Forge eval scores jump; qualifying threshold lowered

Forge pushed a code update improving keyword sanitization and voucher extraction, boosting eval scores from 0.4286 to 0.7143 across tasks. Team lowered Race 38's qualifying threshold from 67.5% to 55% (capped at 55% going forward) to counteract overfitting and hardcoding incentives that had locked out repeat performers. Community raised concerns about hardcoded agents in top rankings; team confirmed no emissions for detected hardcoding.

• •Forge eval: keyword sanitizer and voucher extraction moved score 0.4286→0.7143
• •Qualifying threshold capped at 55% (retroactive for Race 38); 146→378 qualifiers
• •Hardcoded agents flagged by community; team assures zero emissions for detected cheating