diff --git a/evals/fixtures/rubrics/operating-model-fit/clean.json b/evals/fixtures/rubrics/operating-model-fit/clean.json
new file mode 100644
index 0000000..01cb437
--- /dev/null
+++ b/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": []
+  }
+}
diff --git a/evals/fixtures/rubrics/operating-model-fit/fire.json b/evals/fixtures/rubrics/operating-model-fit/fire.json
new file mode 100644
index 0000000..652accb
--- /dev/null
+++ b/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."
+      }
+    ]
+  }
+}
diff --git a/evals/fixtures/rubrics/operating-model-fit/routing-cases.jsonl b/evals/fixtures/rubrics/operating-model-fit/routing-cases.jsonl
new file mode 100644
index 0000000..e3ef33a
--- /dev/null
+++ b/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."}
diff --git a/evals/rubric-coverage.json b/evals/rubric-coverage.json
index 9677a05..a2bbee1 100644
--- a/evals/rubric-coverage.json
+++ b/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": [
diff --git a/evals/run-rubrics.mjs b/evals/run-rubrics.mjs
index 770ee76..07a12de 100644
--- a/evals/run-rubrics.mjs
+++ b/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).
 //
@@ -28,10 +28,16 @@
 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'),
@@ -39,7 +45,8 @@ const checksCatalog = JSON.parse(
 
 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.`);
diff --git a/package-lock.json b/package-lock.json
index 9e218d4..f94a407 100644
--- a/package-lock.json
+++ b/package-lock.json
@@ -12,8 +12,8 @@
         "@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"
@@ -3293,24 +3293,24 @@
       }
     },
     "node_modules/react": {
-      "version": "19.2.6",
-      "resolved": "https://registry.npmjs.org/react/-/react-19.2.6.tgz",
-      "integrity": "sha512-sfWGGfavi0xr8Pg0sVsyHMAOziVYKgPLNrS7ig+ivMNb3wbCBw3KxtflsGBAwD3gYQlE/AEZsTLgToRrSCjb0Q==",
+      "version": "19.2.7",
+      "resolved": "https://registry.npmjs.org/react/-/react-19.2.7.tgz",
+      "integrity": "sha512-HNe9WslTbXmFK8o8cmwgAeJFSBvt1bPdHCVKtaaV+WlAN36mpT4hcRpwbf3fY56ar2oIXzsBpOAiIRHAdY0OlQ==",
       "license": "MIT",
       "engines": {
         "node": ">=0.10.0"
       }
     },
     "node_modules/react-dom": {
-      "version": "19.2.6",
-      "resolved": "https://registry.npmjs.org/react-dom/-/react-dom-19.2.6.tgz",
-      "integrity": "sha512-0prMI+hvBbPjsWnxDLxlCGyM8PN6UuWjEUCYmZhO67xIV9Xasa/r/vDnq+Xyq4Lo27g8QSbO5YzARu0D1Sps3g==",
+      "version": "19.2.7",
+      "resolved": "https://registry.npmjs.org/react-dom/-/react-dom-19.2.7.tgz",
+      "integrity": "sha512-t0BRVXvbiE/o20Hfw669rLbMCDWtYZLvmJigy2f0MxsXF+71pxhR3xOkspmsO8h3ZlNzyibAmtCa3l4lYKk6gQ==",
       "license": "MIT",
       "dependencies": {
         "scheduler": "^0.27.0"
       },
       "peerDependencies": {
-        "react": "^19.2.6"
+        "react": "^19.2.7"
       }
     },
     "node_modules/recma-build-jsx": {
diff --git a/package.json b/package.json
index 22e1a0d..8d27840 100644
--- a/package.json
+++ b/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"
diff --git a/plugins/visual-explainer/.claude/agents/ve-verifier-operating-model.md b/plugins/visual-explainer/.claude/agents/ve-verifier-operating-model.md
new file mode 100644
index 0000000..1b49dab
--- /dev/null
+++ b/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`.
diff --git a/plugins/visual-explainer/references/verification.md b/plugins/visual-explainer/references/verification.md
index 99c2892..64e45a0 100644
--- a/plugins/visual-explainer/references/verification.md
+++ b/plugins/visual-explainer/references/verification.md
@@ -58,7 +58,8 @@ In single-agent mode, YOU produce the declared inputs before running each rubric
 
 2. Build the P-completeness source-vs-render inventory from the source material and the rendered artifact: headings, bullets, table rows, cards, collapsible details, footnotes, and demo-frame summaries. Do not use screenshots for this pass.
