diff --git a/evals/astarpa/evals.ipynb b/evals/astarpa/evals.ipynb
index 236fb1d..e2e11ef 100644
--- a/evals/astarpa/evals.ipynb
+++ b/evals/astarpa/evals.ipynb
@@ -241,11 +241,28 @@
     "    if 'length' in df.columns and 'output_Ok_stats_expanded' in df.columns:\n",
     "        df['band'] = df['output_Ok_stats_expanded'] / (df['stats_seqpairs']* df['length'])\n",
     "\n",
+    "    def bad_k(row):\n",
+    "        # Bad k only makes sense for A*PA with SH-based heuristic.\n",
+    "        if row['algo_name'] != 'AstarPA':\n",
+    "            return False\n",
+    "        \n",
+    "        h = row['job_algo_AstarPA_heuristic_type'].lower()\n",
+    "        if h not in ['sh', 'csh', 'gcsh']:\n",
+    "            return False\n",
+    "        \n",
+    "        # Too small k => too many matches\n",
+    "        if row.k < math.log(row.length, 4) + (2 if row.r == 2 else 0):\n",
+    "            return True\n",
+    "        \n",
+    "        # Too large k => not enough potential\n",
+    "        if row.k > row.r/row.divergence:\n",
+    "            return True\n",
+    "        \n",
+    "        # all is fine\n",
+    "        return False\n",
     "\n",
     "    if 'length' in df.columns:\n",
-    "        df['bad_k'] = df.apply(lambda row: row.k < math.log(row.length, 4) + (2 if row.r == 2 else 0) or row.k > row.r/row.divergence, axis=1)\n",
-    "    #else:\n",
-    "        #df['stats_'] = df.apply(lambda row: row.k < math.log(row.length, 4) + (2 if row.r == 2 else 0) or row.k > row.r/row.divergence, axis=1)\n",
+    "        df['bad_k'] = df.apply(bad_k, axis=1)\n",
     "\n",
     "    def runtime_capped(row):\n",
     "        if not math.isnan(row['runtime']):\n",
@@ -311,7 +328,7 @@
     "         width=None,\n",
     "         height=None,\n",
     "         default_r=None,\n",
-    "         bad_k=False,\n",
+    "         bad_k=True,\n",
     "        ):\n",
     "    \n",
     "    if df[y].isna().all():\n",
@@ -971,7 +988,7 @@
     "df['rn'] = df.apply(lambda row: f\"{row['r']}-{row['length']}\", axis=1)\n",
     "for d, g in df.groupby('errorrate'):\n",
     "    plot(g, file=f'params_e{d}', x='k', y='s_per_pair', xlog=False, ylog=True, connect=True, line_labels=True,\n",
-    "             hue='length', style='r', xlim=(4, 26), bad_k = True,\n",
+    "             hue='length', style='r', xlim=(4, 26),\n",
     "             width=1.3*4.4, height=1.3*3)\n",
     "plt.show()\n",
     "\n",
@@ -995,7 +1012,7 @@
     "        subfig.suptitle(dataset_pretty[k], y=0)\n",
     "        axs = subfig.subplots(1, 2, sharey=True)\n",
     "        for (r, g), ax in zip(g.groupby(['r'], sort=False), axs): \n",
-    "            plot(g, x='k', y='runtime_capped', hue='r', xlog=False, ylog=True, categorical=True, ylim=(0.3, 150), bad_k = True, ax=ax)\n",
+    "            plot(g, x='k', y='runtime_capped', hue='r', xlog=False, ylog=True, categorical=True, ylim=(0.3, 150), ax=ax)\n",
     "            ax.set_xlabel(f'$k$, {\"in\" if r==2 else \"\"}exact matches ($r={r}$)')\n",
     "\n",
     "    plt.savefig(f\"plots/{file}.pdf\", dpi=300, bbox_inches='tight')\n",