Ophthalmology Question Quality Review

Executive Summary

This report synthesizes findings from 200 validated non-gold candidate questions reviewed across eight shards of 25 questions each. The sample is drawn from a pool of 6,703 Ophthalmology questions after excluding benchmark and recent PYQ items.

The overall picture is one of a large, structurally uneven bank. The subject has genuine strengths — a meaningful subset of clinical vignette questions in Glaucoma, Retina, and Pediatric Ophthalmology that approach or meet the benchmark standard — but these are outnumbered by a heavy mass of low-Bloom recall items, a recurring pattern of answer-key errors in negation-format questions, a structurally broken category of image-dependent questions rendered without images, and a non-trivial number of outright factual errors that would actively mislead candidates if deployed.

The Bloom's distribution tells the story quantitatively: 69 Bloom's-1 and 103 Bloom's-2 items account for 86% of the 200-question candidate sample. Only 19 questions reach Bloom's-3 and 9 reach Bloom's-4. The benchmark set, by contrast, mixes recall anchors with genuine clinical reasoning items and image-based diagnosis. The candidate pool is structurally inverted relative to what INI-CET and NEET-PG actually test.

Headline numbers from the reviewed set:

Disposition	Approximate Count	Basis
Keep (usable as-is or with minor polish)	~55–65	Clinically framed, factually sound, adequate distractors
Fix (usable after targeted remediation)	~70–80	Correctible stem, key, tag, or distractor issue
Disable (retire or full rewrite required)	~55–70	Factual error, non-functional, trivial recall, or wrong subject

The five main issue categories identified from the evidence are: (1) Factual Recall Overload, (2) Answer-Key Errors and Factual Inversions, (3) Broken Image Dependency, (4) Negation-Stem Structural Weakness, and (5) Clinical Vignette Inflation and Distractor Poverty. A sixth narrower category — (6) Topic Misclassification and Wrong-Subject Contamination — is operationally distinct enough to warrant separate treatment because it implies a different remediation path (reclassification or deletion rather than content editing).

What Good Looks Like

The benchmark and PYQ sets establish a clear quality ceiling. The best items in those sets share four properties that the candidate pool frequently fails to replicate.

Clinical framing that does real work. Benchmark item 56ec5ba8 (post-cataract endophthalmitis, 2 weeks, Propionibacterium) gives a timeline, a clinical picture, and four plausible diagnoses. The timeline — two weeks, not two days — is the discriminating detail that separates Propionibacterium from Staphylococcus. The scenario is not decorative; it is the question. PYQ item 24349a15 (80-year-old, IOP 50 mmHg, deep AC flare, white cataract → phacolytic glaucoma) similarly requires the candidate to integrate age, IOP asymmetry, and lens status before arriving at the answer. The candidate pool contains many questions that dress a single fact in a clinical costume without requiring any integration.

Distractors that represent genuine clinical alternatives. Benchmark item 301d9006 (optic neuritis) offers four options that each represent a real clinical possibility — bilateral loss, unilateral loss without pain, pain alone, and the correct combination. A candidate who does not know the full triad cannot eliminate options by common sense. Many candidate-pool questions fail this test: the wrong options are either obviously absurd or represent conditions from a completely different clinical domain.

Factual precision on multi-component answers. Benchmark item 106dcf8b (acute hemorrhagic conjunctivitis) correctly identifies the combination of Coxsackie A and Enterovirus 70, not just one agent. Candidate question b30fe928 asks the same question but marks Enterovirus 70 alone as correct — a partial answer that would mislead candidates preparing for an exam that tests the full combination.

Appropriate use of image-based diagnosis. PYQ item dd3e6198 (nodular episcleritis, phenylephrine blanching) is a Bloom's-4 image-based item that cannot be answered without the image. It is functional because the image is present. The candidate pool contains multiple questions that reference images that are absent, making them non-functional by definition.

Negative-format questions used sparingly and precisely. PYQ item 9c35af8a (pterygium NOT a feature — 95% recurrence rate) uses the negative format to test a specific numerical fact that is commonly misremembered. The exception is non-obvious and clinically meaningful. The candidate pool uses the negative format far more broadly, often to test trivial exclusions where the odd-one-out is immediately apparent.