 3. Capture P-diagram inputs one figure at a time. Scroll to each element carrying `data-diagram-role` or `.mermaid`; screenshot that element's bounding region; extract its visible labels; write a one-line content brief for the figure.
-4. Use candidate element lists from each deterministic check's `evidence` and `where` fields when a pass asks for candidate extracts.
+4. For P-operating-model, build a review map with one row per slide or coherent long-form section: stable unit id, unit type (`slide` or `section`), visible title, one-sentence narrative job, and screenshot path. If the source has no narrative job, use the visible claim and mark the route low-confidence. Do not use implementation source to infer intent.
+5. Use candidate element lists from each deterministic check's `evidence` and `where` fields when a pass asks for candidate extracts.
 
 ## 2. Run LLM Passes
 
@@ -86,6 +87,7 @@ Use these passes:
 | P-layout | `llm_passes_required` includes `hierarchy`, or any layout candidate exists | `report.json`; the 4 standard screenshots; candidate lists for clipping, fixed chrome, div-grid tables, slide screenshots, demo frames when referenced | `{{skill_dir}}/scripts/verify/rubrics/pass-layout.md` |
 | P-aesthetic | `llm_passes_required` includes any `aesthetic-*` token, a preset is detected or declared, or any preset candidate exists | `report.json`; active preset name; light/dark screenshots; candidate extracts named in the report | `{{skill_dir}}/scripts/verify/rubrics/pass-aesthetic.md` |
 | P-diagram | diagrams are present | `report.json`; one screenshot per figure; extracted diagram labels; one-line content brief for each figure | `{{skill_dir}}/scripts/verify/rubrics/pass-diagram.md` |
+| P-operating-model | `llm_passes_required` includes `operating-model` | one screenshot per review unit; review map with unit id, unit type, visible title, and one-sentence narrative job; only the brief excerpts needed to resolve intent | `{{skill_dir}}/scripts/verify/rubrics/pass-operating-model.md` |
 | P-completeness | source material or demo evidence exists | source inventory; extracted rendered headings, bullets, table rows, cards, and demo-frame summary; no page screenshots | `{{skill_dir}}/scripts/verify/rubrics/pass-completeness.md` |
 | P-copy | extracted prose exists | excluded-filtered prose text only | `{{skill_dir}}/scripts/verify/rubrics/pass-copy.md` |
 | P-visual-tells | `llm_passes_required` includes `visual-tells` (page/slides/magazine) | `report.json`; light/dark screenshots; candidate extracts | `{{skill_dir}}/scripts/verify/rubrics/pass-visual-tells.md` |
@@ -103,6 +105,20 @@ Load only the active preset section in `pass-aesthetic.md`. Do not judge Nothing
 
 Run only for rendered figures. Use per-figure screenshots, not one full-page screenshot. Apply the diagram Removal Test before emit; after emit, inspect screenshots. If a hand-authored SVG still fails after 2 repair attempts, replace the figure with Mermaid instead of continuing manual repair.
 
+### P-operating-model Questions
+
+Route every slide or coherent page section before judging it:
+
+- `none`: cover, divider, quote, image, agenda, simple CTA, single statistic, or one claim with one supporting fact;
+- `relational`: comparison, before/after, formula, aligned evidence, or exact mappings;
+- `operating-model`: sequence, routing, state, causality, provenance, uncertainty, dependencies, feedback, or resource flow; and
+- `simulated-surface`: the workflow or interface itself is the claim.
