feat(studio): comparison analytics charts for skills/workflow benchmarking (#1104)

* feat(studio): add analytics charts with baseline comparison

Rename Compare tab to Analytics. Add recharts for visualization.
Implement ?baseline=<target> query param on /api/compare endpoint
to compute delta and normalized gain (g) per cell. Add collapsible
analytics section below the aggregated matrix with:
- Normalized gain bar chart (horizontal, color-coded by effect)
- Tag × target pass rate heatmap
- Negative delta regression table
- Score distribution histogram
- Trend-over-time line chart

Closes #1102

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* fix(studio): address code review findings for analytics charts

- Fix serve.ts: build cells with delta/normalized_gain fields upfront
  instead of mutating via type bypass
- Fix query key collision: use distinct keys for compare vs baseline queries
- Add `enabled: !!baseline` guard to prevent unnecessary API calls
- Remove dead CostVsImprovement component and unused recharts imports
- Fix misleading GainRow.testId → experiment naming
- Rename "Compare runs" heading to "Analyze runs"
- Fix biome formatting issues

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* refactor(studio): rename CompareTab to AnalyticsTab and update docs

- Rename CompareTab.tsx → AnalyticsTab.tsx with updated exports
- Update imports in index.tsx and $benchmarkId.tsx route files
- Update studio.mdx docs: rename Compare section to Analytics
- Add analytics charts documentation with baseline selector, normalized
  gain chart, tag heatmap, negative delta table, score distribution,
  and trend-over-time chart descriptions
- Add three new screenshots: aggregated matrix, charts with baseline
  selector, and score trend over time

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

Author: christso

apps/cli/src/commands/results/serve.ts

@@ -695,6 +695,23 @@ async function handleCompare(c: C, { searchDir, agentvDir }: DataContext) {
     }
   }
 
+  // ── Baseline delta / normalized-gain computation ─────────────────────
+  const baselineTarget = c.req.query('baseline') ?? '';
+  if (baselineTarget && !targetsSet.has(baselineTarget)) {
+    return c.json({ error: `Baseline target "${baselineTarget}" does not exist in the data` }, 400);
+  }
+
+  // Build baseline lookup before constructing cells so we can include
+  // delta/normalized_gain in the initial cell objects (no mutation needed).
+  const baselineScores = new Map<string, number>();
+  if (baselineTarget) {
+    for (const entry of cellMap.values()) {
+      if (entry.target === baselineTarget && entry.evalCount > 0) {
+        baselineScores.set(entry.experiment, entry.scoreSum / entry.evalCount);
+      }
+    }
+  }
+
   const cells = [...cellMap.values()].map((entry) => {
     // Deduplicate tests: keep only the latest entry per test_id (last wins by insertion order)
     const dedupMap = new Map<string, (typeof entry.tests)[number]>();
@@ -706,15 +723,28 @@ async function handleCompare(c: C, { searchDir, agentvDir }: DataContext) {
     // Cap to most recent entries to prevent unbounded payloads
     const cappedTests = dedupedTests.slice(-MAX_TESTS_PER_CELL);
 
-    return {
+    const avgScore = entry.evalCount > 0 ? entry.scoreSum / entry.evalCount : 0;
+    const cell: Record<string, unknown> = {
       experiment: entry.experiment,
       target: entry.target,
       eval_count: entry.evalCount,
       passed_count: entry.passedCount,
       pass_rate: entry.evalCount > 0 ? entry.passedCount / entry.evalCount : 0,
-      avg_score: entry.evalCount > 0 ? entry.scoreSum / entry.evalCount : 0,
+      avg_score: avgScore,
       tests: cappedTests,
     };
+
+    // Append baseline comparison fields when a baseline is selected
+    if (baselineTarget && entry.target !== baselineTarget) {
+      const baseAvg = baselineScores.get(entry.experiment);
+      if (baseAvg !== undefined) {
+        cell.delta = Math.round((avgScore - baseAvg) * 1000) / 1000;
+        cell.normalized_gain =
+          baseAvg >= 1.0 ? null : Math.round(((avgScore - baseAvg) / (1 - baseAvg)) * 1000) / 1000;
+      }
+    }
+
+    return cell;
   });
 
   // Per-run entries sorted by timestamp descending (newest first).

apps/studio/package.json

@@ -13,7 +13,8 @@
     "@tanstack/react-query": "^5.75.5",
     "@tanstack/react-router": "^1.120.3",
     "react": "^19.1.0",
-    "react-dom": "^19.1.0"
+    "react-dom": "^19.1.0",
+    "recharts": "^3.8.1"
   },
   "devDependencies": {
     "@tailwindcss/vite": "^4.1.7",