology — that is worth preserving and expanding. The remediation burden is, however, substantial.

What Good Looks Like

In the absence of benchmark or PYQ gold-standard items for this subject, the following characteristics define the best questions observed in the reviewed set. These serve as the internal quality bar.

Clinical anchoring with discriminating detail. The best questions provide a brief but specific clinical scenario — age, sex, symptom timeline, one or two examination findings, a relevant investigation result — that narrows the differential without giving away the answer. Question 95e6d2fb (Rickettsial fever: Kashmir child, specific rash distribution, organomegaly) and 316aed4a (GPA triad: sinusitis, hemoptysis, RBC casts) demonstrate this well. The clinical detail does real work; removing any one element would make a different answer defensible.

Distractors that represent genuine clinical alternatives. In 2bd15277 (SLE in an adolescent), the distractors — C3, AEC, HLA-B5 — are all plausible investigations a candidate might order, and each tests a specific misconception. In 13c9be9b (scleroderma "except"), malar flush is a well-chosen foil because it is a real rheumatological finding, just not of scleroderma. Contrast this with questions where one distractor is obviously absurd (penicillin causing stress ulcers; subungual melanoma as a DVT risk factor), which collapse to three-option questions.

Text-completeness. The best questions in this set are fully answerable from the stem alone, or — where an image genuinely adds value — provide enough textual clinical context that the image enriches rather than enables the answer. Question 4add119b (LBBB criteria) provides a full textual ECG description: rate, axis, QRS duration, lead-specific morphology. Question 8105b9f4 (snakebite management) references a snake image but the clinical narrative — neurotoxic pattern, elapid features — is sufficient to answer the question independently.

Reasoning demand proportionate to the Bloom's tag. Questions like 60350736 (tuberous sclerosis with CT Hounsfield unit values for angiomyolipoma fat content) and 7f9248f7 (Romana sign in a traveller from Peru) require the candidate to integrate epidemiology, pathophysiology, and clinical findings — not retrieve a single memorised fact. The answer cannot be reached by elimination alone.

Factual accuracy and guideline currency. The best management questions reflect current practice. 1e76ab5b (sodium bicarbonate for TCA-induced broad QRS) and 6b4f41b3 (neurotoxic snakebite management nuances) are both factually sound and clinically current.

Main Issue Categories

1. Missing Image Assets: The Dominant Structural Failure

Why this pattern is bad

This is not a content quality issue — it is a functional completeness issue. A question that says "the image shows" or "comment on the diagnosis" with no image present is not a question at all; it is an unanswerable text stub. Candidates cannot reason from absent information. When the image is missing, one of three things happens: the question becomes a random guess; the answer leaks from the option set (if only one option is plausible without context); or the question tests nothing except the candidate's ability to recognise that something is missing. None of these outcomes is acceptable in a validated question bank.

The problem is compounded because many of the affected questions are tagged Bloom's 3–4 and classified as medium or hard difficulty. Their apparent cognitive demand exists only if the image is present. Without it, they are either trivial or non-functional, making the Bloom's and difficulty metadata misleading.

How it shows up

Across all eight shards, this pattern appears in an estimated 40–50% of the 200 questions. It manifests in several distinct forms:

Fully orphaned stems: The entire question is image identification with no clinical text. Examples: eed7fa42 ("Which of the following is correct about the image shown?"), 4044ee11 ("The image shows:"), 7ce9da6b ("What is the most possible diagnosis?" — no stem at all), fd332db7 ("The following image is suggestive of diagnosis of").
ECG-dependent questions: The question asks for rate calculation, rhythm identification, or axis determination from an ECG strip that is not present. Examples: 174d6304 (asks for a specific √RR numeric value), 9aba3882 ("Calculate the atrial rate"), 3b8295bc, 0b6e2dfc, 954628a0, 862f84d6.
Clinical photograph identification: Dermatology, neurology signs, ophthalmology, and procedural images. Examples: 6f250e85 ("The image shows elicitation of:"), a8ac9918 ("Identify the lesion"), 6fbe7654 (Hansen disease image), 180d8820 (deformity image), e741ef2e (gait video/image).
Radiological diagnosis: 778958ac (CT pancreatitis), 55e70a2d (NCCT stroke), e97cc0cc, 4e8ea1ea7 (USG breast).
Blank option sets: In the most severe cases, not only is the image missing but the option text itself is absent. 70dba40d has four blank option texts. 5396084e has options labelled only A, B, C, D with no content. 951dfaee has three blank options with only one populated.
LaTeX/rendering artefacts: 8901d0ae contains a \qquad placeholder in the stem, indicating a formatting failure that has left the question text incomplete.

