Godot SFT: the finalized numbers

The Qwen3.8-27B (SFT) Godot group was rerun and settled. The hero stat goes
57.3 -> 61.4, the abstract sentence follows the paper, and the three table
rows carry the final scores and their much smaller token ledgers.

index.html

@@ -51,7 +51,7 @@
   <div class="wrap">
     <div class="stats" id="hero-stats">
       <div class="stat"><b>+14.3</b><span>Overall on Codex&nbsp;+&nbsp;GPT-5.5 · 50.3&nbsp;→&nbsp;64.5</span></div>
-      <div class="stat"><b>57.3</b><span>Qwen3.8-27B games after experience internalization and RSIGame development — past one-shot GPT-5.5 (50.3)</span></div>
+      <div class="stat"><b>61.4</b><span>Qwen3.8-27B games after experience internalization and RSIGame development — past one-shot GPT-5.5 (50.3)</span></div>
       <div class="stat"><b>1,998</b><span>runnable agent-written games released</span></div>
       <div class="stat"><b>222K</b><span>verified agentic training rows</span></div>
     </div>
@@ -69,7 +69,7 @@
         <h2>Abstract</h2>
         <p>Recent advances in large language models have made automatic game generation increasingly feasible, yet reliably improving generated games beyond a playable version remains challenging. Naive iterative refinement can easily overfit a small set of test cases, producing fragile games with unresolved bugs, missing behaviors, and poor generalization to broader player interactions.</p>
         <p>We introduce <strong>RSIGame</strong>, an autonomous agentic game development framework with recursive self-improvement. RSIGame organizes development into complementary local and global loops. Concretely, a local <em>explore-diagnose-improve</em> loop broadly explores the executable game, diagnoses and prioritizes discovered issues, and performs evidence-grounded revision, where an evolving checklist continually accumulates new testing and improvement guidance. A global loop tracks overall quality, preserves the best checkpoint, and detects saturation or regression over long-horizon development. Beyond test-time improvement, RSIGame further internalizes successful development experience into the generator through training.</p>
-        <p>Across 140 GameCraft-Bench tasks, two game engines, and five generators, RSIGame consistently improves game quality under matched development budgets. Notably, iterative development with GLM-5.3-Flash and Qwen3.8-27B, with its experience internalized, reaches 57.30 on Godot and 58.53 on Phaser, surpassing substantially stronger GPT-5.5-based generators.</p></div>
+        <p>Across 140 GameCraft-Bench tasks, two game engines, and five generators, RSIGame consistently improves game quality under matched development budgets. Notably, iterative development with GLM-5.3-Flash and Qwen3.8-27B, with its experience internalized, reaches 61.38 on Godot and 58.53 on Phaser, surpassing substantially stronger GPT-5.5-based generators.</p></div>
       <aside class="loopcard">
         <h3>How a game develops</h3>
         <section class="loopbox inner">