test: browse reference judge gates on its stated floors; the baseline comparison is recorded, not a ratchet (W7d browse judge baseline ratchet; t08 F7)

expect(regressions).toEqual([]) compared a noisy three-sample mean against
eval-baselines.json (clarity 4), silently raising the stated clarity floor of
3. The comparison stays in judge_reasoning; pass and assertions use
BROWSE_JUDGE_FLOORS (clarity 3, completeness 4, actionability 4). Judge
prompt and schema are unchanged. Stored panel means: three known-good (two the
ratchet rejected) pass; three below-floor panels fail.

test/browse-judge-floors.test.ts

@@ -0,0 +1,13 @@
+import { describe, expect, test } from 'bun:test';
+import * as fs from 'fs';
+import * as path from 'path';
+import { browseJudgeFloorsMet } from './helpers/workflow-judge-cache';
+
+type Scores = { clarity: number; completeness: number; actionability: number };
+const stored = JSON.parse(fs.readFileSync(path.join(import.meta.dir, 'fixtures/browse-judge/panel-means.json'), 'utf8')) as
+  { known_good: Record<string, Scores>; known_bad: Record<string, Scores> };
+
+describe('browse judge floors', () => {
+  test.each(Object.entries(stored.known_good))('passes %s', (_name, scores) => expect(browseJudgeFloorsMet(scores)).toBe(true));
+  test.each(Object.entries(stored.known_bad))('fails %s', (_name, scores) => expect(browseJudgeFloorsMet(scores)).toBe(false));
+});

test/fixtures/browse-judge/panel-means.json

@@ -0,0 +1,13 @@
+{
+  "_comment": "Browse reference judge panel means against baseline {clarity:4, completeness:4, actionability:4}. known_good passes the floors (the old ratchet failed the first two); known_bad fails them.",
+  "known_good": {
+    "clarity-dip-to-3.67": { "clarity": 3.67, "completeness": 4, "actionability": 4 },
+    "clarity-at-floor": { "clarity": 3, "completeness": 4.33, "actionability": 4 },
+    "at-baseline": { "clarity": 4, "completeness": 4, "actionability": 4 }
+  },
+  "known_bad": {
+    "clarity-below-floor": { "clarity": 2.67, "completeness": 4, "actionability": 4 },
+    "completeness-below-floor": { "clarity": 4, "completeness": 3.67, "actionability": 4 },
+    "actionability-below-floor": { "clarity": 4, "completeness": 4, "actionability": 3.33 }
+  }
+}

test/helpers/workflow-judge-cache.ts

@@ -135,3 +135,15 @@ export function prepareWorkflowJudgeCache(opts: WorkflowCacheOptions): {
     },
   };
 }
+
+/**
+ * Pass floors for the browse reference judge panel mean. The stored baseline
+ * is recorded for comparison only: a three-sample mean is too noisy for a
+ * no-dip ratchet, and gating on it silently raised the clarity floor to 4.
+ */
+export const BROWSE_JUDGE_FLOORS = { clarity: 3, completeness: 4, actionability: 4 } as const;
+
+export function browseJudgeFloorsMet(scores: Record<keyof typeof BROWSE_JUDGE_FLOORS, number>): boolean {
+  return (Object.keys(BROWSE_JUDGE_FLOORS) as Array<keyof typeof BROWSE_JUDGE_FLOORS>)
+    .every(dim => scores[dim] >= BROWSE_JUDGE_FLOORS[dim]);
+}

test/skill-llm-eval.test.ts

@@ -18,7 +18,7 @@ import { callJudge, judge, JudgeRefusalError, DEFAULT_JUDGE_MAX_TOKENS, judgePan
 import { ASK_QUESTIONS_HEADING, ENG_REVIEW_EXCERPT } from './helpers/workflow-excerpt';
 import type { JudgeScore } from './helpers/llm-judge';
 import { readWorkflowJudgeInput, buildWorkflowJudgePrompt, QA_DISCOVERY_REFERENCES, WORKFLOW_JUDGE_RESPONSE_SCHEMA, type WorkflowJudgeInput } from './helpers/workflow-judge-input';
-import { prepareWorkflowJudgeCache, validWorkflowJudgeScore } from './helpers/workflow-judge-cache';
+import { BROWSE_JUDGE_FLOORS, browseJudgeFloorsMet, prepareWorkflowJudgeCache, validWorkflowJudgeScore } from './helpers/workflow-judge-cache';
 import { buildCookieWorkflowJudgeInput, COOKIE_WORKFLOW_JUDGE } from './helpers/cookie-workflow-judge-input';
 import { getCookieWorkflowManualReview, type ManualJudgeReview } from './helpers/cookie-workflow-manual-review';
 import { resolveEvalModel } from '../lib/eval-model';
@@ -96,8 +96,8 @@ describeIfSelected('LLM-as-judge quality evals', [
     const t0 = Date.now();
     // Browse carve: snapshot flags + the full command list are the whole
     // generated section file; one judge grades the union. Scores are also
-    // pinned against test/fixtures/eval-baselines.json (UPDATE_BASELINES=1
-    // rewrites the pin).
+    // compared with test/fixtures/eval-baselines.json and the difference is
+    // recorded (UPDATE_BASELINES=1 rewrites it); only the floors gate.
     const section = sliceBrowseSection('## Snapshot Flags');
 
     const samples = await judgePanel(() => judge('browse skill reference (flags + commands)', section));
@@ -119,17 +119,16 @@ describeIfSelected('LLM-as-judge quality evals', [
       name: 'browse/SKILL.md reference',
       suite: 'LLM-as-judge quality evals',
       tier: 'llm-judge',
-      passed: scores.clarity >= 3 && scores.completeness >= 4 && scores.actionability >= 4 && regressions.length === 0,
+      passed: browseJudgeFloorsMet(scores),
       duration_ms: Date.now() - t0,
       cost_usd: 0.02 * samples.length,
       judge_scores: { clarity: scores.clarity, completeness: scores.completeness, actionability: scores.actionability },
       judge_reasoning: regressions.length ? `${judgePanelReasoning(samples)} | ${regressions.join('; ')}` : judgePanelReasoning(samples),
     });
 
-    expect(scores.clarity).toBeGreaterThanOrEqual(3);
-    expect(scores.completeness).toBeGreaterThanOrEqual(4);
-    expect(scores.actionability).toBeGreaterThanOrEqual(4);
-    expect(regressions).toEqual([]);
+    expect(scores.clarity).toBeGreaterThanOrEqual(BROWSE_JUDGE_FLOORS.clarity);
+    expect(scores.completeness).toBeGreaterThanOrEqual(BROWSE_JUDGE_FLOORS.completeness);
+    expect(scores.actionability).toBeGreaterThanOrEqual(BROWSE_JUDGE_FLOORS.actionability);
   }, JUDGE_MS);
 
   testIfSelected('setup block', async () => {