blind-evaluation

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.

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

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

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

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

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

Choose an AI subscription using your own work — private blind comparisons, no API keys, browser-local history, and rigorous study mode.