+
+Apply model-fidelity questions only to the final three routes. Do not require a
+diagram when a table, formula, paired contrast, or aligned rows tell the truth.
+Return a finding only when the chosen or missing structure materially weakens
+or misstates the unit's narrative job.
+
 ### P-completeness Questions
 
 Compare source inventory to extracted rendered content. Do not use screenshots. Missing source sections, decision cards, table rows, collapsible details, footnotes, or demonstrably low-value demo embeds fail this pass.
diff --git a/plugins/visual-explainer/scripts/verify/checks.json b/plugins/visual-explainer/scripts/verify/checks.json
index 12ef184..2db0067 100644
--- a/plugins/visual-explainer/scripts/verify/checks.json
+++ b/plugins/visual-explainer/scripts/verify/checks.json
@@ -1,7 +1,7 @@
 {
   "version": 2,
   "generated": "2026-07-04",
-  "note": "v2: +37 AI-tell/copy checks from impeccable+unslop distillation. Stage counts: {\"static-text\":70,\"static-dom\":69,\"llm-pass\":28,\"browser\":37,\"transcript\":2}",
+  "note": "v2: +37 AI-tell/copy checks from impeccable+unslop distillation. Stage counts: {\"static-text\":70,\"static-dom\":69,\"llm-pass\":29,\"browser\":37,\"transcript\":2}",
   "checks": [
     {
       "family": "F1_slop_global",
@@ -2450,6 +2450,27 @@
         "presentation-focal-point-and-contrast"
       ]
     },
+    {
+      "family": "F5_slides",
+      "id": "operating-model-fit",
+      "title": "Review units use the minimum faithful explanatory treatment",
+      "profiles": [
+        "page",
+        "slides",
+        "magazine"
+      ],
+      "stage": "llm-pass",
+      "severity": "warn",
+      "applies_when": "The artifact is a vertical deck, magazine deck, fixed-stage PresentationDeck, or long-form page with coherent sections. The pass routes every slide or section before judging model fidelity.",
+      "spec": "pass: pass-operating-model. First route each review unit to none, relational, operating-model, or simulated-surface from its narrative job. Judge only relational, operating-model, and simulated-surface units. Warn when the composition materially hides or misstates a required comparison, sequence, routing policy, state transition, causal/evidence relationship, uncertainty, dependency, feedback loop, or resource flow; also warn when an elaborate diagram is imposed on a unit that needs only a table, sentence, quote, image, or single number.",
+      "fixture_violation": "A clinician-routing policy rendered as four equal cards with no visible selection rule or separation between mandatory review, representative measurement, calibration, and frontier discovery.",
+      "fp_guards": "Do not require this treatment on covers, heroes, dividers, quotes, full-bleed images, agendas, FAQs, simple CTAs, single statistics, or one-claim/one-fact units. Lists and genuine peer sets may remain cards. A table, formula, or paired contrast may satisfy a relational unit. Route vague or underspecified units downward rather than inventing an operating model.",
+      "source_rules": [
+        "operating-model-first",
+        "layout-is-a-domain-claim",
+        "minimum-faithful-treatment"
+      ]
+    },
     {
       "family": "F5_slides",
       "id": "composition-variety-visual",
diff --git a/plugins/visual-explainer/scripts/verify/lib/report.mjs b/plugins/visual-explainer/scripts/verify/lib/report.mjs
index 4cefe8f..0436820 100644
--- a/plugins/visual-explainer/scripts/verify/lib/report.mjs
+++ b/plugins/visual-explainer/scripts/verify/lib/report.mjs
@@ -49,5 +49,19 @@ function llmPassesFor(ctx, checks) {
     required.add('copy');
     if (['page', 'slides', 'magazine'].includes(ctx.profile)) required.add('visual-tells');
   }
+  const hasReviewUnits =
+    ['slides', 'magazine'].includes(ctx.profile) ||
+    /data-ve-presentation/i.test(ctx.html) ||
+    /<section(?:\s|>)/i.test(ctx.html);
+  if (
+    hasReviewUnits &&
+    checks.some(
+      (check) =>
+        check.id === 'operating-model-fit' &&
+        check.status === 'llm-required',
+    )
+  ) {
+    required.add('operating-model');
+  }
   return Array.from(required);
 }
diff --git a/plugins/visual-explainer/scripts/verify/rubrics/pass-operating-model.md b/plugins/visual-explainer/scripts/verify/rubrics/pass-operating-model.md
new file mode 100644
index 0000000..596e0aa
--- /dev/null
+++ b/plugins/visual-explainer/scripts/verify/rubrics/pass-operating-model.md
@@ -0,0 +1,54 @@
+Inputs:
+- One screenshot per review unit: a slide, presented stage, or coherent long-form page section.