Main Issue Categories

1. Factual Recall Overload: A Structural Bloom's Imbalance Across the Entire Sample

Why this pattern is bad

INI-CET and NEET-PG questions at the level of the benchmark set require candidates to apply knowledge to clinical scenarios, integrate multiple findings, or reason about mechanisms. A bank dominated by Bloom's-1 single-fact recall items does not prepare candidates for this and does not discriminate between candidates who understand a topic and those who have merely memorised a list. When 86% of the candidate sample sits at Bloom's 1–2, the bank is structurally misaligned with the exam it is meant to serve.

How it shows up

This is the broadest pattern in the reviewed set, appearing in every shard. It manifests as: single-eponym lookups ("snow-storm cataract is also known as"), single-definition items ("synchysis refers to liquefaction of the vitreous"), single-anatomy facts ("thinnest part of the sclera"), single-drug-to-indication mappings ("PASCAL is used for laser photocoagulation"), and single-numerical facts ("Snellen's test uses an angle of 1 minute"). These items have no clinical stem, no reasoning requirement, and no distractors that require knowledge to eliminate.

The pattern is particularly dense in the following topic clusters observed in the reviewed set: optics and refraction anatomy, corneal layer anatomy, basic pharmacology (cycloplegics, antiglaucoma drugs), eponymous signs and classifications, and community ophthalmology administrative facts.

Example question IDs with short explanations

Q529b02a8 — "A true diabetic cataract is also known as: snow-storm cataract." Pure eponym recall. No clinical context. Bloom's-1. Dozens of similar items exist in the bank.
Q3a92ca1b — "Synchysis refers to: liquefaction of the vitreous." Single-term definition. No discriminatory value at PG level.
Q3d5966d8 — "Roth spot = retinal hemorrhage with white center." The topic (Roth spots) is clinically relevant, but the question tests only the definition, not recognition in a clinical scenario or association with a systemic disease.
Qa4914c6b — "PASCAL is used for: laser photocoagulation." Device-name-to-function mapping. Bloom's-1.
Q037df42d — "Which ocular structure has the highest refractive index?" Pure anatomy recall, no clinical relevance.
Qc6cd8c07 — "Snellen's test uses an angle of 1 minute." Rote numerical fact.
Q6ebd6df2 — "Thinnest part of the sclera." Pure anatomy recall with no surgical or trauma hook.
Q73a53983 — "Intercanthal distance for traumatic telecanthus is 40 mm." Isolated numerical fact.
Q82255f60 — "Definition of anisometropia." Single-word definition recall.
Q4f73920f — "Most common site of BCC on eyelid." Isolated fact with no clinical scenario.
Q318f9e5f — "Fastest cycloplegic = tropicamide." Textbook recall with no clinical context.
Q9cb6d1e9 — "Most common refractive surgery for myopia = LASIK." Universally known fact.
Q29ab3001 — "Phoria is a type of latent squint." Single-line definition.
Q5bd9d817 — "Primary site of aqueous humor secretion = pars plicata." Anatomy recall with no clinical hook.
Q2ecdd264 — "Most vascular structure of the eyeball = choroid." Trivially easy anatomy.
Q41b52f90 — NPCB secondary level managed by NGO eye hospital. Administrative community ophthalmology trivia with no clinical application.

Recommended disposition

Items in this category that test facts with no clinical application and where the concept is already covered by stronger items in the benchmark or PYQ sets: Disable. Items where the underlying concept is high-yield but the question is purely recall: Fix by converting to a clinical vignette or applied scenario. The content team should set a minimum standard that any Bloom's-1 item must test a fact that is (a) genuinely high-yield for INI-CET/NEET-PG, (b) not already covered by a stronger item, and (c) not trivially eliminable by common sense.

2. Answer-Key Errors and Factual Inversions: A Patient Safety–Level Quality Risk

Why this pattern is bad

Questions with wrong answer keys do not merely waste a slot in the bank — they actively teach incorrect medicine. A candidate who studies from a question that marks "vitreous hemorrhage" as the primary determinant of refractive power, or "primary open-angle glaucoma" as the cause of painful vision loss, will carry that error into clinical practice and into the exam. This is the highest-severity category in the reviewed set. It is also the category most likely to generate candidate reports and platform credibility damage.

