Connito AI Fixes Eval Seed Prediction, Updates Weight Mechanics

Connito shipped a fix to prevent miners from predicting evaluation data slices in advance by incorporating block hashes into seed generation. The team also published a detailed FAQ clarifying the 8-cycle rolling average weight allocation system, score calculation, and validator evaluation mechanics. An overfitting issue was resolved, though validator loss metrics still require attention.

• •Eval seed now unpredictable; block hashes prevent advance data computation
• •Weight allocation: top 3 miners get 98%, next 15 get 2% per cycle
• •Score: ranked by delta loss (top-1 = 2.25, top-2 = 1.5, top-3 = 1.0)
• •Expect 2-3 cycles of ranking volatility during consensus reconvergence