+- A review map containing, for each unit: stable unit id, unit type (`slide` or `section`), visible title, and one-sentence narrative job. If no narrative job exists, use the visible claim and mark routing confidence `low`.
+- Optional source brief excerpts that define a unit's intended decision. Do not read implementation source.
+
+Route before judging:
+
+Assign exactly one treatment to every review unit.
+
+- `none`: The unit is a cover, hero, divider, quote, full-bleed image, agenda, FAQ, simple CTA, single statistic, or one claim with one supporting fact. It does not need a relational model.
+- `relational`: The claim depends on comparison, before/after, a formula, aligned evidence, or a small set of exact mappings. A table, paired composition, or annotated equation may be sufficient.
+- `operating-model`: The claim depends on sequence, routing, state transitions, causality, provenance, uncertainty or probability, dependencies, feedback, or a resource-to-output flow.
+- `simulated-surface`: The workflow or interface itself is the claim, so a product/environment view is more faithful than an abstract diagram. Use this rarely.
+
+Routing rules:
+
+- Ask whether the viewer must understand relationships among multiple elements to accept the unit's claim. If no, route `none`.
+- Do not escalate because a unit contains three or more items. Lists and genuine peer sets may remain cards or rows.
+- Do not require a diagram when a table, formula, or paired contrast is more faithful.
+- Do not infer an elaborate system from a vague topic title. When intent is missing, prefer the lower treatment and mark confidence `low`.
+- The treatment names explanatory depth, not visual ambition. A well-made `none` unit is not lesser work.
+
+Questions for units routed `relational`, `operating-model`, or `simulated-surface`:
+
+- [operating-model-fit] Does the composition faithfully expose the relationships the viewer needs for the unit's narrative job?
+- Does the visual grammar make a truthful claim? Cards imply independent peers; a pipeline implies ordered transformation; a funnel implies attrition; a tree implies hierarchy or prerequisites; a graph implies meaningful relationships; a control plane implies policy routing; a trace implies sequence and provenance.
+- Can the viewer reconstruct the relevant comparison, sequence, routing rule, state change, evidence path, or dependency without presenter narration?
+- Does each load-bearing position, connector, line style, size, color, or animation have one stable meaning?
+- When uncertainty, abstention, residual probability, or an unsupported state is material, does the unit keep it visible?
+- Is the treatment proportionate? Flag both generic-card flattening and a full systems diagram used for a claim that only needs a table or sentence.
+
+Failure threshold:
+
+Return a finding only when the selected or missing structure materially weakens or misstates the unit's claim. Do not fail a unit merely because it has no diagram, uses cards, or chooses a lower treatment than another unit.
+
+Verdict JSON schema:
+
+`{"pass":true,"routing":[{"unit_id":"s01","unit_type":"slide|section","treatment":"none|relational|operating-model|simulated-surface","confidence":"low|medium|high","reason":"..."}],"findings":[{"check_id":"operating-model-fit","unit_id":"s04","unit_type":"slide|section","evidence":"...","fix":"..."}]}`
+
+Pass examples:
+
+- A cover slide routes `none` and uses one dominant title.
+- A report hero or FAQ section routes `none`; neither needs a system diagram.
+- A benchmark-matrix section routes `relational` and uses aligned rows.
