ProjectTech4DevAI · kaapi-agent-bot · Sep 18, 2026
diff --git a/app/(main)/evaluations/[id]/page.tsx b/app/(main)/evaluations/[id]/page.tsx
@@ -51,7 +51,7 @@
  DownloadIcon,
 } from "@/app/components/icons";

 export default function EvaluationReport() {
  const router = useRouter();
  const params = useParams();
  const toast = useToast();
@@ -399,7 +399,10 @@
                     onResync={handleResync}
                   />
                   {categoryMetrics.length > 0 && (
-                    <CategoryMetricsTable categoryMetrics={categoryMetrics} />
+                    <CategoryMetricsTable
+                      categoryMetrics={categoryMetrics}
+                      isJudgeRun={job.is_judge_run}
+                    />
                   )}
                 </>
               ) : (

diff --git a/app/components/evaluations/CategoryMetricsTable.tsx b/app/components/evaluations/CategoryMetricsTable.tsx
@@ -4,6 +4,7 @@ import { CategoryMetric } from "@/app/lib/types/evaluation";
 
 interface CategoryMetricsTableProps {
   categoryMetrics: CategoryMetric[];
+  isJudgeRun?: boolean;
 }
 
 function formatScore(value: number | null): string {
@@ -13,6 +14,7 @@ function formatScore(value: number | null): string {
 
 export default function CategoryMetricsTable({
   categoryMetrics,
+  isJudgeRun,
 }: CategoryMetricsTableProps) {
   if (!categoryMetrics || categoryMetrics.length === 0) return null;
 
@@ -36,9 +38,11 @@ export default function CategoryMetricsTable({
               <th className="px-5 py-2.5 text-right text-[11px] font-semibold uppercase tracking-wide text-text-secondary">
                 Total Questions
               </th>
-              <th className="px-5 py-2.5 text-right text-[11px] font-semibold uppercase tracking-wide text-text-secondary">
-                Avg Cosine
-              </th>
+              {!isJudgeRun && (
+                <th className="px-5 py-2.5 text-right text-[11px] font-semibold uppercase tracking-wide text-text-secondary">
+                  Avg Cosine
+                </th>
+              )}
               <th className="px-5 py-2.5 text-right text-[11px] font-semibold uppercase tracking-wide text-text-secondary">
                 Avg Correctness
               </th>
@@ -56,9 +60,11 @@ export default function CategoryMetricsTable({
                 <td className="px-5 py-2.5 text-sm text-right tabular-nums text-text-primary">
                   {row.total_evals}
                 </td>
-                <td className="px-5 py-2.5 text-sm text-right tabular-nums text-text-primary">
-                  {formatScore(row.avg_cosine)}
-                </td>
+                {!isJudgeRun && (
+                  <td className="px-5 py-2.5 text-sm text-right tabular-nums text-text-primary">
+                    {formatScore(row.avg_cosine)}
+                  </td>
+                )}
                 <td className="px-5 py-2.5 text-sm text-right tabular-nums text-text-primary">
                   {formatScore(row.avg_correctness)}
                 </td>

diff --git a/app/components/evaluations/DetailedResultsTable.tsx b/app/components/evaluations/DetailedResultsTable.tsx
@@ -12,11 +12,13 @@ import {
   hasSummaryScores,
   isNewScoreObjectV2,
   isGroupedFormat,
+  isCosineScoreName,
 } from "@/app/lib/utils/evaluation";
 import { formatScoreValue, getScoreByName } from "@/app/lib/utils";
 import { InfoTooltip } from "@/app/components/ui";
 import { GroupedResultsTable } from "@/app/components/evaluations";
 import { MarkdownContent } from "@/app/components/chat";
+import { ExternalLinkIcon } from "@/app/components/icons";
 
 interface DetailedResultsTableProps {
   job: EvalJob;
@@ -26,6 +28,7 @@ export default function DetailedResultsTable({
   job,
 }: DetailedResultsTableProps) {
   const scoreObject = getScoreObject(job);
+  const isJudgeRun = job.is_judge_run;
 
   if (!scoreObject || !hasSummaryScores(scoreObject)) {
     return (
@@ -42,6 +45,7 @@ export default function DetailedResultsTable({
       return (
         <GroupedResultsTable
           traces={scoreObject.traces as GroupedTraceItem[]}
+          isJudgeRun={isJudgeRun}
         />
       );
     }
@@ -61,12 +65,16 @@ export default function DetailedResultsTable({
   }
 