Recommended disposition

This category requires a dedicated image-integrity audit before any other quality decision is made. The remediation path splits into three sub-groups:

Image confirmed present and rendering correctly: Proceed to content quality review. Many of these questions are conceptually strong.
Image absent but question is recoverable by adding clinical text: Fix. Rewrite the stem to include the key clinical or ECG findings in text form so the question is self-contained. This is feasible for ECG questions (rate, rhythm, morphology can be described) and for many clinical scenarios (e.g., 4b92eb44 — add "tall stature, lens dislocation, arm span > height" to make Marfan syndrome answerable without the image).
Image absent and question is pure visual identification with no clinical reasoning layer: Disable. Questions like eed7fa42, 4044ee11, fd332db7, 3d628f15 (identify a scientist from a photograph), 70dba40d, 5396084e, and 951dfaee cannot be salvaged by text rewriting because the entire cognitive task is visual recognition. Disable these until images are confirmed embedded and rendering is verified; even then, evaluate whether the question tests reasoning or merely recognition.

2. Factually Incorrect or Guideline-Discordant Answer Keys

Why this pattern is bad

A question with a wrong answer key is worse than no question at all. It actively miseducates candidates, penalises those who know the correct answer, and rewards those who have memorised the wrong fact. In a PG entrance context, where management questions directly map to clinical practice, guideline-discordant answers carry patient safety implications beyond the exam itself. This category is distinct from the image problem because it affects questions that are otherwise structurally sound and text-complete.

How it shows up

This pattern appears as a narrower cluster — approximately 8–12 questions across the reviewed set — but the errors are serious enough to warrant systematic attention.

CPR compression rate (7b353a48): The marked answer is "90 per minute, including neonates." Current AHA guidelines specify 100–120 per minute for adults and children. The neonatal rate uses a 3:1 compression-to-ventilation ratio at approximately 90 compressions per minute, but this is not the same as the adult rate and the framing is misleading. A companion question in the same shard (a8fba74f) correctly states ≥100/min, creating a direct internal contradiction in the bank.
Tension pneumothorax management (df2fc53d): The correct answer is marked as chest tube insertion. For immediate management of tension pneumothorax, needle decompression precedes chest tube. The question asks for "most appropriate immediate management," making the marked answer clinically incorrect.
Septicemia with hypotension (bc836096): "IV fluids alone" is marked as the primary management for septicemia with hypotension and low urine output. This contradicts the Surviving Sepsis Campaign, which mandates vasopressors alongside fluids when hypotension persists after initial resuscitation.
GCS minimum score (55e6a279): The correct answer is marked as 5. The minimum GCS is 3 (E1V1M1). This is a straightforward factual error in a foundational scoring question.
Scorpion sting management (e227d159): "Lytic cocktail" (promethazine + chlorpromazine + pethidine) is marked correct. Current Indian guidelines favour prazosin for autonomic storm in scorpion envenomation. The lytic cocktail is no longer recommended first-line.
Medullary sponge kidney extra-renal manifestation (879d6313): Congenital hepatic fibrosis is marked as the most common extra-renal manifestation of medullary sponge kidney. Congenital hepatic fibrosis is the extra-renal association of autosomal recessive PKD (ARPKD), not medullary sponge kidney. For medullary sponge kidney, the relevant associations are hypercalciuria and nephrolithiasis.
Shock classification with internally inconsistent parameters (e12e51e9): The stem describes excessive bleeding after RTA (implying hypovolaemic shock) combined with increased cardiac output and pulsus paradoxus. These findings do not map to any single shock category. The marked answer (distributive shock) is inconsistent with the haemodynamic parameters given.
Leptospirosis distractor (0953d3f5): Option D states "hepatorenal syndrome in 50% of cases." Weil's disease (severe leptospirosis) occurs in approximately 5–10% of cases, not 50%. The distractor is factually wrong in a way that could mislead candidates about disease severity.

