thumbgate 1.29.2 → 1.31.0

package/public/pricing.html

@@ -4,7 +4,7 @@
   <meta charset="utf-8">
   <meta name="viewport" content="width=device-width, initial-scale=1">
   <title>Pricing — Enforcement that pays for itself | ThumbGate</title>
-  <meta name="description" content="ThumbGate vs DIY prompts, platform hire, or GRC suite. Free evaluate, Pro $19/mo, $499 Diagnostic gate. Countable units that convert.">
+  <meta name="description" content="Choose ThumbGate by the value and scope of the agent risk: free evaluation, Pro for one operator, a $499 managed diagnostic for one workflow, or custom enterprise intake.">
   <link rel="canonical" href="__APP_ORIGIN__/pricing">
   <link rel="alternate" type="text/markdown" title="ThumbGate LLM context" href="__APP_ORIGIN__/llm-context.md">
   <meta property="og:title" content="ThumbGate Pricing — $499 Diagnostic · Pro $19/mo">
@@ -57,6 +57,19 @@
     .fit-card { border-top:3px solid var(--green); background:var(--card); padding:22px; }
     .fit-card h3 { margin:0 0 8px; }
     .fit-card p { margin:0; color:var(--muted); }
+    .value-equation { display:grid; grid-template-columns:1fr auto 1fr auto 1fr auto 1.3fr; gap:12px; align-items:stretch; margin-top:26px; }
+    .value-factor { display:grid; align-content:center; min-height:112px; border:1px solid var(--line); border-radius:14px; background:var(--card); padding:18px; }
+    .value-factor strong { font-size:1.06rem; }
+    .value-factor span { color:var(--muted); font-size:.83rem; }
+    .value-operator { display:grid; place-items:center; color:var(--green); font-size:1.7rem; font-weight:900; }
+    .value-result { border-color:var(--green); background:#e9f3ee; }
+    .segment-grid { display:grid; grid-template-columns:repeat(3,1fr); gap:14px; margin-top:28px; }
+    .segment-card { display:flex; flex-direction:column; border:1px solid var(--line); border-radius:14px; background:var(--card); padding:22px; }
+    .segment-card.recommended { border:2px solid var(--green); }
+    .segment-card h3 { margin:8px 0 6px; }
+    .segment-card p { margin:0 0 14px; color:var(--muted); font-size:.92rem; }
+    .segment-card .fence { min-height:58px; margin-top:auto; color:var(--ink); font-size:.82rem; }
+    .segment-card .button { margin-top:14px; }
     .faq { border-top:1px solid var(--line); }
     .faq-item { border-bottom:1px solid var(--line); }
     .faq-q { width:100%; border:0; background:transparent; padding:18px 0; color:var(--ink); font:inherit; font-weight:800; text-align:left; cursor:pointer; }
@@ -65,22 +78,52 @@
     footer { border-top:1px solid var(--line); padding:28px 0; color:var(--muted); font-size:.85rem; }
     .footer-inner { display:flex; justify-content:space-between; gap:24px; }
     .footer-links { display:flex; gap:18px; }
-    @media (max-width:820px) { .hero,.steps,.fit { grid-template-columns:1fr; } .hero { gap:34px; } .nav-links a:not(.nav-buy) { display:none; } }
+    @media (max-width:820px) { .hero,.steps,.fit,.segment-grid { grid-template-columns:1fr; } .value-equation { grid-template-columns:1fr; } .value-operator { min-height:24px; } .hero { gap:34px; } .nav-links a:not(.nav-buy) { display:none; } }
   </style>
   <script type="application/ld+json">
   {
     "@context": "https://schema.org",
-    "@type": "Service",
-    "name": "ThumbGate Enterprise Workflow Gate",
-    "description": "A bounded implementation for one supported local AI-agent workflow: review, configured pre-action gate, regression test, and rollout and rollback proof.",
-    "provider": { "@type": "Organization", "name": "ThumbGate", "url": "__APP_ORIGIN__/" },
-    "offers": {
-      "@type": "Offer",
-      "price": "499",
-      "priceCurrency": "USD",
-      "url": "__APP_ORIGIN__/pricing#buy",
-      "availability": "https://schema.org/InStock"
-    }
+    "@graph": [
+      {
+        "@type": "Service",
+        "name": "ThumbGate Managed Workflow Diagnostic",
+        "description": "A bounded implementation for one supported local AI-agent workflow: review, configured pre-action gate, regression test, and rollout and rollback proof.",
+        "provider": { "@type": "Organization", "name": "ThumbGate", "url": "__APP_ORIGIN__/" },
+        "audience": { "@type": "BusinessAudience", "audienceType": "Engineering and platform teams with a repeated AI-agent failure" },
