fix(core): derive score from assertions when score absent in code-grader

examples/features/code-grader-sdk/scripts/verify-attachments.ts

@@ -58,11 +58,5 @@ export default defineCodeGrader(({ expectedOutput, output, inputFiles }) => {
     }
   }
 
-  const passed = assertions.filter((a) => a.passed).length;
-  const score = assertions.length === 0 ? 0 : passed / assertions.length;
-
-  return {
-    score,
-    assertions,
-  };
+  return { assertions };
 });

examples/features/copilot-log-eval/graders/transcript-quality.ts

@@ -91,8 +91,5 @@ export default defineCodeGrader(({ output }) => {
     });
   }
 
-  const passed = assertions.filter((a) => a.passed).length;
-  const score = assertions.length > 0 ? passed / assertions.length : 0;
-
-  return { score, assertions };
+  return { assertions };
 });

examples/features/eval-assert-demo/.agentv/graders/keyword-check.ts

@@ -37,10 +37,5 @@ export default defineCodeGrader(({ output }) => {
     assertions.push({ text: 'Answer does not mention France', passed: false });
   }
 
-  const passed = assertions.filter((a) => a.passed).length;
-  const total = assertions.length;
-  return {
-    score: total > 0 ? passed / total : 0,
-    assertions,
-  };
+  return { assertions };
 });

examples/features/eval-assert-demo/.agentv/graders/length-check.ts

@@ -37,10 +37,5 @@ export default defineCodeGrader(({ output }) => {
     assertions.push({ text: `Answer has ${wordCount} words (> 50, too verbose)`, passed: false });
   }
 
-  const passed = assertions.filter((a) => a.passed).length;
-  const total = assertions.length;
-  return {
-    score: total > 0 ? passed / total : 0,
-    assertions,
-  };
+  return { assertions };
 });

examples/features/execution-metrics/scripts/check-metrics-present.ts

@@ -17,10 +17,7 @@ export default defineCodeGrader(({ trace, tokenUsage, costUsd, durationMs }) =>
   const assertions: Array<{ text: string; passed: boolean }> = [];
 
   if (!trace) {
-    return {
-      score: 0,
-      assertions: [{ text: 'No trace provided', passed: false }],
-    };
+    return { assertions: [{ text: 'No trace provided', passed: false }] };
   }
 
   // Check for tokenUsage
@@ -47,11 +44,5 @@ export default defineCodeGrader(({ trace, tokenUsage, costUsd, durationMs }) =>
     assertions.push({ text: 'durationMs not present', passed: false });
   }
 
-  const passed = assertions.filter((a) => a.passed).length;
-  const score = passed / assertions.length;
-
-  return {
-    score,
-    assertions,
-  };
+  return { assertions };
 });

examples/features/file-changes-with-repos/scripts/check-file-changes.ts

@@ -22,7 +22,7 @@ if (!fileChanges || fileChanges.trim().length === 0) {
     passed: false,
     evidence: 'file_changes is empty — workspace not configured or file tracking failed',
   });
-  console.log(JSON.stringify({ score: 0, assertions }));
+  console.log(JSON.stringify({ assertions }));
   process.exit(0);
 }
 
@@ -56,5 +56,4 @@ assertions.push({
   evidence: hasAddFn ? undefined : 'add() function not found in diff',
 });
 
-const passed = assertions.filter((a) => a.passed).length;
-console.log(JSON.stringify({ score: passed / assertions.length, assertions }));
+console.log(JSON.stringify({ assertions }));

examples/features/functional-grading/scripts/functional-check.ts

@@ -19,7 +19,6 @@ const workspacePath: string | null = input.workspace_path;
 if (!workspacePath) {
   console.log(
     JSON.stringify({
-      score: 0,
       assertions: [
         {
           text: 'workspace_path not provided — cannot run functional checks',
@@ -63,13 +62,4 @@ if (compiled) {
   runStage('tests', 'npm', ['test']);
 }
 
-const passed = assertions.filter((a) => a.passed).length;
-const total = assertions.length;
-const score = total > 0 ? passed / total : 0;
-
-console.log(
-  JSON.stringify({
-    score,
-    assertions,
-  }),
-);
+console.log(JSON.stringify({ assertions }));

examples/features/import-claude/graders/transcript-quality.ts

@@ -28,9 +28,5 @@ export default defineCodeGrader(({ output }) => {
     passed: emptyAssistant.length === 0,
   });
 