Recommended disposition

All items in this category require fix before use. The fix path is specific to each item:

7b353a48: Correct the answer to 100–120/min; rewrite the neonatal caveat accurately.
df2fc53d: Change the correct answer to needle decompression, or reframe the stem as "definitive management."
bc836096: Restructure to reflect current sepsis bundle; either correct the answer or narrow the stem to the very first 30-minute fluid bolus phase.
55e6a279: Correct the answer key to 3.
e227d159: Update the correct answer to prazosin, or add a regional guideline qualifier.
879d6313: Clarify whether the condition is ARPKD or medullary sponge kidney and align the answer accordingly.
e12e51e9: Reconcile the haemodynamic parameters to match one shock category, or reframe as a "which finding is inconsistent" question.
0953d3f5: Replace the 50% distractor with an accurate figure or a different distractor entirely.

3. Low-Yield Recall Items Below the PG Entrance Threshold

Why this pattern is bad

A question that any MBBS graduate can answer correctly by simple recall — without clinical reasoning, differential diagnosis, or management judgment — does not discriminate between candidates who are ready for postgraduate training and those who are not. These items inflate the question count, distort difficulty metadata, and consume candidate time that could be spent on genuinely discriminating questions. They are particularly problematic when they carry Bloom's 1 or "easy" flags that the system has already correctly identified, yet they remain in the bank without remediation.

How it shows up

This pattern is broad, appearing across all eight shards. Approximately 20–25 questions in the reviewed set fall into this category. They cluster into recognisable sub-types:

Medical history and eponym trivia: a4bac6bb (who invented percussion?), 1f5accff (who invented the modern stethoscope?), 3d628f15 (identify a scientist from a photograph). These test historical recall with zero clinical application.
Definitional recall with no clinical context: e344cd1a (remittent fever = daily variation >1°C), 3da99430 (puddle sign = ascites), b78d7eb3 (emprosthotonus definition), 665ed447 (pneumatosis cystoides intestinalis definition), c9a58b67 (crenated tongue = lateral border indentation).
Single-fact procedural trivia: 3884f507 (needle gauge for RBC transfusion), 0760910a (parenteral iron formula), 5396084e (insulin injection sites — also image-dependent).
Epidemiological facts with no reasoning demand: 75ff46c6 (most common cause of PE = DVT), 61402e42 (most common cause of aortic aneurysm = atherosclerosis), ba07959d (metabolic abnormality in myeloma = hypercalcemia), cbbb5458 (pleural effusion detected clinically when >500 ml).
Implausible distractor sets that collapse to one-option questions: 62890cfa (stress ulcers caused by all except penicillin therapy — penicillin is an obvious outlier), bf3cd606 (DVT seen in all except subungual melanoma — no plausible mechanism for confusion), 80425c89 (mucinous ascites — pure recall, implausible distractors).

Recommended disposition

For items that are pure Bloom's 1 recall with no clinical context and no plausible path to elevation: disable. This applies to a4bac6bb, 1f5accff, 3d628f15, e344cd1a, 3da99430, b78d7eb3, 665ed447, c9a58b67, 3884f507, 0760910a, 75ff46c6, 61402e42, ba07959d, cbbb5458, 62890cfa, bf3cd606, 80425c89, 012dc381.

For items where the underlying concept is worth testing but the question is too thin: fix by embedding the fact in a clinical scenario. For example, fbbf4ac9 (small BP cuff causing false elevation) is factually correct and text-complete but Bloom's 2 with no clinical anchor. Fix: reframe as an obese patient with a measured BP of 160/100 — what is the most likely source of error? This reaches Bloom's 3 without requiring a full rewrite.

The threshold for General Medicine PG questions should be a minimum of Bloom's 2 with mandatory clinical context, and Bloom's 3 or above for any question that is not a straightforward definition or classification.

4. Subject-Boundary Contamination

Why this pattern is bad

Questions that belong to Surgery, Urology, Nursing, Microbiology, or Pathology inflate the General Medicine question count without adding medicine-specific depth. More importantly, they displace genuinely high-yield General Medicine content and distort coverage analysis. A candidate preparing for General Medicine should not encounter questions about paraphimosis management, breast fibroadenoma ultrasound, or nursing interventions for constipation in this subject's bank. Subject misclassification also means these questions are likely absent from the banks where they belong, creating coverage gaps in those subjects.