How it shows up

The pattern appears in two sub-forms. The first is a direct factual inversion, where the keyed correct answer is simply wrong on the underlying biology or clinical fact. The second is a partial-answer error, where the keyed answer is incomplete relative to the established standard (as in the conjunctivitis causation example), and a more complete answer exists in the benchmark set.

This pattern appears across multiple shards and multiple topic areas, suggesting it is not confined to a single content contributor or topic cluster. It is observed in corneal physiology, glaucoma, uveitis, retinal vascular disease, and systemic ocular manifestations.

Example question IDs with short explanations

Q55cfa205 — Marks "vitreous hemorrhage" as the primary determinant of refractive power of the eye. The cornea contributes approximately 43 D of the eye's total ~60 D refractive power. This is a complete factual inversion. Any candidate who answers correctly on this question learns the wrong fact.
Qceb77783 — Marks "primary open-angle glaucoma" as the cause of painful loss of vision. POAG is the paradigmatic painless glaucoma; acute angle-closure glaucoma causes pain. The answer key is inverted.
Q3407bd87 — States that corneal endothelium "regenerates rapidly" as the correct answer. Corneal endothelium does not regenerate; surviving cells spread and enlarge to cover defects. This is a foundational corneal physiology error.
Q9890ddbd — "Which statement about acute conjunctivitis is FALSE?" marks "vision is typically unaffected" as the correct (i.e., false) answer. Vision IS typically unaffected in acute conjunctivitis. The intended false statement is something else entirely; the answer key is wrong.
Q89106f69 — CRAO EXCEPT question marks "sudden painless loss of vision" as the exception (i.e., false feature). Sudden painless loss of vision is the hallmark presentation of CRAO. The answer key is inverted.
Qb30fe928 — Acute hemorrhagic conjunctivitis: marks "Enterovirus type 70" alone as correct. The benchmark item 106dcf8b establishes that Coxsackie A + Enterovirus 70 is the standard answer. This is a partial-answer error that would mislead candidates preparing for an exam that tests the combination.
Qf2c13883 — Marks syphilis as NOT a cause of panuveitis. Syphilis is a classic and well-recognised cause of panuveitis. The answer key appears inverted.
Q5fcfc165 — Cherry red spot question: Berlin's edema (commotio retinae) is keyed as the primary answer, and "all of the above" includes CRVO, which does not cause cherry red spot. Multiple content errors in a single item.
Q961173a3 — Ionic exchange in corneal endothelium: "activation of anaerobic glycolysis" is keyed as correct. Corneal endothelial pump depends on aerobic metabolism and Na⁺/K⁺-ATPase; the correct answer should relate to interference with aerobic metabolism or the sodium pump, not activation of anaerobic glycolysis.
Qdebeef32 — Marks "Pre-Descemet's membrane" (Dua's layer) as the toughest corneal layer. Standard teaching identifies Bowman's layer as the toughest structural layer. Pre-Descemet's is a recently described layer not conventionally described as the toughest.
Q049b294a — Early sign of papilloedema: marks "blurring of disc margin" as correct. Most authoritative sources (Kanski, AK Khurana) cite blurring of the peripapillary nerve fiber layer as the earliest sign. The answer key is likely wrong.

Recommended disposition

All items in this category: Disable immediately pending expert content review and answer correction. Items where the concept is high-yield and the question structure is otherwise sound (e.g., Q3407bd87, Q961173a3) may be Fixed after expert verification and answer key correction. Items where the error is so fundamental that the entire question premise is compromised (e.g., Q55cfa205, Qceb77783, Q5fcfc165) should be Disabled and replaced with purpose-built items. The content team should run a dedicated sweep of all negation-format questions and all corneal physiology questions for answer-key accuracy, as these two clusters show the highest error density in the reviewed set.

3. Broken Image Dependency: Non-Functional Questions in the Live Bank

Why this pattern is bad

A question that references "the image shown" or "the instrument depicted" without an attached image is not a question — it is an unanswerable prompt. If deployed to candidates, it generates guaranteed reports, erodes platform trust, and in some cases (where the image is the only route to the correct answer) makes the question unsolvable by any reasoning. This is a structural defect, not a content quality issue, but it has immediate operational consequences.

