Recommended CAT Predicted Papers
FINAL_RECOMMENDED_CAT_PREDICTED_PAPERS.pdf
20 KB / free
Balanced Paper PDF
final_predicted_paper_A_balanced.pdf
73 KB / premium
Arithmetic-heavy Paper PDF
final_predicted_paper_B_arithmetic_heavy.pdf
71 KB / premium
Reasoning-heavy Paper PDF
final_predicted_paper_C_reasoning_heavy.pdf
75 KB / premium
Recent-trend Paper PDF
final_predicted_paper_D_recent_trend.pdf
75 KB / premium
Wildcard Paper PDF
final_predicted_paper_E_wildcard.pdf
76 KB / premium
Final recommendation summary
Final Recommended CAT Predicted Papers
Disclaimer: Pattern-based prediction only. This is not a leaked or official paper.
Methodology
PYQ analysis is loaded into SQLite, converted into prediction specs, used to create internal candidate prompts/imports, verified, scored for CAT-likeness, filtered for weakness/duplicates, assembled into variants, backtested, and weight-adjusted.
Selected Papers
| Variant | Type | Expected Overlap | Risk | Diversity | Reason |
|---|---|---|---|---|---|
| Arithmetic-Heavy / QA Expected Paper | arithmetic_heavy | 0.865 | 0.35 | 0.636 | Selected as one of the best-scoring non-wildcard variants. |
| Reasoning-Heavy Paper | reasoning_heavy | 0.858 | 0.35 | 0.727 | Selected as one of the best-scoring non-wildcard variants. |
| Balanced High-Probability Paper | balanced | 0.849 | 0.35 | 0.864 | Selected as one of the best-scoring non-wildcard variants. |
| Wildcard Paper | wildcard | 0.817 | 0.65 | 0.636 | Included as a volatility hedge. |
Candidate Content Status
- Selected candidate pool: 65
- Selected paper-eligible candidates: 37
- Selected LaTeX candidates: 26
- Specs represented: 19
- Final assembly readiness override: ALLOWED WITH RISK FLAGS.
Coverage Risk
CAT_QA_SPEC_14 Circles remains under-covered; final paper variants should not force a weak circle candidate unless required by portfolio diversity.
No rejected Circles candidate was promoted during assembly.
Limitations
- Exact CAT question prediction is not realistic.
- VARC/DILR extraction and tags are rule-based.
- Candidate quality depends on manual LLM/local-model generation and verification.
- Final output should stay limited and high-precision, not become a bulk question bank.
Full pipeline summary
Full Stage 9 Prediction Portfolio Pipeline Summary
What Data Was Used
- Papers loaded: 66
- PYQ QA questions loaded: 540
- Section pattern rows loaded: 750
How Specs Were Created
- Prediction specs created: 37
- QA specs use Stage 4 subtopic/archetype signals.
- DILR specs are set-level, not isolated-question specs.
- VARC specs are passage/question-intention specs.
How Candidates Were Generated
- Candidate rows currently in database: 178
- Without an API/local model, Stage 9 writes prompt batches and a manual import template.
- Candidate generation is internal only; final user-facing output remains limited.
How Verification And Scoring Worked
- Candidate score rows: 178
- Verification checks missing answer, missing solution, incomplete MCQ options, direct formula wording, and basic structure.
- Scoring combines prediction score, evidence strength, CAT-likeness, depth, difficulty match, trap quality, solution clarity, coverage gain, and redundancy.
How Variants Were Assembled
- Variants created: 5
- Candidate placements: 110
- Authoritative selected pool: 65
- Selected from LaTeX pool: 26
- Selected paper-eligible candidates: 37
- Specs represented: 19
- Assembly readiness override: ALLOWED WITH RISK FLAGS
Coverage Risk
CAT_QA_SPEC_14 Circles remains under-covered; final paper variants should not force a weak circle candidate unless required by portfolio diversity.
Rejected Circles candidates were not reclassified or forced into the assembled variants.
How Backtesting Adjusted Weights
- Backtest rows: 20
- Backtest scores adjust final portfolio weights, especially frequency versus CAT-likeness and diversity.
What The Final Output Means
The Stage 9 output is a high-precision portfolio engine scaffold. It is ready to accept manually generated or locally generated candidate questions, verify and score them, remove weak/duplicate rows, assemble variants, backtest, adjust weights, and export final recommended papers.