   // Get all unique score names from the first item
-  const scoreNames =
+  const allScoreNames =
     individual_scores[0]?.trace_scores?.map((s) => s.name) || [];
+  const scoreNames = isJudgeRun
+    ? allScoreNames.filter((name) => !isCosineScoreName(name))
+    : allScoreNames;
 
   const hasAnyCategory = individual_scores.some(
     (s) => (s.category ?? "").trim().length > 0,
   );
+  const hasAnyTraceUrl = individual_scores.some((s) => s.score_trace_url);
 
   const COLUMN_WIDTHS = {
     index: 50,
@@ -75,14 +83,16 @@ export default function DetailedResultsTable({
     groundTruth: 250,
     answer: 250,
     score: 160,
+    trace: 90,
   };
   const tableMinWidth =
     COLUMN_WIDTHS.index +
     (hasAnyCategory ? COLUMN_WIDTHS.category : 0) +
     COLUMN_WIDTHS.question +
     COLUMN_WIDTHS.groundTruth +
     COLUMN_WIDTHS.answer +
-    scoreNames.length * COLUMN_WIDTHS.score;
+    scoreNames.length * COLUMN_WIDTHS.score +
+    (hasAnyTraceUrl ? COLUMN_WIDTHS.trace : 0);
 
   return (
     <div className="border rounded-lg overflow-hidden bg-white border-gray-200">
@@ -132,6 +142,14 @@ export default function DetailedResultsTable({
                   {scoreName}
                 </th>
               ))}
+              {hasAnyTraceUrl && (
+                <th
+                  className="px-4 py-3 text-center text-xs font-semibold uppercase text-bg-primary"
+                  style={{ width: `${COLUMN_WIDTHS.trace}px` }}
+                >
+                  Trace
+                </th>
+              )}
             </tr>
           </thead>
 
@@ -209,6 +227,23 @@ export default function DetailedResultsTable({
                       </td>
                     );
                   })}
+
+                  {hasAnyTraceUrl && (
+                    <td className="px-4 py-3 text-center align-top">
+                      {item.score_trace_url && (
+                        <a
+                          href={item.score_trace_url}
+                          target="_blank"
+                          rel="noopener noreferrer"
+                          className="inline-flex items-center justify-center text-accent-primary hover:opacity-75 transition-opacity"
+                          aria-label="View trace"
+                          title="View trace"
+                        >
+                          <ExternalLinkIcon className="w-4 h-4" />
+                        </a>
+                      )}
+                    </td>
+                  )}
                 </tr>
               );
             })}

diff --git a/app/components/evaluations/EvalRunCard.tsx b/app/components/evaluations/EvalRunCard.tsx
@@ -25,6 +25,7 @@ export default function EvalRunCard({
   const isCompleted = job.status?.toLowerCase() === "completed";
   const scoreObj = getScoreObject(job);
   const statusColor = getStatusColor(job.status || "");
+  const isJudgeRun = job.is_judge_run;
 
   return (
     <div
@@ -59,7 +60,11 @@ export default function EvalRunCard({
 
         {scoreObj && (
           <div className="mt-3">
-            <ScoreDisplay score={scoreObj} errorMessage={job.error_message} />
+            <ScoreDisplay
+              score={scoreObj}
+              errorMessage={job.error_message}
+              isJudgeRun={isJudgeRun}
+            />
           </div>
         )}
 