How it shows up

The pattern appears in at least five questions across the reviewed shards. In each case, the question stem explicitly references a visual exhibit ("the disease depicted in this picture," "the instrument shown," "the given ocular examination," "the fundus shown below," "the spectacle type shown") but no image is present in the question data. These questions cannot be answered from the text alone. Some were originally PYQ items that included images in the original exam context; the images were not carried over during ingestion.

This pattern is likely underrepresented in the reviewed sample because the shard analysis was text-based. The full 6,703-question pool almost certainly contains additional broken image-dependent items that were not flagged in these eight shards.

Example question IDs with short explanations

Qa090bce2 — "Select the FALSE statement pertaining to the disease depicted in this picture." No image present. The question cannot be answered from the stem alone. The topic appears to be glaucoma based on context, but the specific disease depicted is unknown.
Q2f647962 — "All are true about the instrument shown except." No image present. Likely refers to an ophthalmoscope or loupe based on context, but the instrument is unidentifiable from the text.
Q834efdd3 — AIIMS Nov 2018 PYQ referencing a spectacle type shown in an image. No image present. The question is non-functional.
Qcd9e066d — "Identify the finding on ocular examination" (descemetocele context). References "given ocular examination" with no image. Non-functional.
Q2d47d43c — "The fundus shown below indicates presence of:" (Forster-Fuchs' spots context). No fundus image present. Non-functional.

Recommended disposition

All confirmed broken image-dependent questions: Disable immediately from the live candidate pool. Do not deploy to candidates in current state. For each item, the content team should attempt to locate the original source image. If the image can be recovered and verified, the question may be restored. If the image cannot be recovered, the question should either be rewritten as a text-based clinical description (where the clinical finding can be described in words without loss of discriminatory value) or permanently retired. The content team should also conduct a systematic audit of the full 6,703-question pool for image-reference strings ("shown below," "depicted in," "given image," "instrument shown") to identify the full scope of this problem.

4. Negation-Stem Structural Weakness: A Format Overused Beyond Its Legitimate Range

Why this pattern is bad

The "all except / NOT a feature / which is FALSE" format is a legitimate item-writing tool when the exception is non-obvious, clinically meaningful, and not eliminable by common sense. It becomes a quality problem when it is used as a difficulty proxy — when the stem is negative but the correct answer is immediately apparent because the odd-one-out is either obviously unrelated to the topic or is a well-known fact that any candidate would recognise. In this mode, the negative format adds cognitive load through confusion rather than genuine clinical reasoning. It also creates a disproportionate risk of answer-key errors, because the logical inversion required to construct the question introduces an additional step where errors can enter.

How it shows up

Negation-format questions appear in every shard reviewed. Across the 200-question sample, approximately 35–40 questions use this format — roughly 18–20% of the candidate pool. This is a high proportion. The problems cluster into three sub-types:

Sub-type A: Trivially easy exceptions. The odd-one-out is immediately apparent because it belongs to a completely different clinical category (e.g., "metoprolol is NOT an antiglaucoma drug" — metoprolol is a systemic beta-blocker, not an ophthalmic preparation; any candidate who has seen a glaucoma drug list would eliminate it instantly).

Sub-type B: Answer-key errors introduced by logical inversion. As documented in Category 2 above, questions like Q9890ddbd (conjunctivitis), Q89106f69 (CRAO), and Qf2c13883 (panuveitis) all use the negation format and all have inverted answer keys. The format appears to be a contributing factor to keying errors.

Sub-type C: Ambiguous exceptions where multiple options could defensibly be the odd-one-out. Questions like Q36bc0f9b (FHI features) and Qf3cb0b49 (hemolytic glaucoma mechanism) have options where the "correct" exception is debatable, creating a defensibility risk in a competitive exam context.

