EntityProcess · christso · Jun 27, 2026 · Jun 27, 2026
diff --git a/apps/cli/src/commands/results/serve.ts b/apps/cli/src/commands/results/serve.ts
@@ -1203,6 +1203,7 @@ function attachRunDetailReadModelFields<T extends Record<string, unknown>>(
     return {
       ...result,
       ...(record.aggregation && { aggregation: record.aggregation }),
+      ...(record.eval_path && { eval_path: record.eval_path }),
       ...(record.result_dir && { result_dir: record.result_dir }),
       ...(record.summary_path && { summary_path: record.summary_path }),
       ...(record.grading_path && { grading_path: record.grading_path }),

diff --git a/apps/cli/test/commands/results/serve.test.ts b/apps/cli/test/commands/results/serve.test.ts
@@ -2581,18 +2581,30 @@ describe('serve app', () => {
       const filename = '2026-03-25T10-00-00-000Z';
       const runDir = path.join(runsDir, filename);
       mkdirSync(runDir, { recursive: true });
-      writeFileSync(path.join(runDir, 'index.jsonl'), toJsonl(RESULT_A, RESULT_B));
+      writeFileSync(
+        path.join(runDir, 'index.jsonl'),
+        toJsonl(
+          {
+            ...RESULT_A,
+            eval_path: 'evals/demo.eval.yaml',
+            result_dir: 'demo/test-greeting',
+          },
+          RESULT_B,
+        ),
+      );
 
       const app = createApp([], tempDir, tempDir, undefined, { studioDir });
       const res = await app.request(`/api/runs/${filename}`);
       expect(res.status).toBe(200);
       const data = (await res.json()) as {
-        results: { testId: string }[];
+        results: { testId: string; eval_path?: string; result_dir?: string }[];
         source: 'local' | 'remote';
         source_label: string;
       };
       expect(data.results).toHaveLength(2);
       expect(data.results[0].testId).toBe('test-greeting');
+      expect(data.results[0].eval_path).toBe('evals/demo.eval.yaml');
+      expect(data.results[0].result_dir).toBe('demo/test-greeting');
       expect(data.source).toBe('local');
       expect(data.source_label).toBe(filename);
     });

diff --git a/apps/dashboard/src/components/EvalDetail.tsx b/apps/dashboard/src/components/EvalDetail.tsx
@@ -335,8 +335,9 @@ function SourceTab({ result }: { result: EvalResult }) {
         <h4 className="mb-3 text-sm font-medium text-gray-300">Traceability</h4>
         <dl className="grid gap-3 md:grid-cols-2">
           <SourceMetaRow label="Eval file" value={traceability.eval_file?.display_path} />
+          <SourceMetaRow label="Eval" value={result.eval_path} />
           <SourceMetaRow label="Test ID" value={traceability.test_id ?? result.testId} />
-          <SourceMetaRow label="Suite" value={result.suite} />
+          <SourceMetaRow label="Legacy suite" value={result.suite} />
           <SourceMetaRow label="Category" value={result.category} />
           <SourceMetaRow label="Target" value={result.target} />
         </dl>

diff --git a/apps/dashboard/src/components/EvalSourceLabel.tsx b/apps/dashboard/src/components/EvalSourceLabel.tsx
@@ -0,0 +1,18 @@
+import { formatEvalSourceDisplay } from '~/lib/run-detail-context';
+import type { EvalResult } from '~/lib/types';
+
+interface EvalSourceLabelProps {
+  result: Pick<EvalResult, 'eval_path' | 'suite'>;
+  className?: string;
+}
+
+export function EvalSourceLabel({ result, className = '' }: EvalSourceLabelProps) {
+  const display = formatEvalSourceDisplay(result);
+  if (!display) return null;
+
+  return (
+    <span className={`block truncate text-gray-500 ${className}`} title={display.title}>
+      {display.label}
+    </span>
+  );
+}
diff --git a/apps/dashboard/src/components/ResultTable.tsx b/apps/dashboard/src/components/ResultTable.tsx
@@ -10,6 +10,7 @@ import type React from 'react';
 import { Fragment, useEffect, useMemo, useState } from 'react';
 