+        "offers": {
+          "@type": "Offer",
+          "name": "Managed workflow diagnostic",
+          "price": "499",
+          "priceCurrency": "USD",
+          "url": "__APP_ORIGIN__/pricing#buy",
+          "availability": "https://schema.org/InStock"
+        }
+      },
+      {
+        "@type": "SoftwareApplication",
+        "name": "ThumbGate Pro",
+        "applicationCategory": "DeveloperApplication",
+        "operatingSystem": "macOS, Linux, Windows",
+        "offers": [
+          {
+            "@type": "Offer",
+            "name": "ThumbGate Pro monthly",
+            "price": "19",
+            "priceCurrency": "USD",
+            "url": "__APP_ORIGIN__/checkout/pro?plan_id=pro",
+            "availability": "https://schema.org/InStock"
+          },
+          {
+            "@type": "Offer",
+            "name": "ThumbGate Pro annual",
+            "price": "149",
+            "priceCurrency": "USD",
+            "url": "__APP_ORIGIN__/checkout/pro?plan_id=pro",
+            "availability": "https://schema.org/InStock"
+          }
+        ]
+      }
+    ]
   }
   </script>
   <script type="application/ld+json">
@@ -109,8 +152,8 @@
       <div class="nav-links">
         <a href="/guide">Technical setup</a>
         <a href="https://github.com/IgorGanapolsky/ThumbGate/blob/main/docs/VERIFICATION_EVIDENCE.md" target="_blank" rel="noopener">Proof</a>
-        <a class="nav-enterprise" href="#enterprise-gate" data-offer-link data-cta-id="pricing_nav_buy">Get Started · $499</a>
-        <a class="nav-pro" href="/checkout/pro?utm_source=pricing&amp;utm_medium=nav&amp;utm_campaign=rally_style_gtm&amp;cta_id=pricing_nav_pro&amp;plan_id=pro" data-offer-link data-cta-id="pricing_nav_pro">Pro · $19/mo</a>
+        <a class="nav-enterprise" href="#enterprise-gate" data-offer-link data-cta-id="pricing_nav_buy" data-plan-id="sprint_diagnostic" data-value="499" data-segment="workflow_team">Get Started · $499</a>
+        <a class="nav-pro" href="/checkout/pro?utm_source=pricing&amp;utm_medium=nav&amp;utm_campaign=value_packaging_v1&amp;cta_id=pricing_nav_pro&amp;plan_id=pro" data-offer-link data-cta-id="pricing_nav_pro" data-plan-id="pro" data-value="19" data-segment="solo_operator">Pro · $19/mo</a>
       </div>
     </div>
   </nav>
@@ -139,29 +182,74 @@
             <input type="hidden" name="plan_id" value="sprint_diagnostic">
             <input type="hidden" name="utm_source" value="pricing">
             <input type="hidden" name="utm_medium" value="direct_checkout">
-            <input type="hidden" name="utm_campaign" value="rally_style_gtm">
+            <input type="hidden" name="utm_campaign" value="value_packaging_v1">
+            <input id="pricing-buyer-segment" type="hidden" name="campaign_variant" value="workflow_team">
             <input type="hidden" name="cta_id" value="pricing_managed_gate_buy">
             <input type="hidden" name="cta_placement" value="pricing">
             <input type="hidden" name="landing_path" value="/pricing">
             <button class="button" type="submit">Get Started — $499 Diagnostic</button>
           </form>
           <p class="fine">Secure payment. We reply with scheduling and the workflow checklist.</p>
-          <div class="boundary">Detected secret exfiltration and gate-process bypass attempts deny by default. Other destructive actions warn by default and deny when strict enforcement is enabled.</div>
+          <div class="boundary">Detected secret exfiltration denies by default: literal secrets, secret-file paths in uploads/pipes/redirections/command-substitution, secret env vars to network tools, scp/rsync/cloud CLI uploads of credential files, and common interpreter one-liners. Not a full network sandbox. Other destructive actions warn by default and deny when strict enforcement is enabled.</div>
         </aside>
         <aside class="pro-card" id="pro" aria-label="ThumbGate Pro">
           <div class="eyebrow">Self-serve</div>
           <div class="price">$19<span style="font-size:1rem;font-weight:700">/mo</span></div>
           <p class="price-note">Or $149/yr. Individual operators. Cancel anytime.</p>
