Update index.html

index.html

@@ -529,6 +529,7 @@
         <button class="tab active" data-idx="0">Partial Replication</button>
         <button class="tab" data-idx="1">Compilation Success</button>
         <button class="tab" data-idx="2">Coefficient Direction</button>
+        <button class="tab" data-idx="3">Significant Level Correctness</button>
     </div>
 
     <div class="panel show" id="panel-rep">
@@ -543,7 +544,7 @@
             <div class="lb-columns">
                 <span>#</span>
                 <span>Model</span>
-                <span class="col-score">Score</span>
+                <span class="col-score">Score (0‑100)</span>
             </div>
             <div id="rows-rep"><div class="loading"><div class="loading-spinner"></div>Loading rankings…</div></div>
         </div>
@@ -561,7 +562,7 @@
             <div class="lb-columns">
                 <span>#</span>
                 <span>Model</span>
-                <span class="col-score">Score</span>
+                <span class="col-score">Score (0‑100)</span>
             </div>
             <div id="rows-compile"><div class="loading"><div class="loading-spinner"></div>Loading rankings…</div></div>
         </div>
@@ -579,12 +580,31 @@
             <div class="lb-columns">
                 <span>#</span>
                 <span>Model</span>
-                <span class="col-score">Score</span>
+                <span class="col-score">Score (0‑100)</span>
             </div>
             <div id="rows-dir"><div class="loading"><div class="loading-spinner"></div>Loading rankings…</div></div>
         </div>
     </div>
 
+    <!-- New panel for Significant Level Correctness -->
+    <div class="panel" id="panel-sig">
+        <div class="leaderboard-card">
+            <div class="leaderboard-header">
+                <h2>
+                    <svg width="16" height="16" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2"><path d="M18 20V10M12 20V4M6 20v-6"/></svg>
+                    Leaderboard
+                </h2>
+                <span class="badge-official">Official Benchmark</span>
+            </div>
+            <div class="lb-columns">
+                <span>#</span>
+                <span>Model</span>
+                <span class="col-score">Score (0‑100)</span>
+            </div>
+            <div id="rows-sig"><div class="loading"><div class="loading-spinner"></div>Loading rankings…</div></div>
+        </div>
+    </div>
+
     <div id="error-banner" class="error-banner hidden"></div>
 
     <div class="metric-card">
@@ -596,9 +616,10 @@
             <li><strong>Compilation Success:</strong> Generated econometric code executes completely without runtime or syntax errors.</li>
             <li><strong>Partial Replication:</strong> The target treatment coefficient can be reproduced within a 10% relative error threshold.</li>
             <li><strong>Correct Coefficient Direction:</strong> The sign (positive/negative) of the treatment‑effect coefficient matches ground‑truth results.</li>
+            <li><strong>Significant Level Correctness:</strong> The model correctly reproduces the statistical significance level of the treatment‑effect coefficient.</li>
         </ul>
         <div class="metric-note">
-            <em>Note:</em> Codex serves as a strong code‑specialized upper‑bound baseline with high overall metrics across all three evaluation dimensions. However, it only supports one‑shot code generation without agent‑level interactive planning or multi‑round revision capabilities, and its public API is no longer available. MetricsAI outperforms vanilla‑LLM and general‑purpose‑agent baselines for interactive real‑world econometric‑research workflows.
+            <em>Note:</em> Codex serves as a strong code‑specialized upper‑bound baseline with high overall metrics across all four evaluation dimensions. However, it only supports one‑shot code generation without agent‑level interactive planning or multi‑round revision capabilities, and its public API is no longer available. MetricsAI outperforms vanilla‑LLM and general‑purpose‑agent baselines for interactive real‑world econometric‑research workflows.
             <br><br>
             Dataset: <a href="https://huggingface.co/datasets/CamoAiLab/Infernet" target="_blank">CamoAiLab/Infernet</a>
         </div>
@@ -724,12 +745,13 @@ async function loadData() {
         buildRows(data, "Partial Replication (%)", "rows-rep");
         buildRows(data, "Compilation Success (%)", "rows-compile");
         buildRows(data, "Correct Coefficient Direction (%)", "rows-dir");
+        buildRows(data, "Significant Level Correctness (%)", "rows-sig");
     } catch (err) {
         console.error("load csv error", err);
         const banner = document.getElementById("error-banner");
         banner.textContent = `Failed to load ranking data: ${err.message}`;
         banner.classList.remove("hidden");
-        ["rows-rep","rows-compile", "rows-dir"].forEach(id => {
+        ["rows-rep","rows-compile", "rows-dir","rows-sig"].forEach(id => {
             document.getElementById(id).innerHTML = '<div class="loading">Unable to load data</div>';
         });
     }
@@ -738,4 +760,4 @@ async function loadData() {
 loadData();
 </script>
 </body>
-</html>
+</html>