@@ -91,7 +96,7 @@ export default function EvalRunCard({
                           </span>
                         </>
                       )}
-                      {job.cost.embedding && (
+                      {job.cost.embedding && !isJudgeRun && (
                         <>
                           <span>Cosine similarity calculation</span>
                           <span className="text-left">

diff --git a/app/components/evaluations/GroupedResultsTable.tsx b/app/components/evaluations/GroupedResultsTable.tsx
@@ -6,12 +6,15 @@
 
 import { Fragment } from "react";
 import { TraceScore, GroupedTraceItem } from "@/app/lib/types/evaluation";
+import { isCosineScoreName } from "@/app/lib/utils/evaluation";
 import { formatScoreValue } from "@/app/lib/utils";
 import { InfoTooltip } from "@/app/components/ui";
 export default function GroupedResultsTable({
   traces,
+  isJudgeRun,
 }: {
   traces: GroupedTraceItem[];
+  isJudgeRun?: boolean;
 }) {
   if (!traces || traces.length === 0) {
     return (
@@ -171,8 +174,13 @@ export default function GroupedResultsTable({
                   <td className="px-4 pt-1 pb-3 bg-accent-subtle/50" />
 
                   {Array.from({ length: maxAnswers }, (_, answerIndex) => {
-                    const answerScores: TraceScore[] =
+                    const rawAnswerScores: TraceScore[] =
                       group.scores?.[answerIndex] || [];
+                    const answerScores = isJudgeRun
+                      ? rawAnswerScores.filter(
+                          (s) => !isCosineScoreName(s.name),
+                        )
+                      : rawAnswerScores;
                     const answer = group.llm_answers[answerIndex];
 
                     return (

diff --git a/app/components/evaluations/MetricsOverview.tsx b/app/components/evaluations/MetricsOverview.tsx
@@ -1,6 +1,7 @@
 "use client";
 
 import type { EvalJob, SummaryScore } from "@/app/lib/types/evaluation";
+import { isCosineScoreName } from "@/app/lib/utils/evaluation";
 import { Button } from "@/app/components/ui";
 import { RefreshIcon, WarningTriangleIcon } from "@/app/components/icons";
 
@@ -14,11 +15,15 @@ interface MetricsOverviewProps {
 
 export default function MetricsOverview({
   job,
-  summaryScores,
+  summaryScores: allSummaryScores,
   isJobInProgress,
   isResyncing,
   onResync,
 }: MetricsOverviewProps) {
+  const isJudgeRun = job.is_judge_run;
+  const summaryScores = isJudgeRun
+    ? allSummaryScores.filter((s) => !isCosineScoreName(s.name))
+    : allSummaryScores;
   const showPartialNotice =
     summaryScores.some(
       (s) => job.total_items && s.total_pairs < job.total_items,
@@ -38,15 +43,17 @@ export default function MetricsOverview({
         <h3 className="text-sm font-semibold text-text-secondary">
           Metrics Overview
         </h3>
-        <Button
-          variant="primary"
-          size="sm"
-          onClick={onResync}
-          disabled={isResyncing}
-        >
-          <RefreshIcon className={isResyncing ? "animate-spin" : ""} />
-          {isResyncing ? "Resyncing..." : "Resync"}
-        </Button>
+        {!isJudgeRun && (
+          <Button
+            variant="primary"
+            size="sm"
+            onClick={onResync}
+            disabled={isResyncing}
+          >
+            <RefreshIcon className={isResyncing ? "animate-spin" : ""} />
+            {isResyncing ? "Resyncing..." : "Resync"}
+          </Button>
+        )}
       </div>
       {summaryScores.length > 0 ? (
         <div className="flex gap-4 flex-wrap">

diff --git a/app/components/evaluations/ScoreDisplay.tsx b/app/components/evaluations/ScoreDisplay.tsx
@@ -6,16 +6,21 @@
 "use client";
 
 import type { ScoreObject } from "@/app/lib/types/evaluation";
-import { hasSummaryScores } from "@/app/lib/utils/evaluation";
+import {
+  hasSummaryScores,
+  isCosineScoreName,
+} from "@/app/lib/utils/evaluation";
 
 interface ScoreDisplayProps {
   score: ScoreObject | null;
   errorMessage?: string | null;
+  isJudgeRun?: boolean;
 }
 
 export default function ScoreDisplay({
   score,
   errorMessage,
+  isJudgeRun,
 }: ScoreDisplayProps) {
   if (!score) {
     return (
@@ -31,7 +36,9 @@ export default function ScoreDisplay({
 