-          <a class="button" href="/checkout/pro?utm_source=pricing&amp;utm_medium=card&amp;utm_campaign=rally_style_gtm&amp;cta_id=pricing_pro_buy&amp;plan_id=pro" data-offer-link data-cta-id="pricing_pro_buy" style="width:100%;box-sizing:border-box">Start Pro — $19/mo</a>
+          <a class="button" href="/checkout/pro?utm_source=pricing&amp;utm_medium=card&amp;utm_campaign=value_packaging_v1&amp;cta_id=pricing_pro_buy&amp;plan_id=pro" data-offer-link data-cta-id="pricing_pro_buy" data-plan-id="pro" data-value="19" data-segment="solo_operator" style="width:100%;box-sizing:border-box">Start Pro</a>
           <p class="fine">Hosted checkout. Free local evaluate remains free after you subscribe.</p>
         </aside>
       </div>
     </div>
 
+    <section id="value">
+      <div class="eyebrow">Price from your exposure</div>
+      <h2>Value is the recovery work the next repeat does not create.</h2>
+      <p class="section-lede">Use your own incident history. ThumbGate does not claim a universal savings number.</p>
+      <div class="value-equation" role="img" aria-label="Repeat incidents multiplied by recovery hours and loaded hourly cost, plus downstream impact, equals avoidable recovery exposure">
+        <div class="value-factor"><strong>Repeat incidents</strong><span>per quarter</span></div>
+        <div class="value-operator">×</div>
+        <div class="value-factor"><strong>Recovery hours</strong><span>diagnosis, repair, review</span></div>
+        <div class="value-operator">×</div>
+        <div class="value-factor"><strong>Loaded hourly cost</strong><span>the people pulled into recovery</span></div>
+        <div class="value-operator">+</div>
+        <div class="value-factor value-result"><strong>Downstream impact</strong><span>delay, incident response, audit evidence</span></div>
+      </div>
+    </section>
+
+    <section id="segments">
+      <div class="eyebrow">Choose by buyer value</div>
+      <h2>One product. Three buying decisions.</h2>
+      <p class="section-lede">The fence is scope and service level—not an arbitrary discount.</p>
+      <div class="segment-grid">
+        <article class="segment-card">
+          <div class="eyebrow">Solo operator</div>
+          <h3>Evaluate free, then Pro</h3>
+          <p>Pro removes rule limits and adds personal recall, review, dashboard, and export workflows.</p>
+          <div class="fence"><strong>Fence:</strong> one operator, local-first, self-serve support. $19 monthly or $149 annual.</div>
+          <a class="button" href="/checkout/pro?utm_source=pricing&amp;utm_medium=segment_card&amp;utm_campaign=value_packaging_v1&amp;cta_id=pricing_segment_pro&amp;plan_id=pro" data-offer-link data-cta-id="pricing_segment_pro" data-plan-id="pro" data-value="19" data-segment="solo_operator">Choose Pro</a>
+        </article>
+        <article class="segment-card recommended">
+          <div class="eyebrow">Engineering or platform team</div>
+          <h3>Managed diagnostic</h3>
+          <p>Turn one expensive, repeated agent failure into a configured gate and a regression proof.</p>
+          <div class="fence"><strong>Fence:</strong> one supported local workflow, fixed scope, two-business-day delivery. $499 once.</div>
+          <a class="button" href="#enterprise-gate" data-offer-link data-cta-id="pricing_segment_diagnostic" data-plan-id="sprint_diagnostic" data-value="499" data-segment="workflow_team">Harden one workflow</a>
+        </article>
+        <article class="segment-card">
+          <div class="eyebrow">Regulated or multi-workflow team</div>
+          <h3>Enterprise service intake</h3>
+          <p>Scope approval boundaries, rollback evidence, and accountable rollout support after one workflow proves fit.</p>
+          <div class="fence"><strong>Fence:</strong> custom scope after intake. Hosted team features, SSO, and SIEM are not general availability.</div>
+          <a class="button" href="#enterprise-gate" data-offer-link data-cta-id="pricing_segment_enterprise" data-plan-id="enterprise_service" data-value="0" data-segment="regulated_team">Start with one diagnostic</a>
+        </article>
+      </div>
+    </section>
+
     <section>
       <div class="eyebrow">How ThumbGate stacks up</div>
       <h2>Make the alternative look expensive.</h2>
