Add conditional operating-model judge

evals/fixtures/rubrics/operating-model-fit/clean.json

@@ -0,0 +1,36 @@
+{
+  "judge_response": {
+    "pass": true,
+    "routing": [
+      {
+        "unit_id": "cover",
+        "unit_type": "slide",
+        "treatment": "none",
+        "confidence": "high",
+        "reason": "The slide introduces the deck with one title."
+      },
+      {
+        "unit_id": "review-economics",
+        "unit_type": "slide",
+        "treatment": "relational",
+        "confidence": "high",
+        "reason": "The claim is a repeated-cost versus one-time-cost comparison."
+      },
+      {
+        "unit_id": "review-routing",
+        "unit_type": "slide",
+        "treatment": "operating-model",
+        "confidence": "high",
+        "reason": "The claim depends on policy-based routing."
+      },
+      {
+        "unit_id": "clinical-ide",
+        "unit_type": "slide",
+        "treatment": "simulated-surface",
+        "confidence": "high",
+        "reason": "The physician's interaction with evidence and hypotheses is the claim."
+      }
+    ],
+    "findings": []
+  }
+}

evals/fixtures/rubrics/operating-model-fit/fire.json

@@ -0,0 +1,23 @@
+{
+  "judge_response": {
+    "pass": false,
+    "routing": [
+      {
+        "unit_id": "review-routing",
+        "unit_type": "slide",
+        "treatment": "operating-model",
+        "confidence": "high",
+        "reason": "The claim depends on policy-based routing into distinct clinician queues."
+      }
+    ],
+    "findings": [
+      {
+        "check_id": "operating-model-fit",
+        "unit_id": "review-routing",
+        "unit_type": "slide",
+        "evidence": "Four equal cards name review jobs but show no selection rule, mandatory path, protected probability sample, or deduplication.",
+        "fix": "Render the universal checks and policy router, then separate mandatory, representative, calibration, and frontier lanes."
+      }
+    ]
+  }
+}

evals/fixtures/rubrics/operating-model-fit/routing-cases.jsonl

@@ -0,0 +1,16 @@
+{"id":"section-hero-single-claim","unit_type":"section","narrative_job":"Open a report with one proposition.","visible_claim":"Clinical reasoning is the scarce asset.","expected_treatment":"none","expected_applicable":false,"rationale":"A dominant claim does not need a relational model."}
+{"id":"section-quote","unit_type":"section","narrative_job":"Let one expert observation carry the argument.","visible_claim":"A physician quote with attribution.","expected_treatment":"none","expected_applicable":false,"rationale":"A quote is evidence, not a system."}
+{"id":"section-single-kpi","unit_type":"section","narrative_job":"Establish one baseline number.","visible_claim":"Universal review consumes five minutes per task.","expected_treatment":"none","expected_applicable":false,"rationale":"One number and one qualifier need hierarchy, not a diagram."}
+{"id":"section-faq","unit_type":"section","narrative_job":"Answer implementation questions in plain language.","visible_claim":"Four independent questions and answers.","expected_treatment":"none","expected_applicable":false,"rationale":"Independent answers remain a list."}
+{"id":"section-pricing-compare","unit_type":"section","narrative_job":"Compare internal annotation cost with external licensing value.","visible_claim":"Two price bases and their ratio.","expected_treatment":"relational","expected_applicable":true,"rationale":"Aligned quantities make the comparison inspectable."}
+{"id":"section-before-after","unit_type":"section","narrative_job":"Show how review allocation changes after launch.","visible_claim":"Universal stamping becomes sampled review.","expected_treatment":"relational","expected_applicable":true,"rationale":"A paired composition exposes the change without implying a process."}
+{"id":"section-benchmark-matrix","unit_type":"section","narrative_job":"Compare benchmark coverage and saturation.","visible_claim":"Tasks mapped against modalities and measured abilities.","expected_treatment":"relational","expected_applicable":true,"rationale":"A matrix is the faithful form for exact mappings."}
+{"id":"section-review-router","unit_type":"section","narrative_job":"Explain which work reaches clinicians and why.","visible_claim":"Policy routes cases into mandatory, representative, calibration, and frontier queues.","expected_treatment":"operating-model","expected_applicable":true,"rationale":"The selection rules and separate queues are the claim."}
+{"id":"section-state-transition","unit_type":"section","narrative_job":"Explain how a case moves from evidence collection to adjudication.","visible_claim":"Case states have explicit gates and abstention paths.","expected_treatment":"operating-model","expected_applicable":true,"rationale":"State and gate semantics must remain visible."}
+{"id":"section-causal-feedback","unit_type":"section","narrative_job":"Show how clinician labels improve future sampling.","visible_claim":"Errors update the judge, which changes the next frontier sample.","expected_treatment":"operating-model","expected_applicable":true,"rationale":"The feedback loop is causal, not a list of peers."}
+{"id":"section-resource-to-gate","unit_type":"section","narrative_job":"Connect clinician hours to evidence and release decisions.","visible_claim":"Annotation creates eval evidence that controls model release.","expected_treatment":"operating-model","expected_applicable":true,"rationale":"The resource-to-decision chain is the strategic argument."}
+{"id":"section-clinical-workspace","unit_type":"section","narrative_job":"Show how a physician investigates a multimodal case.","visible_claim":"A physician manipulates evidence, hypotheses, and next actions in one workspace.","expected_treatment":"simulated-surface","expected_applicable":true,"rationale":"The interaction model is more important than an abstract workflow."}
+{"id":"slide-cover","unit_type":"slide","narrative_job":"Name the argument.","visible_claim":"Evals are strategic IP.","expected_treatment":"none","expected_applicable":false,"rationale":"The cover needs one focal statement."}
+{"id":"slide-break-even","unit_type":"slide","narrative_job":"Make the review-cost threshold legible.","visible_claim":"Per-task cost times volume compared with benchmark investment.","expected_treatment":"relational","expected_applicable":true,"rationale":"An equation and paired totals are sufficient."}
+{"id":"slide-review-router","unit_type":"slide","narrative_job":"Explain the new clinician allocation policy.","visible_claim":"Only selected cases reach distinct expert queues.","expected_treatment":"operating-model","expected_applicable":true,"rationale":"Policy routing is the claim."}
+{"id":"slide-clinical-ide","unit_type":"slide","narrative_job":"Make the future training environment concrete.","visible_claim":"Physicians reason over multimodal evidence and probability-bearing hypotheses.","expected_treatment":"simulated-surface","expected_applicable":true,"rationale":"The product surface is the most faithful explanation."}