How it shows up

This pattern appears as a moderate-sized cluster — approximately 15–20 questions across the reviewed set — with several distinct sub-types:

Surgery content: df2fc53d (tension pneumothorax management — Cardiothoracic Surgery), 4e8ea1ea7 (USG breast — Surgery), 83ea1068 (esophageal carcinoma — Surgery/GI), f5c53ddc (paraphimosis — Urology), d955218f (gastric lavage tube identification — Surgery/Emergency), 33b4d72e (thyroid/neck swelling with bosselated appearance — Surgery).
Microbiology content: 9521e299 (Yersinia pestis — Microbiology), cdcaf1ca (Echinococcus granulosus life cycle — Microbiology/Parasitology).
Pathology content: 4044ee11 (peripheral blood smear identification — Pathology/Haematology).
Nursing/allied health content: 025b30a8 (nursing intervention for constipation), 1e36e0ef ("it burns when I urinate" as a UTI indicator). These are written at a nursing-school cognitive level and carry no NEET-PG tags.
Neurology/Orthopaedics content: e741ef2e (gait abnormality in polio — Neurology/Orthopaedics), 180d8820 (ulnar nerve deformity — Surgery/Neurology).

Recommended disposition

Questions that are factually sound and well-constructed but belong to another subject: reassign to the appropriate subject bank rather than disabling. This applies to df2fc53d, 4e8ea1ea7, 83ea1068, f5c53ddc, 33b4d72e, 9521e299, cdcaf1ca, e741ef2e, 180d8820.

Questions that are both misclassified and below PG threshold or structurally broken: disable. This applies to 025b30a8, 1e36e0ef, 4044ee11 (also image-dependent), d955218f (also image-dependent).

A systematic subject-tag audit of the full 218-question set is warranted. The contamination rate in this sample suggests that a meaningful fraction of the bank's apparent breadth is borrowed from other subjects rather than earned through genuine General Medicine coverage.

5. The "All Except" Format Used as a Substitute for Clinical Reasoning

Why this pattern is bad

The "all are true except" or "which of the following is incorrect" format is not inherently problematic — it is a legitimate item type when the exception tests a genuine misconception and the other options are all plausible true statements that require knowledge to confirm. The problem in this sample is that the format is overused (appearing in approximately 28–35% of questions across shards) and is frequently deployed in ways that reduce rather than increase reasoning demand. Specifically: when one distractor is obviously wrong by common sense, the question collapses to a three-option positive question. When the format is combined with image dependency, it creates a double failure — the candidate cannot interpret the image and cannot evaluate the exception logic. When the format is used to test a single isolated fact (the exception), it rewards memorisation of one outlier rather than integrated understanding.

How it shows up

"Except" + image dependency (double failure): 3b1e7744, 2d1ab0c1, 9521e299, b7a3859b, 259cd6be, 239a2bdd, e11c9245, 1d1a9fb6, 50a5abc8, 416686d4, bf475a81. These questions are non-functional without the image and the exception logic cannot be evaluated independently.
"Except" with an obvious outlier distractor: 8a9767a9 (hypercalcemia vs. hypocalcemia after massive transfusion — near-giveaway), 62890cfa (penicillin as a stress ulcer cause — obviously wrong), bf3cd606 (subungual melanoma as DVT risk — obviously wrong).
"Except" testing a single memorised fact with no clinical scenario: f6600a9b (chylous ascites causes), 665ed447 (pneumatosis cystoides intestinalis), b78d7eb3 (emprosthotonus).
"Any of the above" as the keyed correct answer: 8e63e21e (recurrent aphthous ulcers). This is a well-documented item-writing flaw that rewards test-taking strategy over knowledge.

Recommended disposition

The format itself does not require elimination, but the following rules should apply:

Any "except" question that is also image-dependent: fix by resolving the image dependency first; if the image cannot be confirmed, disable.
Any "except" question where one distractor is obviously implausible: fix by replacing the weak distractor with a genuine foil, or convert to a positive-stem clinical reasoning question.
Any "except" question that tests a single isolated fact with no clinical context: fix by adding a clinical scenario, or disable if the concept is already well-covered by stronger items.
8e63e21e ("any of the above"): fix by replacing the keyed answer with a specific single correct answer and rewriting the option set accordingly.

The content team should consider a bank-wide audit of "except" format prevalence in General Medicine and set a ceiling (e.g., no more than 20% of questions in any topic cluster should use this format).