-      <p class="section-lede">Same move as high-converting SaaS pricing: DIY prompts cost nothing and stop nothing; platform hire and GRC suites cost months.</p>
+      <p class="section-lede">DIY prompts look cheap but still consume model and review time. Platform hiring and GRC suites require larger commitments.</p>
       <div style="overflow-x:auto;margin-top:22px;border:1px solid var(--line);border-radius:14px;background:var(--card);">
         <table style="width:100%;border-collapse:collapse;font-size:.92rem;min-width:620px;">
           <thead>
@@ -176,24 +264,24 @@
           <tbody>
             <tr>
               <td style="padding:12px 14px;border-bottom:1px solid var(--line);color:var(--muted);">Monthly cost</td>
-              <td style="padding:12px 14px;border-bottom:1px solid var(--line);">$0</td>
-              <td style="padding:12px 14px;border-bottom:1px solid var(--line);">$15k+ loaded</td>
-              <td style="padding:12px 14px;border-bottom:1px solid var(--line);">$$$$</td>
+              <td style="padding:12px 14px;border-bottom:1px solid var(--line);">Model + review time</td>
+              <td style="padding:12px 14px;border-bottom:1px solid var(--line);">Loaded engineering cost</td>
+              <td style="padding:12px 14px;border-bottom:1px solid var(--line);">Contract + implementation</td>
               <td style="padding:12px 14px;border-bottom:1px solid var(--line);font-weight:800;color:var(--green);">$0–$19 · $499 once</td>
             </tr>
             <tr>
               <td style="padding:12px 14px;border-bottom:1px solid var(--line);color:var(--muted);">Time to first gate</td>
-              <td style="padding:12px 14px;border-bottom:1px solid var(--line);">Never reliable</td>
+              <td style="padding:12px 14px;border-bottom:1px solid var(--line);">Manual upkeep</td>
               <td style="padding:12px 14px;border-bottom:1px solid var(--line);">Weeks</td>
               <td style="padding:12px 14px;border-bottom:1px solid var(--line);">Months</td>
               <td style="padding:12px 14px;border-bottom:1px solid var(--line);font-weight:800;">Minutes after install / 2 business days managed</td>
             </tr>
             <tr>
               <td style="padding:12px 14px;color:var(--muted);">When agent “looks competent”</td>
-              <td style="padding:12px 14px;">Ignores the prompt</td>
+              <td style="padding:12px 14px;">Prompt may be skipped</td>
               <td style="padding:12px 14px;">Custom glue</td>
               <td style="padding:12px 14px;">Gateway only</td>
-              <td style="padding:12px 14px;font-weight:800;">Hard block at tool call</td>
+              <td style="padding:12px 14px;font-weight:800;">Configured gate at tool call</td>
             </tr>
           </tbody>
         </table>
@@ -254,13 +342,31 @@
       } catch (_) {}
     }
 
-    function offerProps(ctaId, placement) {
-      return { ctaId: ctaId, ctaPlacement: placement, planId: 'sprint_diagnostic', value: 499, currency: 'USD' };
+    function offerProps(ctaId, placement, planId, value, segment) {
+      return {
+        ctaId: ctaId,
+        ctaPlacement: placement,
+        planId: planId,
+        value: Number(value || 0),
+        currency: 'USD',
+        segment: segment,
+        experimentId: 'value_packaging_v1'
+      };
     }
 
     document.querySelectorAll('[data-offer-link]').forEach(function (link) {
       link.addEventListener('click', function () {
-        const props = offerProps(link.dataset.ctaId || 'pricing_anchor', 'pricing_anchor');
+        if (link.getAttribute('href') === '#enterprise-gate') {
+          const buyerSegment = document.querySelector('#pricing-buyer-segment');
+          if (buyerSegment) buyerSegment.value = link.dataset.segment || 'workflow_team';
+        }
+        const props = offerProps(
+          link.dataset.ctaId || 'pricing_anchor',
+          'pricing_anchor',
+          link.dataset.planId || 'unknown',
+          link.dataset.value || '0',
+          link.dataset.segment || 'unknown'
+        );
         window.plausible('pricing_cta_click', { props: props });
         sendFirstPartyTelemetry('cta_click', props);
       });
@@ -269,7 +375,14 @@
