feat(report-ui): local report viewer for iterating on evals

packages/core/src/report/metadata.ts

@@ -1,7 +1,7 @@
 import { z } from "zod";
 import { HarnessRunSchema, ToolCallSchema } from "../harness";
 import { JsonObjectSchema, JsonValueSchema } from "../json";
-import { isJsonObject, NullableFiniteNumberSchema } from "../schema-utils";
+import { NullableFiniteNumberSchema, isJsonObject } from "../schema-utils";
 
 /** Harness metadata stored by vitest-evals on Vitest task metadata. */
 export const HarnessMetaSchema = z
@@ -31,6 +31,8 @@ export const EvalMetaSchema = z
   .object({
     scores: z.array(EvalScoreSchema).optional(),
     avgScore: NullableFiniteNumberSchema,
+    input: JsonValueSchema.optional(),
+    expected: JsonValueSchema.optional(),
     output: JsonValueSchema.optional(),
     thresholdFailed: z.boolean().optional(),
     toolCalls: z.array(ToolCallSchema).optional(),

packages/report-ui/package.json

@@ -18,9 +18,7 @@
   "type": "module",
   "main": "./dist/index.js",
   "module": "./dist/index.js",
-  "files": [
-    "dist"
-  ],
+  "files": ["dist"],
   "publishConfig": {
     "access": "public"
   },
@@ -41,13 +39,15 @@
   },
   "devDependencies": {
     "@tailwindcss/vite": "^4.3.0",
+    "@tanstack/react-router": "^1.170.32",
     "@types/react": "^19.2.7",
     "@types/react-dom": "^19.2.3",
-    "@vitejs/plugin-react": "^5.1.1",
+    "@vitejs/plugin-react": "^6.1.1",
     "react": "^19.2.3",
     "react-dom": "^19.2.3",
+    "react18-json-view": "^0.2.10",
     "tailwindcss": "^4.3.0",
-    "vite": "^7.3.0",
+    "vite": "^8.2.2",
     "vite-tsconfig-paths": "^6.1.1"
   }
 }

packages/report-ui/src/app/App.test.ts

@@ -1,7 +1,9 @@
-import { afterEach, describe, expect, test, vi } from "vitest";
 import type { ReportWorkspace } from "@vitest-evals/core";
+import { afterEach, describe, expect, test, vi } from "vitest";
 import {
+  adjacentVisibleCase,
   loadWorkspace,
+  nextSortSearch,
   resolveSelectedCase,
   resolveSelectedCaseId,
   summarizeVisibleWorkspace,
@@ -40,6 +42,46 @@ afterEach(() => {
   vi.unstubAllGlobals();
 });
 
+describe("nextSortSearch", () => {
+  test("starts a new column with its default direction and toggles the active one", () => {
+    expect(nextSortSearch(undefined, "asc", "score")).toEqual({
+      dir: "desc",
+      sort: "score",
+    });
+    expect(nextSortSearch("score", "desc", "score")).toEqual({
+      dir: "asc",
+      sort: "score",
+    });
+    expect(nextSortSearch("score", "asc", "case")).toEqual({
+      dir: "asc",
+      sort: "case",
+    });
+    expect(nextSortSearch(undefined, "asc", "model")).toEqual({
+      dir: "asc",
+      sort: "model",
+    });
+  });
+});
+
+describe("adjacentVisibleCase", () => {
+  test("opens the first or last case when nothing is selected", () => {
+    expect(adjacentVisibleCase(cases, undefined, 1)?.id).toBe("failed-case");
+    expect(adjacentVisibleCase(cases, undefined, -1)?.id).toBe("passed-case");
+  });
+
+  test("moves to the next visible case and stays at the ends", () => {
+    expect(adjacentVisibleCase(cases, "failed-case", 1)?.id).toBe(
+      "passed-case",
+    );
+    expect(adjacentVisibleCase(cases, "passed-case", 1)?.id).toBe(
+      "passed-case",
+    );
+    expect(adjacentVisibleCase(cases, "passed-case", -1)?.id).toBe(
+      "failed-case",
+    );
+  });
+});
+
 describe("case selection", () => {
   test("keeps selection scoped to visible filtered cases", () => {
     const visibleCases = [cases[1]!];
@@ -109,6 +151,7 @@ describe("visible summary", () => {
         },
         {
           query: "",
+          model: "all",
           runId: "run-2",
           status: "passed",
         },
@@ -134,6 +177,7 @@ describe("visible summary", () => {
         runs,
         {
           query: "",
+          model: "all",
           runId: "run-2",
           status: "passed",
         },
@@ -151,6 +195,7 @@ describe("visible summary", () => {
         runs,
         {
           query: "",
+          model: "all",
           runId: "all",
           status: "all",
         },
@@ -183,6 +228,7 @@ describe("visible summary", () => {
         },
         {
           query: "",
+          model: "all",
           runId: "all",
           status: "all",
         },