Fix duplicate WebSocket log output; update benchmark numbers to latest run (avg 75.9%, ~10min)

README.md

@@ -72,7 +72,7 @@ export MODEL_NAME="meta-llama/Llama-3.1-8B-Instruct"
 export HF_TOKEN="hf_your_token_here"
 export ENV_BASE_URL="https://therubberduckdebuggers-cascade-containment.hf.space"
 
-# Full LLM+GRPO evaluation (~20 minutes, emits [START][STEP][END] logs)
+# Full LLM+GRPO evaluation (~10 minutes, emits [START][STEP][END] logs)
 python inference.py
 
 # Local validation + greedy benchmark
@@ -214,9 +214,9 @@ Results from `baseline/run.py` (Llama 3.3 70B via Groq, runtime ~19.8 minutes):
 | Task | Greedy (D0) | LLM+GRPO | Lift |
 | --- | --- | --- | --- |
 | Easy | 42.8% (breach 60%) | **90.8%** | +48pp |
-| Medium | 42.7% (breach 80%) | **71.6%** | +29pp |
-| Hard | 33.0% (breach 100%) | **66.1%** | +33pp |
-| **Average** | **39.5%** | **76.2%** | **+37pp** |
+| Medium | 42.7% (breach 80%) | **78.0%** | +35pp |
+| Hard | 33.0% (breach 100%) | **61.1%** | +28pp |
+| **Average** | **39.5%** | **75.9%** | **+36pp** |
 
 The greedy baseline (always allocates to D0) scores 33–43% with 60–100% hospital breach rates — no trivial exploit path. LLM+GRPO reaches 66–91% with zero breaches, demonstrating that genuine triage reasoning is required and rewarded.
 

inference.py

@@ -73,6 +73,8 @@ def run_rollout(
     rewards      = []
 
     log_start(task=f"{task_name}-r{rollout_idx}", env=BENCHMARK, model=MODEL_NAME)
+    seen_steps: set = set()  # deduplicate WebSocket replay artefacts
+    end_logged: bool = False
 
     try:
         while not done:
@@ -85,7 +87,9 @@ def run_rollout(
                 result = env.step(action)
             except Exception as e:
                 log_step(step=step + 1, action=action_str, reward=0.0, done=True, error=str(e)[:80])
-                log_end(success=False, steps=step, score=0.0, rewards=rewards)
+                if not end_logged:
+                    end_logged = True
+                    log_end(success=False, steps=step, score=0.0, rewards=rewards)
                 return total_reward, step, trajectory, 0.0
 
             next_obs      = result.observation
@@ -96,7 +100,11 @@ def run_rollout(
 
             rewards.append(reward)
             trajectory.append({"obs": obs, "action": action, "reward": reward})
-            log_step(step=step, action=action_str, reward=reward, done=done, error=None)
+
+            # Only log each step number once — WebSocket can replay buffered responses
+            if step not in seen_steps:
+                seen_steps.add(step)
+                log_step(step=step, action=action_str, reward=reward, done=done, error=None)
 
             obs = next_obs
             if done:
@@ -114,10 +122,14 @@ def run_rollout(
         success = score >= 0.40
 
     except Exception:
-        log_end(success=False, steps=step, score=0.0, rewards=rewards)
+        if not end_logged:
+            end_logged = True
+            log_end(success=False, steps=step, score=0.0, rewards=rewards)
         return total_reward, step, trajectory, 0.0
 
-    log_end(success=success, steps=step, score=score, rewards=rewards)
+    if not end_logged:
+        end_logged = True
+        log_end(success=success, steps=step, score=score, rewards=rewards)
     return total_reward, step, trajectory, score
 
 

scripts/test_local.py

@@ -11,8 +11,8 @@ from server.grader import grade_trajectory
 # LLM+GRPO reference scores — update this dict after each baseline/run.py session.
 GRPO_SCORES = {
     "easy":   {"score": 0.8848, "containment": 1.000, "hospital": 0.996, "efficiency": 0.900},
-    "medium": {"score": 0.7539, "containment": 0.469, "hospital": 0.978, "efficiency": 0.987},
-    "hard":   {"score": 0.6311, "containment": 0.462, "hospital": 0.952, "efficiency": 0.533},
+    "medium": {"score": 0.7798, "containment": 0.533, "hospital": 0.974, "efficiency": 0.960},
+    "hard":   {"score": 0.6110, "containment": 0.420, "hospital": 0.940, "efficiency": 0.520},
 }
 
