ai-sherpa/rank-learner-orthogonal-treatment-ranking-repro

C5 third pass: MIMIC-III measured on the open ODbL demo; all three Table-4 benchmarks now at n=1,000

C5 second pass: results/rework_c5_v2.json

C5 second pass: code/bundle_ec2_c5v2.py

C5 second pass: code/rework_c5_v2.py

C5 second pass: logbook.json

C5 second pass: pages/conclusion/page.md

C5 second pass: pages/failure-boundaries/page.md

C5 second pass: pages/00-scorecard/page.md

C5 second pass: pages/claim-5-semi-synthetic-benchmarks/page.md

Add anchored claims C4 (synthetic AUTOC sweep, 5 sizes x 20 seeds), C5 (semi-synthetic MovieLens + CPS at n=1,000; MIMIC-III blocked by PhysioNet credentialing, reported as partial coverage) and C6 (fixed-pool pair-subsampling saturation). Claims 1-3 pages unchanged.

Update logbook: Repro — Rank-Learner: Orthogonal Ranking of Treatment Effects

initial commit