 import { useFeedback } from '~/lib/api';
+import { evalResultPath } from '~/lib/navigation';
 import {
   RESULT_TABLE_VIEW_PRESETS,
   type RepeatRunGroup,
@@ -651,8 +652,8 @@ function TrialResultCell({
       return <TargetCell target={row.targetLabel} tone="text-gray-500" />;
     case 'score':
       return <PassRatePill rate={trial.score ?? 0} />;
-    case 'suite':
-      return <TruncatedMuted value={row.suiteLabel} tone="text-gray-500" />;
+    case 'eval':
+      return <TruncatedMuted value={row.evalLabel} tone="text-gray-500" />;
     case 'category':
       return <TruncatedMuted value={row.categoryLabel} tone="text-gray-500" />;
     case 'duration':
@@ -833,8 +834,8 @@ function ResultCell({
       ) : (
         <PassRatePill rate={row.result.score} />
       );
-    case 'suite':
-      return <TruncatedMuted value={row.suiteLabel} />;
+    case 'eval':
+      return <TruncatedMuted value={row.evalLabel} />;
     case 'category':
       return <TruncatedMuted value={row.categoryLabel} />;
     case 'duration':
@@ -910,11 +911,10 @@ function ResultDetailPanel({
   onOpenTrialDetail: (trial: EvalCaseTrial, initialTab?: DetailTab) => void;
   onClose: () => void;
 }) {
-  const evalDetailHref = buildEvalDetailHref({
+  const evalDetailHref = evalResultPath(runId, row.testId, {
     projectId,
-    runId,
-    evalId: row.testId,
     resultDir: row.result.result_dir,
+    evalPath: row.result.eval_path,
   });
   const title = selectedTrialPath ? `${row.testId} · ${selectedTrialPath}` : row.testId;
   const showAggregateRepeatDetail = repeatGroup && !selectedTrial;
@@ -934,7 +934,7 @@ function ResultDetailPanel({
           </h4>
           <p className="mt-1 truncate text-xs text-gray-500" title={row.targetLabel}>
             {row.targetLabel}
-            {row.suiteLabel ? ` · ${row.suiteLabel}` : ''}
+            {row.evalLabel ? ` · ${row.evalLabel}` : ''}
           </p>
         </div>
         <div className="flex shrink-0 items-center gap-2">
@@ -1003,19 +1003,6 @@ function ExpanderCell({
   );
 }
 
-function buildEvalDetailHref(options: {
-  projectId?: string;
-  runId: string;
-  evalId: string;
-  resultDir?: string;
-}): string {
-  const base = options.projectId
-    ? `/projects/${encodeURIComponent(options.projectId)}/evals/${encodeURIComponent(options.runId)}/${encodeURIComponent(options.evalId)}`
-    : `/evals/${encodeURIComponent(options.runId)}/${encodeURIComponent(options.evalId)}`;
-  if (!options.resultDir) return base;
-  return `${base}?result_dir=${encodeURIComponent(options.resultDir)}`;
-}
-
 function scrollPanelIntoView(panel: HTMLElement | null) {
   if (!panel) return;
   window.requestAnimationFrame(() => {

diff --git a/apps/dashboard/src/components/Sidebar.tsx b/apps/dashboard/src/components/Sidebar.tsx
@@ -33,13 +33,18 @@ import {
   useRunList,
   useStudioConfig,
 } from '~/lib/api';
-import { shouldShowSuiteLabels } from '~/lib/run-detail-context';
+import {
+  evalResultIdentityKey,
+  evalResultSearchParams,
+  matchesEvalResultIdentity,
+} from '~/lib/navigation';
+import { shouldShowEvalSourceLabels } from '~/lib/run-detail-context';
 import { formatRunDisplay } from '~/lib/run-label';
 import { useSidebarContext } from '~/lib/sidebar-context';
 import type { EvalResult } from '~/lib/types';
 
 import { BrandName } from './BrandName';
-import { EvalSuiteLabel } from './EvalSuiteLabel';
+import { EvalSourceLabel } from './EvalSourceLabel';
 
 /** Responsive <aside> wrapper. Handles mobile overlay and desktop static placement. */
 function SidebarShell({ children }: { children: ReactNode }) {
@@ -104,11 +109,11 @@ function SidebarRunText({ display }: { display: ReturnType<typeof formatRunDispl
 function EvalSidebarItemContent({
   result,
   passThreshold,
-  showSuiteLabel,
+  showEvalSourceLabel,
 }: {
   result: EvalResult;
   passThreshold: number;
-  showSuiteLabel: boolean;
+  showEvalSourceLabel: boolean;
 }) {
   const passed = isPassing(result.score, passThreshold);
 
@@ -119,14 +124,23 @@ function EvalSidebarItemContent({
       </span>
       <span className="min-w-0 flex-1">
         <span className="block truncate">{result.testId}</span>
-        {showSuiteLabel ? (
-          <EvalSuiteLabel suite={result.suite} className="mt-1 max-w-full text-[11px] leading-4" />
+        {showEvalSourceLabel ? (
+          <EvalSourceLabel result={result} className="mt-1 max-w-full text-[11px] leading-4" />
         ) : null}
       </span>
     </>
   );
 }
 
