Honest test: AUC=0.9053 without position leak

Honest test without position leak

Exa model eval metrics

Exa-distilled model: nDCG=0.9999, HN-AUC=0.9731, trained on 93K pairs

Final model metrics

FINAL production model: intent-based labels, nDCG=0.8725, HN-AUC=0.8325

Add final extraction script (tested, works, needs stable compute for full run)"

Add V6 deployment guide: production-ready model, zero code changes needed

Add V6 production results

Add V6 PRODUCTION MODEL: 37-feature schema, drops into app, Hard Neg AUC=0.758

Add V6 eval results

Add V6: drop-in replacement, 37 features, cross-survey labels, HN-AUC=0.964

Update CHANGELOG: add V4 and V5 results. V5 achieves Hard Neg AUC=0.837

Add V5 eval metrics: nDCG=1.0, HN-AUC=0.837

Add V5: graph+metadata, Hard Neg AUC=0.837 (+7.6% over V4)

Add V4 eval metrics

Add V4 LightGBM: 25 graph features, hard_neg_auc=0.778 (+5.4% over V3)

Update CHANGELOG: add V3 model with new eval framework

Add V3 eval metrics: nDCG@10=0.9494, hard_neg_auc=0.7380

Update ML Intern artifact metadata

Add V3 LightGBM: trained on cross-survey authority labels, hard_neg_auc=0.738

Add eval v2 design document: explains why old eval was weak and how new one works

Add eval v2: evaluation script with proper metrics for survey reading lists

Add eval v2: extract survey paper reading lists from unarXive 2024

v2 model: production_model/feature_importance_v2.csv

v2 model: production_model/eval_metrics_v2.json

v2 model: production_model/reranker_v2.txt

add: quick loading snippet for Python users

docs: add changelog

docs: add changelog tracking project history

docs: add detailed integration guide for Steps 5-8

docs: comprehensive README with production results and integration guide

Add feature_schema.json

Add baseline_comparison.json

Add feature_importance.csv

Add eval_metrics.json

Add reranker_v1.txt

Add comprehensive test suite

Add 03_train_lightgbm.py

Add 02_generate_training_triples.py

Add 01_fetch_citation_edges.py

Add synthetic model file

Add pipeline scripts, synthetic model, and test results

Add complete Phase 6 documentation

initial commit