Example question IDs with short explanations

Q84a30014 — "All true about chalazion EXCEPT: more common in children." The exception is a genuine clinical fact (chalazion is more common in adults), but the other options are so obviously true that the question is trivially easy. The negative format adds no value here.
Q3f4c0ee4 — Absolute angle closure features EXCEPT. The exception (phthisis bulbi) is a late-stage complication, not a feature of the acute attack — a reasonable distinction, but the other options are so obviously correct that the question resolves without reasoning.
Q90d4e4c2 — Glaucoma drugs EXCEPT. Metoprolol as the exception is immediately apparent to anyone who has seen a glaucoma drug list.
Q1c1f1d16 — Duane's retraction syndrome EXCEPT (narrowing of palpebral fissure in abduction). This is a well-constructed exception — the narrowing occurs in adduction, not abduction — and represents the legitimate use of the format. Keep.
Q7e58ce60 — NOT a feature of trachoma: ectropion. Trachoma causes entropion, not ectropion — a commonly confused point. This is also a legitimate use of the format. Keep.
Q9890ddbd, Q89106f69, Qf2c13883 — All negation-format questions with inverted answer keys (see Category 2 above for full details).

Recommended disposition

The content team should apply a two-step filter to all negation-format questions in the bank. Step 1: Is the exception non-obvious and clinically meaningful? If no, convert to a positive stem or disable if the underlying concept is already covered by a stronger item. Step 2: Is the answer key correct and unambiguous? If there is any doubt, disable pending expert review. Questions that pass both steps (like Q1c1f1d16 and Q7e58ce60) should be kept. The content team should not attempt to fix negation-format questions by adjusting distractors alone — if the exception is trivially obvious, the format itself is the problem and conversion to a positive stem is the correct remediation.

5. Clinical Vignette Inflation and Distractor Poverty: The Quality Ceiling Problem

Why this pattern is bad

The benchmark standard requires that clinical vignettes do real diagnostic or reasoning work — the scenario should contain details that are necessary to arrive at the correct answer, and the distractors should represent genuine clinical alternatives that a candidate without full knowledge might plausibly select. When a vignette is present but the answer is immediately apparent from a single detail in the stem, the question is functionally a Bloom's-1 recall item with a clinical costume. When distractors are implausible (e.g., "retinal detachment" as a mechanism of POAG), the question loses its ability to discriminate between candidates who understand the concept and those who are guessing.

How it shows up

This pattern appears in two distinct forms across the reviewed set.

Form A: Vignette inflation (cosmetic clinical framing). The question has a patient, a setting, and a presenting complaint, but the answer is determined by a single obvious detail that requires no integration. Examples: Q3dce05a3 (Air Force pilot with subconjunctival hemorrhage — the diagnosis is apparent from the single finding of subconjunctival hemorrhage with no reasoning required), Qb5d4cb05 ("when to remove IOL — never" dressed as a clinical scenario), Q60eb5b5f (anisometropic amblyopia vignette where the diagnosis is given by the refractive error alone, but the presenting complaint of "pain in the right eye" is atypical for amblyopia and misleads toward organic pathology).

Form B: Distractor poverty in otherwise sound vignettes. The vignette and correct answer are clinically appropriate, but the wrong options are implausible enough that the question can be answered by elimination without knowing the correct answer. Examples: Qe35497d5 (POAG pathway question — correct answer is retinal ganglion cell apoptosis, but distractors include "retinal detachment" and "phototransduction in rods/cones," which are not plausible mechanisms for POAG), Q3b8117f0 (acute angle closure management — "None of the options" as a distractor), Qdfef6a3b (glaucoma vignette with insufficient clinical detail to distinguish from cluster headache or migraine).

Example question IDs with short explanations

Q3dce05a3 — Air Force pilot, subconjunctival hemorrhage. Clinical wrapper adds nothing; the diagnosis is apparent from the single finding. Tagged Bloom's-3 but functionally Bloom's-1.
Qb5d4cb05 — "When to remove IOL — never." Trivially obvious to any medical student. The clinical scenario adds no reasoning requirement.
Q60eb5b5f — Anisometropic amblyopia: "pain in the right eye" is an atypical and misleading presenting complaint for amblyopia. The vignette is internally inconsistent.
Qdfef6a3b — Glaucoma vignette (35F, unilateral headache, nausea, vomiting). Lacks IOP, corneal findings, and pupil data. Cannot distinguish acute angle-closure glaucoma from cluster headache or migraine on the information given.
Qe35497d5 — POAG pathway: correct answer (retinal ganglion cell apoptosis) is sound, but distractors (retinal detachment, phototransduction) are implausible and reduce discrimination.
Qccb88a20 — Malignant change in choroidal nevus: "all of the above" as the correct answer. The individual features are all correct, but the format eliminates discrimination between candidates who know one feature and those who know all three.
Q21e687ab — Amblyopia: "all of the above" as the correct answer. Individual options are imprecisely worded, and the format is easily guessed.