+function useCurrentEvalIdentitySearch() {
+  const location = useLocation();
+  const search = location.search as Record<string, string | undefined>;
+  return {
+    resultDir: search.result_dir,
+    evalPath: search.eval_path,
+  };
+}
+
 type ProjectTabId = 'runs' | 'experiments' | 'analytics' | 'targets';
 
 const projectNavItems: { id: ProjectTabId; label: string; description: string }[] = [
@@ -409,8 +423,9 @@ function RunDetailSidebar({ currentRunId }: { currentRunId: string }) {
 function EvalSidebar({ runId, currentEvalId }: { runId: string; currentEvalId: string }) {
   const { data } = useRunDetail(runId);
   const { data: config } = useStudioConfig();
+  const currentIdentity = useCurrentEvalIdentitySearch();
   const passThreshold = config?.threshold ?? config?.pass_threshold ?? 0.8;
-  const showSuiteLabels = shouldShowSuiteLabels(data?.results ?? []);
+  const showEvalSourceLabels = shouldShowEvalSourceLabels(data?.results ?? []);
 
   return (
     <SidebarShell>
@@ -434,13 +449,18 @@ function EvalSidebar({ runId, currentEvalId }: { runId: string; currentEvalId: s
         </div>
 
         {data?.results.map((result) => {
-          const isActive = result.testId === currentEvalId;
+          const search = evalResultSearchParams({
+            resultDir: result.result_dir,
+            evalPath: result.eval_path,
+          });
+          const isActive = matchesEvalResultIdentity(result, currentEvalId, currentIdentity);
 
           return (
             <Link
-              key={result.testId}
+              key={evalResultIdentityKey(result)}
               to="/evals/$runId/$evalId"
               params={{ runId, evalId: result.testId }}
+              search={search}
               className={`mb-0.5 flex items-start gap-2 rounded-md px-2 py-1.5 text-sm transition-colors ${
                 isActive
                   ? 'bg-gray-800 text-cyan-400'
@@ -450,7 +470,7 @@ function EvalSidebar({ runId, currentEvalId }: { runId: string; currentEvalId: s
               <EvalSidebarItemContent
                 result={result}
                 passThreshold={passThreshold}
-                showSuiteLabel={showSuiteLabels}
+                showEvalSourceLabel={showEvalSourceLabels}
               />
             </Link>
           );
@@ -490,12 +510,17 @@ function SuiteSidebar({ runId, suite }: { runId: string; suite: string }) {
 
         {suiteResults.map((result) => {
           const passed = isPassing(result.score, passThreshold);
+          const search = evalResultSearchParams({
+            resultDir: result.result_dir,
+            evalPath: result.eval_path,
+          });
 
           return (
             <Link
-              key={result.testId}
+              key={evalResultIdentityKey(result)}
               to="/evals/$runId/$evalId"
               params={{ runId, evalId: result.testId }}
+              search={search}
               className="mb-0.5 flex items-center gap-2 rounded-md px-2 py-1.5 text-sm text-gray-400 transition-colors hover:bg-gray-800/50 hover:text-gray-200"
             >
               <span className={`text-xs ${passed ? 'text-emerald-400' : 'text-red-400'}`}>
@@ -609,8 +634,9 @@ function ProjectEvalSidebar({
 }) {
   const { data } = useProjectRunDetail(projectId, runId);
   const { data: config } = useStudioConfig(projectId);
+  const currentIdentity = useCurrentEvalIdentitySearch();
   const passThreshold = config?.threshold ?? config?.pass_threshold ?? 0.8;
-  const showSuiteLabels = shouldShowSuiteLabels(data?.results ?? []);
+  const showEvalSourceLabels = shouldShowEvalSourceLabels(data?.results ?? []);
 
   return (
     <SidebarShell>
@@ -632,12 +658,17 @@ function ProjectEvalSidebar({
           Evaluations
         </div>
         {data?.results.map((result) => {
-          const isActive = result.testId === currentEvalId;
+          const search = evalResultSearchParams({
+            resultDir: result.result_dir,
+            evalPath: result.eval_path,
+          });
+          const isActive = matchesEvalResultIdentity(result, currentEvalId, currentIdentity);
           return (
             <Link
-              key={result.testId}
+              key={evalResultIdentityKey(result)}
               to="/projects/$projectId/evals/$runId/$evalId"
               params={{ projectId, runId, evalId: result.testId }}
+              search={search}
               className={`mb-0.5 flex items-start gap-2 rounded-md px-2 py-1.5 text-sm transition-colors ${
                 isActive
                   ? 'bg-gray-800 text-cyan-400'
@@ -647,7 +678,7 @@ function ProjectEvalSidebar({
               <EvalSidebarItemContent
                 result={result}
                 passThreshold={passThreshold}
-                showSuiteLabel={showSuiteLabels}
+                showEvalSourceLabel={showEvalSourceLabels}
               />
             </Link>
           );
@@ -693,11 +724,16 @@ function ProjectSuiteSidebar({
         </div>
         {suiteResults.map((result) => {
           const passed = isPassing(result.score, passThreshold);
+          const search = evalResultSearchParams({
+            resultDir: result.result_dir,
+            evalPath: result.eval_path,
+          });
           return (
             <Link
-              key={result.testId}
+              key={evalResultIdentityKey(result)}
               to="/projects/$projectId/evals/$runId/$evalId"
               params={{ projectId, runId, evalId: result.testId }}
+              search={search}
               className="mb-0.5 flex items-center gap-2 rounded-md px-2 py-1.5 text-sm text-gray-400 transition-colors hover:bg-gray-800/50 hover:text-gray-200"
             >
               <span className={`text-xs ${passed ? 'text-emerald-400' : 'text-red-400'}`}>

