Parallel eval and scoring methodology changes coming

Trajectory-RL is shipping parallel evaluation this week, reducing eval time from ~60 minutes to 30 minutes by running 2 scenarios per miner concurrently. The team also confirmed they will switch from stake-weighted to average scoring from qualified validators in the next release to reduce variance. Current top miner sits at 8.21 out of 8.6 maximum possible score.

• •Parallel eval rolling out this week; cuts eval time 50% initially
• •Next release: switch to average validator scores, dropping stake weighting
• •Eval time now ~30 minutes per full evaluation cycle
• •Three scenarios have hard caps: path-tracing, swe-bench astropy2 (0.8 each), break-filter-js-from-html unsolvable