evals/rubric-coverage.json

@@ -1,12 +1,16 @@
 {
-  "note": "Tracks executable-fixture coverage for the 30 llm-pass/transcript catalog checks that evals/run.mjs cannot exercise (see plans/007-DESIGN-NOTE.md). Separate from evals/parity-allowlist.json, which is scoped to deterministic-stage (static-text/static-dom/browser) checks only.",
-  "total_llm_pass_and_transcript_checks": 30,
+  "note": "Tracks executable-fixture coverage for the 31 llm-pass/transcript catalog checks that evals/run.mjs cannot exercise (see plans/007-DESIGN-NOTE.md). Separate from evals/parity-allowlist.json, which is scoped to deterministic-stage (static-text/static-dom/browser) checks only. operating-model-fit also has a JSONL routing set covering slide and long-form section decisions.",
+  "behavioral_fixture_sets": {
+    "operating-model-fit": "evals/fixtures/rubrics/operating-model-fit/routing-cases.jsonl"
+  },
+  "total_llm_pass_and_transcript_checks": 31,
   "covered_by_evals_run_rubrics": [
     "text-visibly-clipped",
     "fixed-ui-obscures-content",
     "diagram-type-coherent",
     "preset-both-mode-visual",
     "deck-content-completeness",
+    "operating-model-fit",
     "reel-caption-matches-transcript"
   ],
   "remaining_uncovered": [

evals/run-rubrics.mjs

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-// Regression suite for the 30 llm-pass/transcript catalog checks that
+// Regression suite for the 31 llm-pass/transcript catalog checks that
 // evals/run.mjs cannot exercise (it filters to static-text/static-dom/browser
 // stages by design — see plans/007-DESIGN-NOTE.md for why).
 //
@@ -28,18 +28,25 @@
 import { readFileSync, readdirSync, existsSync } from 'node:fs';
 import { dirname, join, resolve } from 'node:path';
 import { fileURLToPath } from 'node:url';
+import { buildReport } from '../plugins/visual-explainer/scripts/verify/lib/report.mjs';
 
 const EVAL_ROOT = dirname(fileURLToPath(import.meta.url));
 const REPO_ROOT = resolve(EVAL_ROOT, '..');
 const RUBRICS_FIXTURES_ROOT = join(EVAL_ROOT, 'fixtures', 'rubrics');
+const OPERATING_MODEL_ROUTING_CASES = join(
+  RUBRICS_FIXTURES_ROOT,
+  'operating-model-fit',
+  'routing-cases.jsonl',
+);
 
 const checksCatalog = JSON.parse(
   readFileSync(resolve(REPO_ROOT, 'plugins/visual-explainer/scripts/verify/checks.json'), 'utf8'),
 ).checks;
 
 const severityById = new Map(checksCatalog.map((check) => [check.id, check.severity]));
 
