test(evals): keep the review judge at default effort; only lift its output cap

Truncation was the defect: thinking and the answer share max_tokens, and the
~64k-token review bundle once spent all 8192. The registration keeps default
effort and response format and gets ship's 64k streaming cap; the guard test
pins exactly that.

test/skill-llm-eval.test.ts

@@ -669,12 +669,11 @@ describeIfSelected('Other skill evals', [
       judgeContext: 'a pre-landing review with bounded exploratory QA',
       judgeGoal: 'how to review and explore changed behavior even for small diffs without a plan or server, preserve report-only discovery and the test_stub ASK gate, handle incomplete probes honestly, and rerun affected evidence after approved repairs',
       agentCapability: 'frontier',
-      // This ~64k-token bundle is the largest workflow input. At default effort, thinking reached 6.9k output
-      // tokens and 88 s in 24 measured samples, and a CI sample spent all 8192 tokens without an answer.
-      // Medium kept 24 of 24 under 5k tokens and 63 s with the same gating scores; the budget is no cap.
+      // This ~64k-token bundle is the largest workflow input. Thinking and the answer share max_tokens:
+      // at default effort, 24 measured samples used up to 6.9k output tokens and a CI sample spent all
+      // 8192 without an answer. The ship-sized streaming cap removes truncation; effort stays default.
       maxTokens: 65_536,
       stream: true,
-      effort: 'medium',
     });
   }, WORKFLOW_JUDGE_TEST_MS);
 

test/workflow-judge-cache.test.ts

@@ -399,13 +399,13 @@ test('Ship sends its authorized 64k cap and compact response contract through th
   expect(options.maxTokens).toBe(65_536);
   expect(options.stream).toBe(true);
   expect(options.effort).toBe('medium');
-  // Only measured registrations bound judge effort: ship and the ~64k-token review bundle.
-  expect(source.match(/^\s+effort: '/gm)).toHaveLength(2);
+  expect(source.match(/^\s+effort: '/gm)).toHaveLength(1);
+  // The ~64k-token review bundle gets only ship's output cap and streaming; its effort and response format stay default.
   const review = source.match(/testIfSelected\('review\/SKILL\.md workflow',[\s\S]*?await runWorkflowJudge\(\{([\s\S]*?)\n    \}\);/);
   expect(review).not.toBeNull();
   const reviewOptions = new Function('QA_DISCOVERY_REFERENCES', `return ({${review![1]}});`)(QA_DISCOVERY_REFERENCES);
   expect([reviewOptions.maxTokens, reviewOptions.stream, reviewOptions.effort, reviewOptions.schemaTransport, reviewOptions.compactReasoning])
-    .toEqual([65_536, true, 'medium', undefined, undefined]);
+    .toEqual([65_536, true, undefined, undefined, undefined]);
   expect(WORKFLOW_JUDGE_RESPONSE_SCHEMA.properties.reasoning).not.toHaveProperty('pattern');
   expect(WORKFLOW_JUDGE_RESPONSE_SCHEMA.properties.reasoning).not.toHaveProperty('maxLength');
   expect(source.match(/compactReasoning: true/g)).toHaveLength(1);