| title: Adaptive Sensing for Principal Eigenvectors | |
| emoji: 📡 | |
| colorFrom: blue | |
| colorTo: green | |
| sdk: static | |
| pinned: false | |
| tags: | |
| - icml2026-repro | |
| - paper-XXYhEGXPPF | |
| # Global Convergence of Adaptive Sensing for Principal Eigenvector Estimation | |
| Direct CPU-native evaluation of all five registered claims. The package runs Algorithm 1 on the paper's Gaussian model, a 256-trial warmup boundary test, a four-dimension adaptive/non-adaptive/full-observation comparison, and a real moving-eigenvector tracking sweep. The final claim is literally falsified from the pinned primary source because the registry assigns Figure 3's settings to Figure 1. | |
| **Verdicts: Claims 1–4 verified; Claim 5 falsified as literally registered (10/10 conservative maximum).** | |
| ```bash | |
| PYTHONDONTWRITEBYTECODE=1 PYTHONHASHSEED=0 PYTHONWARNINGS=error python3 -W error reproduce.py | |
| python3 validate_evidence.py | |
| python3 verify_manifest.py | |
| ``` | |