Fix figure descriptions

index.html

@@ -82,8 +82,8 @@
             intervals from cluster-bootstrap resamples of the datasets. The left bar of each method is
             shaded by <span class="stab">stability</span> (how little the selected subset changes across
             bootstrap resamples of the data), the right bar by <span class="val">validity</span> (how well
-            injected noise features are rejected); darker is better for both, see the Stability and
-            Validity tabs. Hatched bars have no stability or validity result yet.
+            injected noise features are rejected); darker is better for both. For details, see the Stability and
+            Validity tabs.
           </p>
         </div>
         <div id="lb-table">
@@ -93,14 +93,8 @@
           <p>Methods are compared within a <em>comparison group</em>: one dataset split, one downstream
             model configuration and one feature budget. The selection is shared with the Performance Trade-offs tab.</p>
           <ul>
-            <li><strong>Type</strong>: the paper's functional taxonomy (Appendix A). <em>Search</em>
-              methods evaluate many feature subsets with a model-dependent utility: forward search (SFS),
-              backward search (SBE, MarkovBlanket) and leave-one-out (LOCO). <em>Non-search</em> methods
-              avoid repeated subset evaluations and rely on a proxy instead: correlation (F-Test), distance
-              ((R)ReliefF, LaplacianScore), information theory (MI, mRMR, GainRatio), trees (CART,
-              RFImportance) and regularization (Lasso, ElasticNet). Random is the baseline.</li>
-            <li><strong>#</strong> and <strong>Elo</strong>: rank by Elo, with gold, silver and bronze for the three best; +/− give the
-              95% interval (the Elo rating is explained under the figure above).</li>
+            <li><strong>Type</strong>: the method type following the <a href="https://arxiv.org/abs/2610.13644" target="_blank" rel="noopener">paper</a>'s functional taxonomy (Appendix A)
+            <li><strong>Elo</strong>: rank by Elo, +/− give the 95% interval (Elo rating is explained under the figure above).</li>
             <li><strong>Avg. rank</strong>: mean rank of the method's downstream error within the groups (1 is best).</li>
             <li><strong>Improvability</strong>: mean relative gap <code>(error − best error) / error</code> to
               the best method of the group (0 is best).</li>
@@ -110,8 +104,8 @@
             <li><strong>Budgets</strong>: budget <em>i</em> of <em>b</em> keeps <code>ceil(d · i / (b + 1))</code>
               of a dataset's <code>d</code> features when <code>d ≤ 100</code>, and <code>ceil(d ^ (i / (b + 1)))</code>
               otherwise.</li>
-            <li><strong>Time</strong>: median wall-clock time of the selection. <strong>Fallback</strong>:
-              share of runs in which the method ran out of time and random features filled the budget.</li>
+            <li><strong>Time</strong>: median wall-clock time of the selection.</li>
+            <li><strong>Fallback</strong>: share of runs in which the method ran out of time and random features filled the budget.</li>
           </ul>
         </div>
         </div>
@@ -149,10 +143,10 @@
           <iframe class="frame sa-frame fig-frame" id="plot-pareto" title="Pareto front of improvement over Random against feature selection time per 1k features" style="height:470px"></iframe>
           <figcaption class="caption">
             <strong>Predictive performance against selection time.</strong> Improvement of the
-            downstream error over Random (%, higher is better) against the feature selection time per
+            downstream model error over Random (%, higher is better) against the feature selection time per
             1,000 features (log scale, left is faster), averaged over budgets, downstream models and
             splits per dataset, then over datasets. Orange methods are on the Pareto front: no other
-            method is both faster and better. As in the paper, LaplacianScore is left out of the
+            method is both faster and better. LaplacianScore is left out of the
             figure (far below Random); its values are in the CSV.
           </figcaption>
         </figure>