6. Clinically Thin Vignettes That Collapse to Single-Fact Recall

Why this pattern is bad

This category is distinct from the pure recall items in Category 3. These questions have a clinical scenario — a patient, a symptom, sometimes a finding — but the scenario adds no discriminating information. The clinical wrapper is cosmetic: removing it would not change which answer is correct, because the vignette contains no detail that differentiates between the distractors. These questions are tagged Bloom's 3–4 because they have a clinical stem, but they function as Bloom's 1–2 recall items. This inflates the apparent cognitive demand of the bank and misleads difficulty calibration.

How it shows up

This pattern appears as a moderate cluster — approximately 12–18 questions across the reviewed set:

75ff46c6: DVT → PE. The vignette adds no detail that distinguishes PE from other causes of dyspnoea. The answer (DVT as the most common cause of PE) is a pure recall fact.
68a906b1: Bee sting → distributive shock. The clinical scenario is present but the answer is immediately apparent from the mechanism; no differential reasoning is required.
eedb1423: Marathon runner with heat stroke. The symptom list maps directly to the answer without any competing diagnosis.
cbbb5458: Pleural effusion detection threshold. The "clinical scenario" is a single sentence that adds nothing to the recall fact.
62d2535d: Most common symptom of duodenal ulcer. No clinical detail that would distinguish duodenal from gastric ulcer or other GI pathology.
ccdaff25: Achalasia + bird's beak — the answer is embedded in the stem itself ("bird's beak appearance on barium swallow" is the diagnostic finding for achalasia; asking "what is the diagnosis?" when the pathognomonic finding is already named is circular).
abe6180b: Tachycardia + BP <100 + postural hypotension → bleeding peptic ulcer. The same haemodynamic findings occur in any hypovolaemic state; no discriminating detail (melena, NSAID use, epigastric pain) is provided.
ac16b2f8: GERD-to-stricture vignette without the key discriminating feature (progressive dysphagia for solids before liquids).

Recommended disposition

These questions are worth fixing rather than disabling because the underlying clinical concept is appropriate for PG-level assessment. The fix in each case is the same: add one or two discriminating clinical details that make the vignette do real diagnostic work.

ccdaff25: Disable (answer is in the stem; no fix path that preserves the question).
abe6180b, ac16b2f8, 68a906b1, eedb1423: Fix by adding discriminating clinical detail.
75ff46c6, cbbb5458, 62d2535d: Disable — the concept is too basic for PG level and is better covered by stronger items elsewhere in the bank.

7. Structurally Broken Items (Orphaned Stems, Blank Options, Rendering Failures)

Why this pattern is bad

This category covers questions that are non-functional for reasons other than missing images — specifically, stems with no clinical content at all, option sets that are partially or fully blank, and rendering artefacts that have corrupted the question text. These are the most straightforward disable calls in the bank because there is no content to evaluate; the question does not exist in any meaningful sense.

How it shows up

This is a narrow cluster — approximately 8–10 questions in the reviewed set — but each item is a complete failure:

Fully orphaned stems with no clinical content: 7ce9da6b ("What is the most possible diagnosis?" — no narrative, no image), 160305b9 ("The above marked structures are seen in a condition where the most common cause of death is?" — no image, no description of the structures).
Blank or near-blank option sets: 70dba40d (all four option texts are blank), 5396084e (options labelled A, B, C, D with no text content), 951dfaee (three of four options are blank; only "Alport syndrome" is populated).
LaTeX/rendering artefacts: 8901d0ae contains a \qquad placeholder in the stem, indicating a formatting failure that has left the question text incomplete.
Single-sentence stems with no clinical information: 193ad435 ("Comment on the diagnosis" — one sentence, no patient details, no ECG description), 862f84d6 ("Comment on the diagnosis" — single sentence, zero clinical information).

Recommended disposition

All items in this category: disable immediately. There is no content to evaluate and no fix path that does not amount to writing a new question from scratch. If the underlying concept is worth testing, a new question should be authored rather than attempting to reconstruct these stubs.

Prioritization

The seven issue categories are not equally urgent. The following priority order reflects both the volume of affected questions and the severity of the quality failure.

Priority 1 — Image Asset Integrity Audit (Category 1) This must happen before any other remediation. Approximately 40–50% of the 200 questions are affected. Until image attachment status is confirmed for every question, it is impossible to know how many of the apparent Bloom's 3–4 questions are actually functional. The audit should produce three lists: confirmed-present, confirmed-absent-but-recoverable, and confirmed-absent-not-recoverable. Remediation for the latter two groups follows the fix and disable paths described above.