Recommended disposition

Vignette inflation (Form A): Fix by either (a) adding the clinical details that make the scenario genuinely discriminating, or (b) stripping the vignette and converting to an honest Bloom's-1 recall item if the concept is high-yield enough to justify a recall question. Do not maintain the fiction of a Bloom's-3 label on a Bloom's-1 question.

Distractor poverty (Form B): Fix by replacing implausible distractors with genuine clinical alternatives. For POAG mechanism questions, appropriate distractors include decreased axonal transport, excitotoxicity from glutamate, and trabecular meshwork dysfunction. For glaucoma management questions, appropriate distractors are other antiglaucoma drug classes, not "none of the above." The content team should establish a distractor quality standard: each wrong option must represent a clinical entity or mechanism that a candidate without full knowledge might plausibly select.

"All of the above" as correct answer: Fix by converting to a specific composite stem or a "which is NOT true" format with a clearly false option. The "all of the above" format should be retired from the bank as a keyed correct answer.

6. Topic Misclassification and Wrong-Subject Contamination

Why this pattern is bad

Topic misclassification has two distinct operational consequences. First, it distorts topic-level analytics — if chalazion questions are filed under "Orbital Diseases," the Orbital Diseases topic appears to have more coverage than it does, while Diseases of the Eyelids appears under-covered. This misleads content planning decisions. Second, in the most severe cases (wrong-subject contamination), a question from a completely different specialty appears in the Ophthalmology bank, consuming a slot and potentially being served to candidates as an Ophthalmology question.

How it shows up

The reviewed set contains examples across a spectrum of severity. At the most severe end, Qde280ade is an orthopaedics question (Jersey finger, mallet finger, boxer knuckle) filed under "Diseases of the Cornea" in the Ophthalmology bank — a complete subject-level error. At the moderate end, several questions are filed under the wrong Ophthalmology subtopic: chalazion under "Orbital Diseases" (Q84a30014), subconjunctival hemorrhage under "Diseases of the Sclera" (Q3dce05a3), canaliculitis under "Ocular Manifestations of Systemic Disorders" (Qb3e5b0f4), a drug-causing-cataract question under "Contact Lens" (Q46584127), and a glutathione/lens opacity question under "Contact Lens" (Q71481f0b). The CHARGE syndrome question (Q61843495) is tagged "Neuro-Ophthalmology" when it is a syndromic/pediatric topic.

The eyelid, conjunctival, and adnexal conditions appear to be the most inconsistently classified cluster in the reviewed set, with conditions being distributed across Orbital Diseases, Diseases of the Sclera, Diseases of the Conjunctiva, and Oculoplasty depending on the question author rather than any consistent taxonomy.

Example question IDs with short explanations

Qde280ade — Jersey finger, mallet finger, boxer knuckle. Filed under "Diseases of the Cornea." Orthopaedics question. Complete subject-level error. Disable immediately.
Q84a30014 — Chalazion filed under "Orbital Diseases." Should be Diseases of the Eyelids or Oculoplasty.
Q3dce05a3 — Subconjunctival hemorrhage filed under "Diseases of the Sclera." Should be Diseases of the Conjunctiva or Ocular Trauma.
Qb3e5b0f4 — Canaliculitis (Actinomyces israelii) filed under "Ocular Manifestations of Systemic Disorders." Should be Lacrimal System or Adnexal Diseases.
Q46584127 — Dexamethasone causing posterior subcapsular cataract filed under "Contact Lens." Should be Ophthalmic Pharmacology or Diseases of the Lens.
Q71481f0b — Glutathione and lens opacity filed under "Contact Lens." Should be Diseases of the Lens or Basic Sciences.
Q61843495 — CHARGE syndrome filed under "Neuro-Ophthalmology." Should be Pediatric Ophthalmology.