diff --git a/apps/dashboard/src/lib/navigation.test.ts b/apps/dashboard/src/lib/navigation.test.ts
@@ -3,9 +3,13 @@ import { describe, expect, it } from 'bun:test';
 import {
   categoryPath,
   evalPath,
+  evalResultIdentityKey,
+  evalResultPath,
+  evalResultSearchParams,
   experimentPath,
   initialProjectRedirectStorageKey,
   jobPath,
+  matchesEvalResultIdentity,
   resolveIndexRoute,
   resolveInitialProjectRedirect,
   runPath,
@@ -58,6 +62,15 @@ describe('route path helpers', () => {
     expect(evalPath('run::1', 'case/a', 'demo project')).toBe(
       '/projects/demo%20project/evals/run%3A%3A1/case%2Fa',
     );
+    expect(
+      evalResultPath('run::1', 'case/a', {
+        projectId: 'demo project',
+        resultDir: 'evals/auth.eval.yaml/case-a',
+        evalPath: 'evals/auth.eval.yaml',
+      }),
+    ).toBe(
+      '/projects/demo%20project/evals/run%3A%3A1/case%2Fa?result_dir=evals%2Fauth.eval.yaml%2Fcase-a',
+    );
     expect(jobPath('job/1', 'demo project')).toBe('/projects/demo%20project/jobs/job%2F1');
     expect(categoryPath('run::1', 'Safety > PII', 'demo project')).toBe(
       '/projects/demo%20project/runs/run%3A%3A1/category/Safety%20%3E%20PII',
@@ -74,11 +87,42 @@ describe('route path helpers', () => {
   it('keeps unscoped paths for legacy single-project routes', () => {
     expect(runPath('run::1')).toBe('/runs/run%3A%3A1');
     expect(evalPath('run::1', 'case/a')).toBe('/evals/run%3A%3A1/case%2Fa');
+    expect(evalResultPath('run::1', 'case/a', { evalPath: 'evals/smoke.eval.yaml' })).toBe(
+      '/evals/run%3A%3A1/case%2Fa?eval_path=evals%2Fsmoke.eval.yaml',
+    );
     expect(jobPath('job/1')).toBe('/jobs/job%2F1');
     expect(categoryPath('run::1', 'Safety')).toBe('/runs/run%3A%3A1/category/Safety');
     expect(suitePath('run::1', 'evals/smoke.eval.yaml')).toBe(
       '/runs/run%3A%3A1/suite/evals%2Fsmoke.eval.yaml',
     );
     expect(runsHomePath()).toBe('/?tab=runs');
   });
+
+  it('prefers result_dir over eval_path for eval result query identity', () => {
+    expect(
+      evalResultSearchParams({
+        resultDir: 'opaque/case',
+        evalPath: 'evals/smoke.eval.yaml',
+      }),
+    ).toEqual({ result_dir: 'opaque/case' });
+    expect(evalResultSearchParams({ evalPath: 'evals/smoke.eval.yaml' })).toEqual({
+      eval_path: 'evals/smoke.eval.yaml',
+    });
+  });
+
+  it('matches and keys eval results by result_dir before eval_path', () => {
+    const result = {
+      testId: 'shared',
+      target: 'codex',
+      result_dir: 'opaque/shared',
+      eval_path: 'evals/auth.eval.yaml',
+    };
+
+    expect(evalResultIdentityKey(result)).toBe('opaque/shared');
+    expect(matchesEvalResultIdentity(result, 'shared', { resultDir: 'opaque/shared' })).toBe(true);
+    expect(matchesEvalResultIdentity(result, 'shared', { resultDir: 'other/shared' })).toBe(false);
+    expect(matchesEvalResultIdentity(result, 'shared', { evalPath: 'evals/auth.eval.yaml' })).toBe(
+      true,
+    );
+  });
 });