Priority 2 — Factually Incorrect Answer Keys (Category 2) Approximately 8–12 questions have wrong or guideline-discordant correct answers. These must be fixed before any question in this category is served to candidates. The errors are specific and the fixes are well-defined. This is a small but high-severity group.

Priority 3 — Structurally Broken Items (Category 7) Approximately 8–10 questions are completely non-functional. These should be disabled immediately as a batch operation; no content review is needed.

Priority 4 — Low-Yield Recall Items (Category 3) Approximately 20–25 questions fall below the PG entrance threshold. Disabling these is straightforward for the pure Bloom's 1 items; the borderline items require a fix decision. This is a medium-volume, medium-effort remediation.

Priority 5 — Subject-Boundary Contamination (Category 4) Approximately 15–20 questions belong to other subjects. Reassignment is lower urgency than the above categories but should be completed before the next coverage analysis to avoid false impressions of General Medicine breadth.

Priority 6 — "All Except" Format Overuse (Category 5) This is a diffuse pattern that requires individual item review rather than batch action. The highest-priority sub-group (except + image dependency) is already captured in Priority 1. The remaining items can be addressed in a second pass.

Priority 7 — Clinically Thin Vignettes (Category 6) These questions are functional and factually correct; they are just not discriminating enough. Fixing them improves bank quality but does not prevent harm. Address after the higher-priority categories are resolved.

Example Keep / Fix / Disable Calls

The following calls are drawn from across the reviewed set and are intended to illustrate the application of the issue categories above. They are not exhaustive.

KEEP

Question ID	Rationale
`95e6d2fb`	Rickettsial fever vignette with geographic context, specific rash distribution, organomegaly. Bloom's 4, text-complete, plausible distractors. Keep as-is.
`316aed4a`	GPA triad (sinusitis, hemoptysis, RBC casts). Well-constructed, appropriate distractors, Bloom's 4. Keep as-is.
`1e76ab5b`	TCA toxicity with ECG findings and management. Self-contained, factually current, good distractor logic. Keep.
`2bd15277`	SLE in adolescent girl. Clean vignette, ANA as correct answer is defensible, distractors test specific misconceptions. Keep.
`6d7c7f18`	Myasthenia gravis, ice pack test. Good clinical vignette, high-yield concept, text-complete. Keep.
`d910d9c6`	Peutz-Jeghers (melena + lip pigmentation). Clean stem, unambiguous answer, good distractors, PYQ-tagged. Keep.
`60350736`	Tuberous sclerosis with CT HU values for angiomyolipoma. Multi-system integration, Bloom's 4. Keep pending image confirmation.
`7f9248f7`	Romana sign + Triatomine bug (Chagas disease) in traveller from Peru. Epidemiological context well-crafted. Keep.
`13c9be9b`	Scleroderma vignette with anti-Scl-70, malar flush as the correct exception. Bloom's 4, good clinical reasoning demand. Keep.
`27e47ff3`	RTA type I → hypokalemia → ECG findings. Multi-step reasoning, Bloom's 3, clinically sound. Keep.
`ac28b2d6`	Syndrome least associated with obesity (Sipple/MEN 2A). Good negative-association reasoning, plausible distractors. Keep.
`8105b9f4`	Neurotoxic snakebite management "except." Nuanced, Bloom's 4, text-sufficient even with image reference. Keep.