 

server/app.py

@@ -525,7 +525,7 @@ body{font-family:var(--mono);background:var(--bg);color:var(--text);min-height:1
     <div class="brand-icon">🦠</div>
     <div>
       <div class="brand-name">Cascade Containment</div>
-      <div class="brand-sub">OpenEnv Benchmark &middot; Meta PyTorch Hackathon &times; SST 2026 &middot; LLM avg 75.7%</div>
+      <div class="brand-sub">OpenEnv Benchmark &middot; Meta PyTorch Hackathon &times; SST 2026 &middot; LLM avg 75.9%</div>
     </div>
   </div>
   <div class="header-right">
@@ -568,7 +568,7 @@ body{font-family:var(--mono);background:var(--bg);color:var(--text);min-height:1
     </div>
     <div class="hero-stats">
       <div class="hs-item"><div class="hs-val">3</div><div class="hs-label">Task levels</div></div>
-      <div class="hs-item"><div class="hs-val">75.7%</div><div class="hs-label">LLM+GRPO avg</div></div>
+      <div class="hs-item"><div class="hs-val">75.9%</div><div class="hs-label">LLM+GRPO avg</div></div>
       <div class="hs-item"><div class="hs-val">+36pp</div><div class="hs-label">vs greedy</div></div>
     </div>
   </div>
@@ -766,7 +766,8 @@ body{font-family:var(--mono);background:var(--bg);color:var(--text);min-height:1
         Llama 3.3 70B via Groq, with GRPO-style episodic memory across 2&ndash;3 rollouts per task.
         Each rollout injects advantage-gated memory from prior rollouts into the prompt.
         Reproduced by running <code style="color:var(--text);">python baseline/run.py</code>.
-        Total runtime: <strong style="color:var(--text);">19.8 minutes</strong>.
+        Llama 3.3 70B via Groq, with GRPO-style episodic memory across 2&ndash;3 rollouts per task.
+        Total runtime: <strong style="color:var(--text);">~10 minutes</strong>.
       </div>
       <div class="grid-3" style="margin-bottom:1rem;">
         <div class="card-sm" style="border-top:2px solid var(--green);">
@@ -781,7 +782,7 @@ body{font-family:var(--mono);background:var(--bg);color:var(--text);min-height:1
         </div>
         <div class="card-sm" style="border-top:2px solid var(--amber);">
           <div style="font-size:0.62rem;letter-spacing:0.08em;text-transform:uppercase;color:var(--amber);margin-bottom:0.5rem;">Medium &middot; 3 rollouts</div>
-          <div style="font-family:var(--serif);font-size:2rem;font-weight:700;color:#fff;line-height:1;">75.4%</div>
+          <div style="font-family:var(--serif);font-size:2rem;font-weight:700;color:#fff;line-height:1;">78.0%</div>
           <div style="font-size:0.68rem;color:var(--muted);margin-top:0.25rem;">Best of 4 &middot; no breach</div>
           <div style="margin-top:0.75rem;display:flex;flex-direction:column;gap:0.3rem;">
             <div style="display:flex;justify-content:space-between;font-size:0.68rem;"><span style="color:var(--muted);">Containment</span><span>42%</span></div>
@@ -791,7 +792,7 @@ body{font-family:var(--mono);background:var(--bg);color:var(--text);min-height:1
         </div>
         <div class="card-sm" style="border-top:2px solid var(--red);">
           <div style="font-size:0.62rem;letter-spacing:0.08em;text-transform:uppercase;color:var(--red);margin-bottom:0.5rem;">Hard &middot; 3 rollouts</div>
-          <div style="font-family:var(--serif);font-size:2rem;font-weight:700;color:#fff;line-height:1;">63.1%</div>
+          <div style="font-family:var(--serif);font-size:2rem;font-weight:700;color:#fff;line-height:1;">61.1%</div>
           <div style="font-size:0.68rem;color:var(--muted);margin-top:0.25rem;">Best of 4 &middot; no breach</div>
           <div style="margin-top:0.75rem;display:flex;flex-direction:column;gap:0.3rem;">
             <div style="display:flex;justify-content:space-between;font-size:0.68rem;"><span style="color:var(--muted);">Containment</span><span>51%</span></div>
@@ -811,13 +812,13 @@ body{font-family:var(--mono);background:var(--bg);color:var(--text);min-height:1
           </tr>
           <tr>
             <td><span class="badge badge-amber">Medium</span></td>
-            <td>59.6%</td><td>66.2%</td><td>75.4%</td>
-            <td><strong style="color:var(--amber);">75.4%</strong></td>
+            <td>56.3%</td><td>78.0%</td><td>64.0%</td>
+            <td><strong style="color:var(--amber);">78.0%</strong></td>
