Affine evaluation queue bottleneck sparks system overhaul debate

Model evaluation backlogs exceeding 6 days have triggered intense community pressure to either revert the current queue-based evaluation system or implement architectural fixes. Team acknowledged GPU constraints and committed to discussing evaluation speed improvements. Core tensions: current mechanism favors incumbent champions through spam-submission strategies, while long evaluation cycles discourage genuine model training. Proposed solutions include sequential environment evaluation, increased registration costs, and higher burn decay rates to reduce queue volume.

• •6+ day model evaluation delays drive calls for system rollback or redesign
• •Team acknowledges GPU shortage, commits to evaluation speed discussion
• •Miners propose sequential environment evaluation to reduce wasted compute
• •Registration cost cooldown increase (20 mins → 1-2 hours) suggested to limit spam
• •Incumbent miners allegedly flooding queue with bulk submissions; real trainers reportedly leaving