Recommended disposition

Qde280ade: Disable immediately — wrong subject entirely. All other misclassified items: Fix by reclassifying to the correct topic. The content team should conduct a systematic audit of the eyelid/adnexal/conjunctival topic cluster, as this appears to be the most inconsistently classified area in the reviewed set. A topic taxonomy reference document should be established and shared with content contributors to prevent recurrence.

Prioritization

The six categories above are not equal in urgency. The following priority order is recommended for the content operations team.

Priority 1 — Immediate action required (before next deployment cycle)

Category 3 (Broken Image Dependency): All confirmed broken image-dependent questions must be disabled from the live pool immediately. They are non-functional and will generate candidate reports on first deployment. Estimated count in the full pool: unknown but likely larger than the 5 confirmed in this sample. Audit the full pool for image-reference strings.
Category 2 (Answer-Key Errors and Factual Inversions): All confirmed wrong-key items must be disabled immediately. These actively teach incorrect medicine. The 11 confirmed examples in this sample represent a minimum — the full pool likely contains more. Run a dedicated expert review pass on all negation-format questions and all corneal physiology questions as the two highest-risk clusters.
Category 6 (Wrong-Subject Contamination): Qde280ade (orthopaedics question in Ophthalmology) must be disabled immediately. Conduct a spot-check of other topic-name outliers for similar cross-subject contamination.

Priority 2 — High-value remediation (next content sprint)

Category 4 (Negation-Stem Structural Weakness): Apply the two-step filter (is the exception non-obvious? is the key correct?) to all ~35–40 negation-format questions in the candidate sample. Expected outcome: approximately 15–20 conversions to positive stems or disables, 10–15 keeps, and 5–10 fixes.
Category 5 (Clinical Vignette Inflation and Distractor Poverty): Fix the highest-priority vignette items — particularly those with Bloom's-3 labels that are functionally Bloom's-1, and those with implausible distractors in high-yield topics (Glaucoma, Retina, Pediatric Ophthalmology). Retire all "all of the above" keyed-correct items.

Priority 3 — Structural improvement (ongoing)

Category 1 (Factual Recall Overload): This is the largest category by volume but the lowest individual-item urgency — these questions are not harmful, just low-value. Disable the most trivial items (pure eponym recall, single-numerical facts, device-name lookups) and invest in converting high-yield concepts to clinical vignettes. Target: reduce Bloom's-1 proportion from ~35% to ~20% of the candidate pool, and increase Bloom's-3/4 from ~14% to ~25%.
Category 6 (Topic Misclassification, non-contamination cases): Reclassify the confirmed misclassified items and establish a topic taxonomy reference. This is low urgency for candidate experience but important for content analytics accuracy.

Example Keep / Fix / Disable Calls

The following table provides concrete disposition calls drawn from the reviewed set, organized to illustrate the reasoning applied across categories.