-// The first implementation slice from plan 007 step 3: 6 checks, the
+// The first implementation slice from plan 007 step 3, extended with the
+// operating-model routing contract: 7 checks. The
 // 13 error-severity uncovered checks are the priority pool this was drawn
 // from. `mode: 'stub'` checks are asserted via deriveStatus() against a
 // canned judge_response fixture; `mode: 'algorithm'` runs the real scorer.
@@ -49,6 +56,7 @@ const SLICE = [
   { id: 'diagram-type-coherent', mode: 'stub' },
   { id: 'preset-both-mode-visual', mode: 'stub' },
   { id: 'deck-content-completeness', mode: 'stub' },
+  { id: 'operating-model-fit', mode: 'stub' },
   { id: 'reel-caption-matches-transcript', mode: 'algorithm' },
 ];
 
@@ -154,6 +162,118 @@ function runTranscriptCheck(id) {
   return rows;
 }
 
+// --- Operating-model contract: routing set + section pass selection --------
+
+function readJsonl(path) {
+  return readFileSync(path, 'utf8')
+    .split('\n')
+    .map((line) => line.trim())
+    .filter(Boolean)
+    .map((line, index) => {
+      try {
+        return JSON.parse(line);
+      } catch (error) {
+        throw new Error(`${path}:${index + 1}: ${error.message}`);
+      }
+    });
+}
+
+function runOperatingModelRoutingCases() {
+  const allowedTreatments = new Set([
+    'none',
+    'relational',
+    'operating-model',
+    'simulated-surface',
+  ]);
+  const allowedUnitTypes = new Set(['slide', 'section']);
+  const cases = readJsonl(OPERATING_MODEL_ROUTING_CASES);
+  const seenIds = new Set();
+  const rows = [];
+
+  for (const fixture of cases) {
+    const errors = [];
+    if (!fixture.id || seenIds.has(fixture.id)) errors.push('id must be present and unique');
+    if (!allowedUnitTypes.has(fixture.unit_type)) errors.push('unit_type must be slide or section');
+    if (!allowedTreatments.has(fixture.expected_treatment)) errors.push('invalid expected_treatment');
+    if (typeof fixture.expected_applicable !== 'boolean') errors.push('expected_applicable must be boolean');
+    if (fixture.expected_applicable !== (fixture.expected_treatment !== 'none')) {
+      errors.push('expected_applicable must be false only for treatment=none');
+    }
+    if (!fixture.narrative_job || !fixture.visible_claim || !fixture.rationale) {
+      errors.push('narrative_job, visible_claim, and rationale are required');
+    }
+    seenIds.add(fixture.id);
+    rows.push({
+      id: 'operating-model-routing-case',
+      caseName: fixture.id || 'missing-id',
+      status: errors.length ? 'FAIL' : 'pass',
+      message: errors.length
+        ? errors.join('; ')
+        : `${fixture.unit_type} -> ${fixture.expected_treatment}`,
+    });
+  }
+
+  const coverageAssertions = [
+    ['covers-slide-units', cases.some((fixture) => fixture.unit_type === 'slide')],
+    ['covers-section-units', cases.some((fixture) => fixture.unit_type === 'section')],
+    ...Array.from(allowedTreatments, (treatment) => [
+      `covers-${treatment}`,
+      cases.some((fixture) => fixture.expected_treatment === treatment),
+    ]),
+  ];
+  for (const [caseName, pass] of coverageAssertions) {
+    rows.push({
+      id: 'operating-model-routing-coverage',
+      caseName,
+      status: pass ? 'pass' : 'FAIL',
+      message: pass ? 'covered' : 'missing',
+    });
+  }
+  return rows;
+}
+
+function runOperatingModelPassSelectionCases() {
+  const operatingModelCheck = {
+    id: 'operating-model-fit',
+    stage: 'llm-pass',
+    status: 'llm-required',
+    severity: 'warn',
+  };
+  const cases = [
+    {
+      name: 'sectioned-page',
+      ctx: { profile: 'page', preset: 'custom', presetHint: '', html: '<main><section>Case review</section></main>', filePath: 'sectioned.html' },
+      expected: true,
+    },
+    {
+      name: 'simple-page',
+      ctx: { profile: 'page', preset: 'custom', presetHint: '', html: '<main><div>One statement</div></main>', filePath: 'simple.html' },
+      expected: false,
+    },
+    {
+      name: 'vertical-deck',
+      ctx: { profile: 'slides', preset: 'custom', presetHint: '', html: '<main></main>', filePath: 'slides.html' },
+      expected: true,
+    },
+    {
+      name: 'presentation-stage',
+      ctx: { profile: 'page', preset: 'custom', presetHint: '', html: '<main data-ve-presentation></main>', filePath: 'stage.html' },
+      expected: true,
+    },
+  ];
+
+  return cases.map(({ name, ctx, expected }) => {
+    const report = buildReport(ctx, [operatingModelCheck]);
+    const actual = report.llm_passes_required.includes('operating-model');
+    return {
+      id: 'operating-model-pass-selection',
+      caseName: name,
+      status: actual === expected ? 'pass' : 'FAIL',
+      message: `expected=${expected} actual=${actual}`,
+    };
+  });
+}
+
 // --- Driver ------------------------------------------------------------
 