-  const passed = assertions.filter((a) => a.passed).length;
-  return {
-    score: assertions.length > 0 ? passed / assertions.length : 0,
-    assertions,
-  };
+  return { assertions };
 });

examples/features/workspace-artifact/scripts/check-csv-artifact.ts

@@ -25,7 +25,7 @@ if (!fileChanges || fileChanges.trim().length === 0) {
     passed: false,
     evidence: 'file_changes is empty — workspace snapshot or git baseline may not be configured',
   });
-  console.log(JSON.stringify({ score: 0, assertions }));
+  console.log(JSON.stringify({ assertions }));
   process.exit(0);
 }
 
@@ -65,7 +65,4 @@ assertions.push({
   evidence: hasDataRow ? undefined : 'No data rows found after the header',
 });
 
-const passed = assertions.filter((a) => a.passed).length;
-const score = passed / assertions.length;
-
-console.log(JSON.stringify({ score, assertions }));
+console.log(JSON.stringify({ assertions }));

examples/showcase/cross-repo-sync/scripts/validate-sync.ts

@@ -26,10 +26,7 @@ defineCodeGrader(({ fileChanges, config }) => {
 
   if (!fileChanges) {
     assertions.push({ text: 'No file changes captured', passed: false });
-    return {
-      score: 0,
-      assertions,
-    };
+    return { assertions };
   }
 
   // Parse diff blocks
@@ -57,12 +54,7 @@ defineCodeGrader(({ fileChanges, config }) => {
     }
   }
 
-  const passed = assertions.filter((a) => a.passed).length;
-  const total = assertions.length;
-  const score = total > 0 ? passed / total : 0;
-
   return {
-    score,
     assertions,
     details: {
       files_checked: expectedFiles.length,

packages/core/src/evaluation/graders/code-grader.ts

@@ -262,12 +262,6 @@ export class CodeGrader implements Grader {
       // Plain-text fallback: exit code is pass/fail, stdout is the assertion text.
       // For numeric scores or multi-aspect results, use the JSON protocol instead.
       const passed = exitCode === 0;
-      const score =
-        parsed != null
-          ? clampScore(typeof parsed.score === 'number' ? parsed.score : 0)
-          : passed
-            ? 1
-            : 0;
       const assertions: AssertionEntry[] =
         parsed != null && Array.isArray(parsed?.assertions)
           ? parsed.assertions
@@ -285,6 +279,19 @@ export class CodeGrader implements Grader {
           : parsed == null
             ? [{ text: stdout.trim() || (passed ? 'exit 0' : `exit ${exitCode}`), passed }]
             : [];
+      // When the script omits `score` but returns `assertions`, derive score as passing/total.
+      const score =
+        parsed != null
+          ? clampScore(
+              typeof parsed.score === 'number'
+                ? parsed.score
+                : assertions.length > 0
+                  ? assertions.filter((a) => a.passed).length / assertions.length
+                  : 0,
+            )
+          : passed
+            ? 1
+            : 0;
       // Capture optional structured details from code judge output
       const details =
         parsed?.details && typeof parsed.details === 'object' && !Array.isArray(parsed.details)

packages/core/test/evaluation/graders/code-grader-plain-text.test.ts

@@ -67,6 +67,34 @@ describe('code-grader plain-text fallback', () => {
     expect(result.assertions[0].text).toBe('ok');
   });
 
+  it('assertions without score → derived as passing/total', async () => {
+    const result = await grader(
+      `echo '{"assertions":[{"text":"a","passed":true},{"text":"b","passed":false},{"text":"c","passed":true}]}'`,
+    ).evaluate(ctx);
+    expect(result.score).toBeCloseTo(2 / 3);
+    expect(result.assertions).toHaveLength(3);
+  });
+
+  it('assertions all passing without score → score 1', async () => {
+    const result = await grader(
+      `echo '{"assertions":[{"text":"a","passed":true},{"text":"b","passed":true}]}'`,
+    ).evaluate(ctx);
+    expect(result.score).toBe(1);
+  });
+
+  it('assertions all failing without score → score 0', async () => {
+    const result = await grader(`echo '{"assertions":[{"text":"a","passed":false}]}'`).evaluate(
+      ctx,
+    );
+    expect(result.score).toBe(0);
+  });
+
+  it('empty assertions array without score → score 0', async () => {
+    const result = await grader(`echo '{"assertions":[]}'`).evaluate(ctx);
+    expect(result.score).toBe(0);
+    expect(result.assertions).toHaveLength(0);
+  });
+
   it('script with stderr on non-zero exit → surfaces as error assertion', async () => {
     const result = await grader('echo "bad" >&2; exit 1').evaluate(ctx);
     expect(result.score).toBe(0);