Home / Synthesis

Five-Subject Pilot Synthesis

Status Note

This report is a pilot-only synthesis built from the first five manually reviewed subjects and from the earlier smaller candidate packets. It should be treated as the strongest current conceptual taxonomy document, but not as a full-bank conclusion. The separate full-scope synthesis on the site covers all indexed subjects at the stats layer and is broader but less evidentially deep.

Subjects reviewed

  • Anatomy
  • Physiology
  • Pathology
  • Pharmacology
  • Microbiology

Each subject report was built from fresh packets after filtering candidates to:

  • validation_status = validated
  • exclude recent PYQs
  • exclude INI-CET April 2026 Benchmark Test questions

Cross-subject patterns

1. The dominant problem is not factual incorrectness

Across all five subjects, the core issue is concept selection and question design. Many questions are factually correct, but they are still weak because they test something too small, too trivial, too repetitive, or too poorly aligned with the exam you want to serve.

2. The bank repeatedly collapses toward one-step recall

This appears in different forms by subject:

  • Anatomy: structures, boundaries, supply, "except" stems
  • Physiology: nuclei, ranges, elementary molecular biology
  • Pathology: one-line associations and pathology labels
  • Pharmacology: class labels, adverse effects, drug of choice
  • Microbiology: vector, organism, antigen, family, species trivia

This is the single most consistent disable-worthy category.

3. Pseudo-clinical formatting is common

Several subjects contain questions that start with a clinical story but still ask for a single memorized fact. The vignette creates the appearance of sophistication without increasing reasoning depth. This is especially visible in Anatomy, Physiology, Pharmacology, and parts of Microbiology.

4. Subject drift is real

The pilot shows multiple examples where content is technically adjacent but feels misplaced:

  • Anatomy has comparative-anatomy and non-anatomy spillover
  • Physiology has research-style molecular biology drift
  • Pathology has strong forensic spillover
  • Pharmacology includes regulatory/FDA approval trivia
  • Microbiology contains public-health/current-affairs wrappers over simple recall

5. Recent PYQs are useful but not automatically a perfect gold standard

Every subject surfaced some easy PYQ/benchmark examples. This means "appeared before" is not enough as a retention criterion. Future curation should compare candidate questions against the best gold examples, not against the entire gold bucket indiscriminately.

Subject-by-subject headline

  • Anatomy: too much low-yield trivia and comparative-anatomy noise
  • Physiology: split between trivial recall and over-generated molecular biomedical content
  • Pathology: forensic spillover plus one-line association questions
  • Pharmacology: flashcard-style drug trivia is the main problem
  • Microbiology: strongest good examples, but diluted by species/vector/antigen trivia

Recommended curation principle emerging from the pilot

For the first cleanup pass, the highest-confidence disable buckets appear to be:

  • direct recall at Bloom's 1 with no meaningful reasoning
  • negative-stem or "except" questions where the underlying concept is already weak
  • topic-drift questions that do not feel central to the subject
  • repetitive drug / organism / association flashcards

The best fix candidates are:

  • questions built around an important concept but executed too simply
  • clinical vignettes where the concept is worthwhile but the options or framing are weak
  • asset-dependent spotters that could be rehabilitated with clearer media and a better ask

Most important takeaway

The problem is not just that the bank has "bad questions." It has recognizable families of bad questions. That is good news operationally, because it means the cleanup can be policy-driven instead of purely anecdotal. The first production curation pass should aggressively target trivial recall and subject-drift categories before worrying about subtler improvements.