Question ID	Topic	Issue	Disposition	Rationale
Q3045d111	Sympathetic ophthalmia	Clean Bloom's-3 vignette, PYQ-tagged, benchmark-comparable	Keep	Penetrating injury + fellow eye inflammation 3 weeks later. Requires integration of timeline and mechanism. Directly comparable to benchmark `56ec5ba8`.
Q53659763	Retina — CSME	Current evidence-based management, Bloom's-4, NEET-PG tagged	Keep	Intravitreal anti-VEGF as treatment of choice for CSME reflects current practice shift from focal laser. Applied knowledge item.
Q1c14ed73	Glaucoma — traumatic	Bloom's-3 clinical vignette, hyphema context	Keep	Well-constructed clinical scenario with relevant distractors. Suitable for INI-CET level.
Qd66b0312	Lens — Fincham's test	Specific clinical sign, Bloom's-2/3, good distractors	Keep	Tests ability to distinguish cataract from glaucoma using a specific clinical test. High-yield. Minor difficulty re-tagging needed.
Q4ce14504	Neuro-ophthalmology	Visual pathway localisation, Bloom's-2	Keep	Superior quadrantic hemianopia → temporal lobe. Tests applied anatomy. Clean distractors.
Q17d59ede	Oculoplasty — ptosis	Fasanella-Servat in Horner's, Bloom's-3, NEET-PG tagged	Keep	Clinically specific, tests pharmacology-surgery interface. High-yield.
Q1c1f1d16	Strabismus — Duane's	Negation format used correctly	Keep	Exception (palpebral fissure narrowing in abduction vs. adduction) is non-obvious and clinically meaningful. Legitimate use of format.
Q55cfa205	Optics	Vitreous hemorrhage as primary refractive element — factually wrong	Disable	Complete factual inversion. Cornea is the primary refractive element. Actively misleads candidates.
Qceb77783	Glaucoma	POAG as cause of painful vision loss — factually wrong	Disable	POAG is painless. Answer key inverted.
Q3407bd87	Cornea	Corneal endothelium "regenerates rapidly" — factually wrong	Disable	Corneal endothelium does not regenerate. Fix only after expert correction of answer key and explanation.
Q5fcfc165	Retina	Cherry red spot keyed to Berlin's edema; CRVO listed as cause	Disable	Multiple content errors. CRAO is the classic cause. CRVO does not cause cherry red spot.
Qde280ade	Wrong subject	Orthopaedics question (Jersey finger) in Ophthalmology bank	Disable	Complete subject-level contamination. No ophthalmology content.
Qa090bce2	Glaucoma	Image-dependent, no image present	Disable	Non-functional. Cannot be answered from text alone. Restore image or rewrite.
Q2d47d43c	Retina	Fundus image reference, no image present	Disable	Non-functional. Forster-Fuchs' spots cannot be identified without fundus image.
Q529b02a8	Lens	Snow-storm cataract eponym, Bloom's-1, no clinical context	Disable	Pure eponym recall. Covered by stronger items. No discriminatory value at PG level.
Q3a92ca1b	Vitreous	Synchysis definition, Bloom's-1	Disable	Single-term definition. No clinical application.
Q41b52f90	Community ophthalmology	NPCB administrative trivia	Disable	Low-yield administrative fact with no clinical application.
Qb30fe928	Conjunctiva	Acute hemorrhagic conjunctivitis — partial answer (EV70 only)	Fix	Revise correct option to include Coxsackie A + Enterovirus 70, consistent with benchmark `106dcf8b`.
Q9890ddbd	Conjunctiva	"Vision unaffected in conjunctivitis" marked as FALSE — wrong key	Fix	Vision IS typically unaffected. Review and correct answer key; identify the intended false statement.
Q89106f69	Retina — CRAO	"Sudden painless loss of vision" marked as EXCEPT — wrong key	Fix	This IS a hallmark of CRAO. Full option revision required before use.
Qdfef6a3b	Glaucoma	Vignette too sparse to distinguish from cluster headache	Fix	Add IOP, corneal haziness, mid-dilated pupil, periocular redness to make acute angle-closure the unambiguous answer.
Q60eb5b5f	Pediatric — amblyopia	"Pain in right eye" as presenting complaint — atypical for amblyopia	Fix	Amblyopia is typically asymptomatic. Change presenting complaint to reduced vision or failed school screening.
Qe35497d5	Glaucoma — POAG	Correct answer sound; distractors implausible	Fix	Replace "retinal detachment" and "phototransduction" with decreased axonal transport, excitotoxicity, or trabecular dysfunction.
Q46584127	Pharmacology/Lens	Dexamethasone → cataract filed under "Contact Lens"	Fix	Reclassify to Ophthalmic Pharmacology or Diseases of the Lens. Content is correct.
Qeda07b12	Optics — retinoscopy	Calculation question with incomplete stem	Fix	Add neutralisation finding and working distance to make the calculation unambiguous.
Q84a30014	Eyelid	Chalazion filed under "Orbital Diseases"	Fix	Reclassify to Diseases of the Eyelids. Also verify "more common in children" claim — chalazion is more common in adults.
Q961173a3	Cornea	Ionic exchange — anaerobic glycolysis keyed as correct	Fix	Expert review required. Correct answer should relate to aerobic metabolism/Na⁺K⁺-ATPase interference.
Q783000a4	Retina	Cherry red spot — "all of the above" as correct answer	Fix / Disable	Replace with clinical vignette distinguishing CRAO from storage disorders. If rewrite is not feasible, disable.
Q21e687ab	Pediatric — amblyopia	"All of the above" as correct answer, imprecise options	Fix	Replace with specific composite stem or "which is NOT true" format with clearly false option.