 const catalogIds = new Set(checksCatalog.map((check) => check.id));
@@ -168,6 +288,8 @@ for (const entry of SLICE) {
   const rows = entry.mode === 'algorithm' ? runTranscriptCheck(entry.id) : runStubCheck(entry.id);
   allRows.push(...rows);
 }
+allRows.push(...runOperatingModelRoutingCases());
+allRows.push(...runOperatingModelPassSelectionCases());
 
 console.log('check_id,case,status,detail');
 for (const row of allRows) {
@@ -184,4 +306,4 @@ if (failures.length) {
 }
 
 const checksCovered = new Set(allRows.map((row) => row.id)).size;
-console.log(`\nAll ${allRows.length} assertions passed across ${checksCovered} checks (fire + no-fire each).`);
+console.log(`\nAll ${allRows.length} assertions passed across ${checksCovered} check families.`);

package.json

@@ -21,14 +21,14 @@
     "ve:eval-rubrics": "node evals/run-rubrics.mjs",
     "check:manifests": "node scripts/check-manifests.mjs",
     "ve:learn": "node scripts/ve-mdx/learn.mjs",
-    "test": "node --test visual-explainer-mdx/diagram-layout.test.mjs visual-explainer-mdx/presentation-core.test.mjs scripts/ve-mdx/design-systems.test.mjs scripts/ve-mdx/learn-extractors.test.mjs"
+    "test": "node --test visual-explainer-mdx/diagram-layout.test.mjs visual-explainer-mdx/presentation-core.test.mjs scripts/ve-mdx/design-systems.test.mjs scripts/ve-mdx/export-react-singleton.test.mjs scripts/ve-mdx/learn-extractors.test.mjs"
   },
   "dependencies": {
     "@mdx-js/rollup": "^3.1.1",
     "@tailwindcss/vite": "^4.3.0",
     "@vitejs/plugin-react": "^6.0.1",
-    "react": "^19.2.6",
-    "react-dom": "^19.2.6",
+    "react": "19.2.7",
+    "react-dom": "19.2.7",
     "tailwindcss": "^4.3.0",
     "typescript": "^6.0.3",
     "vite": "^8.0.12"

plugins/visual-explainer/.claude/agents/ve-verifier-operating-model.md

@@ -0,0 +1,30 @@
+---
+name: ve-verifier-operating-model
+description: Route each slide or coherent page section to the minimum explanatory treatment and judge operating-model fidelity only where applicable. Returns only verdict JSON.
+tools: Read
+---
+
+# ve-verifier-operating-model
+
+Run one verification pass. Do not fix files.
+
+## Read
+
+1. `plugins/visual-explainer/scripts/verify/rubrics/pass-operating-model.md`
+2. One screenshot per review unit: slide, presented stage, or coherent page section.
+3. The review map supplied by the orchestrator: unit id, unit type, visible title, and one-sentence narrative job.
+4. Only the brief excerpts needed to resolve a unit's intended decision.
+
+Route every unit before judging it. Do not penalize `none` units for lacking a
+diagram. Do not infer intent from implementation source.
+
+## Output
+
+Return only one valid JSON object matching the rubric schema:
+
+```json
+{"pass":true,"routing":[],"findings":[]}
+```
+
+Do not include prose, markdown, code fences, summaries, or recommendations
+outside `routing` and `findings`.