FIX

Question ID	Issue	Recommended Fix
`7b353a48`	CPR rate marked as 90/min — factually wrong	Correct answer to 100–120/min; rewrite neonatal caveat accurately
`55e6a279`	GCS minimum marked as 5 — should be 3	Correct answer key to 3
`bc836096`	IV fluids alone marked correct for septicemia with hypotension	Correct answer or narrow stem to initial 30-minute resuscitation phase
`df2fc53d`	Chest tube marked as immediate management for tension pneumothorax	Change correct answer to needle decompression, or reframe as "definitive management"
`e227d159`	Lytic cocktail marked correct for scorpion sting	Update to prazosin or add regional guideline qualifier
`879d6313`	Congenital hepatic fibrosis attributed to medullary sponge kidney	Clarify condition (ARPKD vs. MSK) and align answer
`e12e51e9`	Internally inconsistent haemodynamic parameters	Reconcile parameters to one shock category or reframe as "which finding is inconsistent"
`174d6304`	Asks for √RR numeric value from absent ECG	Embed ECG or rewrite as a conceptual calculation question with explicit values
`9aba3882`	"Calculate atrial rate" with no ECG	Embed atrial flutter ECG or rewrite with textual description of sawtooth pattern
`b587ab10`	"Constipation" listed as a finding in hyperthyroidism — creates second defensible answer	Replace "constipation" with a true feature of hyperthyroidism
`8e63e21e`	"Any of the above" as keyed correct answer	Replace with a specific single correct answer; rewrite option set
`4b92eb44`	Disease unnamed, image absent (Marfan syndrome implied)	Add clinical descriptors (tall stature, lens dislocation, arm span > height)
`ac16b2f8`	GERD stricture vignette missing key discriminating feature	Add "progressive dysphagia for solids before liquids"
`0953d3f5`	Distractor states hepatorenal syndrome in 50% of leptospirosis cases	Replace with accurate figure (~5–10%) or different distractor
`fbbf4ac9`	BP cuff artefact — factually correct but Bloom's 2, no clinical anchor	Reframe as obese patient with measured BP 160/100 — source of error?
`c1e61c73`	PSVT on Holter — image absent	Embed Holter strip or provide full textual ECG description
`33b4d72e`	Neck swelling question references phantom image	Rewrite stem as self-contained: "A patient with multinodular goiter…"

DISABLE

Question ID	Rationale
`eed7fa42`	Pure image identification (Dracunculus), no clinical stem, Bloom's 1. Non-functional without image.
`4044ee11`	"The image shows:" — three-word stem, peripheral smear identification only, no clinical reasoning layer.
`70dba40d`	All four option texts are blank. Completely non-functional.
`5396084e`	Options labelled A–D with no text content. Non-functional.
`951dfaee`	Three of four options are blank. Structurally broken.
`7ce9da6b`	"What is the most possible diagnosis?" — no stem, no image. Orphaned.
`160305b9`	"The above marked structures…" — no image, no description. Orphaned.
`a4bac6bb`	"Who discovered percussion?" — medical history trivia, Bloom's 1, no clinical application.
`1f5accff`	"Who invented the modern stethoscope?" — same category as above.
`3d628f15`	"Identify the scientist in the image" (Pasteur). Bloom's 1, image-dependent, no medical reasoning.
`0760910a`	Parenteral iron formula — pure numerical recall, Bloom's 1, no clinical decision-making.
`3884f507`	Needle gauge for RBC transfusion — procedural trivia, Bloom's 1.
`75ff46c6`	DVT as most common cause of PE — first-year MBBS fact, flagged easy, no clinical scenario.
`cbbb5458`	Pleural effusion detected when >500 ml — Bloom's 1, no clinical context.
`c9a58b67`	Crenated tongue definition — definitional recall, flagged easy.
`e344cd1a`	Remittent fever definition — Bloom's 1, flagged easy.
`3da99430`	Puddle sign = ascites — Bloom's 1, flagged easy, no NEET-PG tag.
`012dc381`	Menghini needle = liver biopsy — Bloom's 1, flagged easy.
`025b30a8`	Nursing intervention for constipation — not NEET-PG content, below cognitive threshold.
`1e36e0ef`	"It burns when I urinate" as UTI indicator — nursing-level, Bloom's 2, no NEET-PG tag.
`62890cfa`	Stress ulcers "except penicillin" — obvious outlier, no vignette, flagged easy.
`bf3cd606`	DVT "except subungual melanoma" — obvious outlier, no vignette, flagged easy.
`80425c89`	Mucinous ascites — pure recall, implausible distractors, flagged easy.
`665ed447`	Pneumatosis cystoides intestinalis definition — Bloom's 1, no clinical reasoning.
`fd332db7`	"The following image is suggestive of" — one-line stem, image-only, Bloom's 2 at best.
`ccdaff25`	Answer (achalasia) is embedded in the stem (bird's beak appearance named explicitly). Circular question.
`e741ef2e`	Gait abnormality — entirely image/video dependent, no text fallback, also misclassified (Neurology/Orthopaedics).
`f5c53ddc`	"What is the clinical diagnosis?" — no clinical text, four urological options, entirely image-dependent. Also misclassified (Urology).
`d955218f`	"What is the function of the tube shown below?" — no image, no description. Also misclassified (Surgery/Emergency).