Clarify source versus SFT measurements

index.html

@@ -500,7 +500,7 @@
           <div class="tr-run-head">
             <div>
               <strong>Recorded training diagnostics</strong>
-              <div class="text-small text-muted">Initial and final scheduled evaluations · NLL, lower is better</div>
+              <div class="text-small text-muted">Source and scheduled SFT evaluations · NLL, lower is better</div>
             </div>
           </div>
           <div class="tr-result-chart" role="img" aria-label="Held-out SFT NLL decreased from 1.7628 to 1.2209. PIQA normalized accuracy increased from 68.39 percent at the refinement source to 69.31 percent at selected SFT epoch 2.">
@@ -516,14 +516,14 @@
             </div>
           </div>
           <div class="tr-legend text-small" style="margin-top:30px">
-            <span class="tr-swatch"></span><span>Initial</span>
-            <span class="tr-swatch routed"></span><span>Final</span>
+            <span class="tr-swatch"></span><span>Source</span>
+            <span class="tr-swatch routed"></span><span>SFT measurement</span>
           </div>
         </div>
         <div class="card">
           <table class="table table-sm">
             <thead>
-              <tr><th>Measurement</th><th class="text-end">Initial</th><th class="text-end">Final</th></tr>
+              <tr><th>Measurement</th><th class="text-end">Source</th><th class="text-end">SFT checkpoint</th></tr>
             </thead>
             <tbody>
               <tr><td>Held-out SFT NLL</td><td class="text-end">1.7628 @ source</td><td class="text-end">1.2209 @ epoch 3</td></tr>