   // Handle score format with summary_scores (V1, V2, or Basic)
   if (hasSummaryScores(score)) {
-    const summaryScores = score.summary_scores || [];
+    const summaryScores = (score.summary_scores || []).filter(
+      (s) => !isJudgeRun || !isCosineScoreName(s.name),
+    );
 
     if (summaryScores.length === 0) {
       return (

diff --git a/app/components/icons/evaluations/ExternalLinkIcon.tsx b/app/components/icons/evaluations/ExternalLinkIcon.tsx
@@ -0,0 +1,21 @@
+interface IconProps {
+  className?: string;
+}
+
+export default function ExternalLinkIcon({ className }: IconProps) {
+  return (
+    <svg
+      className={className}
+      fill="none"
+      viewBox="0 0 24 24"
+      stroke="currentColor"
+      strokeWidth={2}
+      strokeLinecap="round"
+      strokeLinejoin="round"
+    >
+      <path d="M18 13v6a2 2 0 01-2 2H5a2 2 0 01-2-2V8a2 2 0 012-2h6" />
+      <polyline points="15 3 21 3 21 9" />
+      <line x1="10" y1="14" x2="21" y2="3" />
+    </svg>
+  );
+}
diff --git a/app/components/icons/index.tsx b/app/components/icons/index.tsx
@@ -29,6 +29,7 @@ export { default as EditIcon } from "./evaluations/EditIcon";
 export { default as MenuIcon } from "./evaluations/MenuIcon";
 export { default as DatabaseIcon } from "./evaluations/DatabaseIcon";
 export { default as GroupIcon } from "./evaluations/GroupIcon";
+export { default as ExternalLinkIcon } from "./evaluations/ExternalLinkIcon";
 
 // Document Icons
 export { default as DocumentFileIcon } from "./document/DocumentFileIcon";

diff --git a/app/lib/types/evaluation.ts b/app/lib/types/evaluation.ts
@@ -15,6 +15,7 @@ export interface TraceItem {
   question_id?: number;
   category?: string;
   scores: TraceScore[];
+  score_trace_url?: string;
 }
 
 export interface GroupedTraceItem {
@@ -42,6 +43,7 @@ export interface IndividualScore {
     response_id?: string;
   };
   trace_scores: TraceScore[];
+  score_trace_url?: string;
 }
 
 export interface CategoryMetric {
@@ -129,6 +131,7 @@ export interface EvalJob {
   id: number;
   run_name: string;
   run_mode?: RunMode;
+  is_judge_run?: boolean;
   dataset_name: string;
   dataset_id: number;
   batch_job_id: number;

diff --git a/app/lib/utils/evaluation.ts b/app/lib/utils/evaluation.ts
@@ -26,6 +26,10 @@ export function getScoreObject(job: EvalJob): ScoreObject | null {
   return job.scores || job.score || null;
 }
 
+export function isCosineScoreName(name: string): boolean {
+  return /cosine/i.test(name);
+}
+
 export function isGroupedFormat(
   traces: TraceItem[] | GroupedTraceItem[],
 ): traces is GroupedTraceItem[] {
@@ -47,6 +51,7 @@ export function normalizeToIndividualScores(
         output: { answer: trace.llm_answer },
         metadata: { ground_truth: trace.ground_truth_answer },
         trace_scores: trace.scores,
+        score_trace_url: trace.score_trace_url,
       };
     }
     return { trace_id: "", trace_scores: [] };