Reported evidence
Designer-RSI → changes → GenEval2 execution success rate
The abstract reports a rise in GenEval2 execution success from 72.7% to 99.3% on Claude Sonnet 4 under the complete Designer-RSI setup. This is an experimental result, not an isolated causal estimate for memory expansion, revision, or replay checks, and not a production acceptance rate.
Read the evidence
The abstract reports that, over five rounds involving 1,406 real user briefs and 1,869 automatically graded execution trajectories, the system increased its skill count from 76 to 139 without weight updates or human labels. GenEval2 execution success on Claude-Sonnet-4 rose from 72.7% to 99.3%, and generation quality increased by 11.99 points.[S4]Read the full research