ai-sherpa/nash-welfare-linear-bandits-repro

Retrofit to leader 5-claim decomposition: add claim-4 (p-means interpolation/generalization) + claim-5 (self-normalized concentration + elliptical-potential proof mechanisms), both verified $0/CPU

Rework Claim 3 (5WVIbxWqwA): real MSLR-WEB10K vs from-scratch LinNash -- logbook.json

Rework Claim 3 (5WVIbxWqwA): real MSLR-WEB10K vs from-scratch LinNash -- pages/index.md

Rework Claim 3 (5WVIbxWqwA): real MSLR-WEB10K vs from-scratch LinNash -- pages/conclusion/page.md

Rework Claim 3 (5WVIbxWqwA): real MSLR-WEB10K vs from-scratch LinNash -- pages/failure-boundaries/page.md

Rework Claim 3 (5WVIbxWqwA): real MSLR-WEB10K vs from-scratch LinNash -- pages/methods-provenance/page.md

Rework Claim 3 (5WVIbxWqwA): real MSLR-WEB10K vs from-scratch LinNash -- pages/claim-3-real-world-outperformance/page.md

Rework Claim 3 (5WVIbxWqwA): real MSLR-WEB10K vs from-scratch LinNash -- pages/00-scorecard/page.md

Update logbook: Repro - Improved Algorithms for Nash Welfare in Linear Bandits

initial commit