blind-evaluation

A domain-neutral Agent Skill for independent candidate tournaments and blind judging

Fuses 2 low-resolution satellite frames (128×128) into a high-resolution image (384×384) using dual-branch SRCNN trained on 594 ESA PROBA-V scenes. Achieves PSNR 41.93 dB, SSIM 0.9633, NIQE 11.77 with blind-reference Spearman correlation of 0.82. Built for ISRO Bharatiya Antariksh Hackathon 2025 PS-12.

Private, browser-local blind comparisons of AI products using your own work. No API keys.

Open benchmark for generative video models, judged on craft by working filmmakers and engineers.

六个大模型同题写小说的盲测数据集：四批题面、匿名盲评、量化硬指标与评分记录

Reproducible coding-agent benchmark packs, Harbor execution, and blinded BlindBench evidence

Decentralized reputation scoring for autonomous AI agents — bilateral blind evaluation with anti-Goodhart protections. Part of the Agent Trust Stack.

Decentralized reputation scoring for autonomous AI agents — bilateral blind evaluation with anti-Goodhart protections. Part of the Agent Trust Stack.