           </tr>
           <tr>
             <td><span class="badge badge-red">Hard</span></td>
-            <td>63.1%</td><td>53.9%</td><td>59.0%</td>
-            <td><strong style="color:var(--red);">63.1%</strong></td>
+            <td>61.1%</td><td>57.7%</td><td>60.3%</td>
+            <td><strong style="color:var(--red);">61.1%</strong></td>
           </tr>
         </table>
       </div>
@@ -848,21 +849,21 @@ body{font-family:var(--mono);background:var(--bg);color:var(--text);min-height:1
           </tr>
           <tr>
             <td><span class="badge badge-amber">Medium</span></td>
-            <td style="color:var(--muted);">42.7%</td><td><strong>75.4%</strong></td>
-            <td class="pos">+33pp</td>
+            <td style="color:var(--muted);">42.7%</td><td><strong>78.0%</strong></td>
+            <td class="pos">+35pp</td>
             <td><span class="badge badge-green">Strong</span></td>
             <td><span class="badge badge-green">None &mdash; 8% eff, 80% breach</span></td>
           </tr>
           <tr>
             <td><span class="badge badge-red">Hard</span></td>
-            <td style="color:var(--muted);">33.0%</td><td><strong>63.1%</strong></td>
-            <td class="pos">+33pp</td>
+            <td style="color:var(--muted);">33.0%</td><td><strong>61.1%</strong></td>
+            <td class="pos">+28pp</td>
             <td><span class="badge badge-green">Strong</span></td>
             <td><span class="badge badge-green">None &mdash; 0% eff, 100% breach</span></td>
           </tr>
           <tr style="border-top:1px solid var(--border2);">
             <td><strong>Average</strong></td>
-            <td style="color:var(--muted);">39.5%</td><td><strong>75.7%</strong></td>
+            <td style="color:var(--muted);">39.5%</td><td><strong>75.9%</strong></td>
             <td class="pos"><strong>+36pp</strong></td>
             <td><span class="badge badge-green">Strong</span></td>
             <td><span class="badge badge-green">No exploits found</span></td>
@@ -883,7 +884,7 @@ body{font-family:var(--mono);background:var(--bg);color:var(--text);min-height:1
           <div style="display:flex;flex-direction:column;gap:0.45rem;">
             <div><div style="display:flex;justify-content:space-between;font-size:0.68rem;margin-bottom:0.2rem;"><span style="color:var(--muted);">Greedy (D0)</span><span>43%</span></div><div class="bar-track" style="height:8px;"><div class="bar-fill" style="width:43%;background:var(--dim);"></div></div></div>
             <div><div style="display:flex;justify-content:space-between;font-size:0.68rem;margin-bottom:0.2rem;"><span style="color:var(--amber);">LLM+GRPO</span><span>75%</span></div><div class="bar-track" style="height:8px;"><div class="bar-fill" style="width:75%;background:var(--amber);"></div></div></div>
-            <div style="font-size:0.68rem;color:var(--muted);">Lift: <span style="color:var(--amber);font-weight:700;">+33pp</span></div>
+            <div style="font-size:0.68rem;color:var(--muted);">Lift: <span style="color:var(--amber);font-weight:700;">+35pp</span></div>
           </div>
         </div>
         <div class="card-sm">
@@ -891,7 +892,7 @@ body{font-family:var(--mono);background:var(--bg);color:var(--text);min-height:1
           <div style="display:flex;flex-direction:column;gap:0.45rem;">
             <div><div style="display:flex;justify-content:space-between;font-size:0.68rem;margin-bottom:0.2rem;"><span style="color:var(--muted);">Greedy (D0)</span><span>33%</span></div><div class="bar-track" style="height:8px;"><div class="bar-fill" style="width:33%;background:var(--dim);"></div></div></div>
             <div><div style="display:flex;justify-content:space-between;font-size:0.68rem;margin-bottom:0.2rem;"><span style="color:var(--red);">LLM+GRPO</span><span>63%</span></div><div class="bar-track" style="height:8px;"><div class="bar-fill" style="width:63%;background:var(--red);"></div></div></div>
-            <div style="font-size:0.68rem;color:var(--muted);">Lift: <span style="color:var(--red);font-weight:700;">+30pp</span></div>
+            <div style="font-size:0.68rem;color:var(--muted);">Lift: <span style="color:var(--red);font-weight:700;">+28pp</span></div>
           </div>
         </div>
       </div>