+- A review-cost slide routes `relational` and shows the repeated unit, equation, and break-even without a systems diagram.
+- A clinician-review slide routes `operating-model` and shows the policy router separating mandatory review, representative sampling, calibration, and frontier queues.
+- A clinical IDE slide routes `simulated-surface` because the physician's interaction with evidence, hypotheses, and next actions is the claim.
+
+Fail examples:
+
+- A routing policy appears as four equal cards, leaving no selection rule or separation between mandatory review and sampled measurement.
+- A causal feedback loop is drawn as a one-way pipeline.
+- A probability-bearing diagnostic graph presents named diagnoses without residual uncertainty.
+- A quote slide is forced into a node graph even though the relationship adds no information.
diff --git a/scripts/ve-mdx/export-react-singleton.test.mjs b/scripts/ve-mdx/export-react-singleton.test.mjs
new file mode 100644
index 0000000..e775e4e
--- /dev/null
+++ b/scripts/ve-mdx/export-react-singleton.test.mjs
@@ -0,0 +1,58 @@
+import test from 'node:test';
+import assert from 'node:assert/strict';
+import {
+  cpSync,
+  mkdirSync,
+  mkdtempSync,
+  readFileSync,
+  rmSync,
+  writeFileSync,
+} from 'node:fs';
+import os from 'node:os';
+import path from 'node:path';
+import { spawnSync } from 'node:child_process';
+import { fileURLToPath } from 'node:url';
+
+const REPO_ROOT = path.resolve(path.dirname(fileURLToPath(import.meta.url)), '../..');
+
+test('client export dedupes React when the source lives in another package tree', () => {
+  const externalRoot = mkdtempSync(path.join(os.tmpdir(), 'artifacture-react-singleton-'));
+  try {
+    mkdirSync(path.join(externalRoot, 'node_modules'), { recursive: true });
+    cpSync(
+      path.join(REPO_ROOT, 'node_modules/react'),
+      path.join(externalRoot, 'node_modules/react'),
+      { recursive: true },
+    );
+
+    const source = path.join(externalRoot, 'external-deck.tsx');
+    const output = path.join(externalRoot, 'external-deck.html');
+    writeFileSync(
+      source,
+      `import React, { useState } from 'react';
+
+export default function ExternalDeck() {
+  const [slide] = useState('one');
+  return <main data-ve-presentation data-slide-id={slide}>External deck</main>;
+}
+`,
+    );
+
+    const result = spawnSync(
+      process.execPath,
+      ['scripts/ve-mdx/export.mjs', source, '--out', output, '--draft'],
+      { cwd: REPO_ROOT, encoding: 'utf8' },
+    );
+    assert.equal(result.status, 0, `${result.stdout}\n${result.stderr}`);
+
+    const html = readFileSync(output, 'utf8');
+    const hookImplementations = html.match(/useState=function/g) ?? [];
+    assert.equal(
+      hookImplementations.length,
+      1,
+      `expected one React hook implementation, found ${hookImplementations.length}`,
+    );
+  } finally {
+    rmSync(externalRoot, { recursive: true, force: true });
+  }
+});
diff --git a/scripts/ve-mdx/export.mjs b/scripts/ve-mdx/export.mjs
index 16b1a59..ff7b4af 100644
--- a/scripts/ve-mdx/export.mjs
+++ b/scripts/ve-mdx/export.mjs
@@ -97,6 +97,13 @@ async function main() {
       base: './',
       logLevel: 'warn',
       plugins: [veMdxPreflightPlugin(source, draft), mdx(), react(), tailwindcss()],
+      // A source outside the Artifacture package can have its own physical
+      // React install. Without dedupe, the generated bundle renders with
+      // Artifacture's ReactDOM while hooks import the source package's React,
+      // leaving that second copy's dispatcher null at runtime.
+      resolve: {
+        dedupe: ['react', 'react-dom', 'react/jsx-runtime', 'react/jsx-dev-runtime'],
+      },
       server: {
         fs: {
           allow: [repoRoot, tmp],