     const checkoutForm = document.querySelector('[data-primary-checkout]');
     if (checkoutForm) {
       checkoutForm.addEventListener('submit', function () {
-        const props = offerProps('pricing_managed_gate_buy', 'pricing');
+        const buyerSegment = document.querySelector('#pricing-buyer-segment');
+        const props = offerProps(
+          'pricing_managed_gate_buy',
+          'pricing',
+          'sprint_diagnostic',
+          499,
+          buyerSegment ? buyerSegment.value : 'workflow_team'
+        );
         window.plausible('checkout_start', { props: props });
         sendFirstPartyTelemetry('checkout_start', props);
       });
@@ -283,7 +396,7 @@
       });
     });
 
-    sendFirstPartyTelemetry('pricing_page_view', { landingPath: '/pricing' });
+    sendFirstPartyTelemetry('pricing_page_view', { landingPath: '/pricing', experimentId: 'value_packaging_v1' });
   </script>
 </body>
 </html>

package/public/whitepaper.html

@@ -0,0 +1,189 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+<meta charset="UTF-8">
+<meta name="viewport" content="width=device-width, initial-scale=1.0">
+<meta name="generator" content="ThumbGate">
+<meta name="author" content="Igor Ganapolsky">
+<title>How We Know ThumbGate Works — Evaluation White Paper</title>
+<meta name="description" content="Evaluation white paper for ThumbGate: golden datasets, offline regression, tool-call correctness, latency, cost, human review, and production monitoring — with diagrams and re-run commands.">
+<meta property="og:title" content="How We Know ThumbGate Works">
+<meta property="og:description" content="Seven proof dimensions for a pre-action gate: goldens, offline tests, tool-call metrics, latency, cost, human review, production monitoring.">
+<meta property="og:type" content="article">
+<meta property="og:url" content="https://thumbgate.ai/whitepaper">
+<link rel="canonical" href="https://thumbgate.ai/whitepaper">
+<link rel="icon" type="image/png" href="/thumbgate-icon.png">
+<script defer data-domain="thumbgate.ai" src="https://plausible.io/js/script.js"></script>
+<script type="application/ld+json">
+{
+  "@context": "https://schema.org",
+  "@type": "TechArticle",
+  "headline": "How We Know ThumbGate Works",
+  "description": "Buyer-facing evaluation white paper covering golden datasets, offline regression, tool-call correctness, latency, cost, human review, and production monitoring.",
+  "url": "https://thumbgate.ai/whitepaper",
+  "dateModified": "2026-07-29",
+  "author": { "@type": "Person", "name": "Igor Ganapolsky", "url": "https://github.com/IgorGanapolsky" }
+}
+</script>
+<style>
+  :root {
+    --bg:#0b0f14; --panel:#111823; --border:#1e2a3a; --text:#e6edf3;
+    --muted:#8b98a5; --cyan:#39c5cf; --green:#3fb950; --red:#f85149;
+  }
+  * { margin:0; padding:0; box-sizing:border-box; }
+  body { background:var(--bg); color:var(--text); font-family:-apple-system,BlinkMacSystemFont,"Segoe UI",Roboto,sans-serif; line-height:1.65; }
+  nav { padding:1rem 2rem; border-bottom:1px solid var(--border); display:flex; gap:1.25rem; flex-wrap:wrap; align-items:center; }
+  nav a { color:var(--muted); text-decoration:none; font-size:0.9rem; }
+  nav a:hover { color:var(--cyan); }
+  nav .brand { color:var(--text); font-weight:700; }
+  .container { max-width:900px; margin:0 auto; padding:2.5rem 1.5rem 4rem; }
+  h1 { font-size:2rem; margin-bottom:0.4rem; }
+  h2 { font-size:1.3rem; margin:2.3rem 0 0.7rem; color:var(--cyan); }
+  h3 { font-size:1.05rem; margin:1.3rem 0 0.4rem; }
+  .subtitle { color:var(--muted); font-size:1.05rem; margin-bottom:1.25rem; }
+  p, li { margin:0.65rem 0; }
+  ul, ol { padding-left:1.3rem; }
+  table { width:100%; border-collapse:collapse; margin:1rem 0; font-size:0.92rem; }
+  th, td { text-align:left; padding:0.5rem 0.65rem; border-bottom:1px solid var(--border); vertical-align:top; }
+  th { color:var(--muted); }
+  code { background:#0d1420; border:1px solid var(--border); border-radius:4px; padding:0.1rem 0.35rem; font-size:0.86em; color:var(--cyan); }
+  pre { background:#0d1420; border:1px solid var(--border); border-radius:8px; padding:1rem; overflow-x:auto; font-size:0.85rem; margin:1rem 0; }
+  .panel { background:var(--panel); border:1px solid var(--border); border-radius:10px; padding:1.1rem 1.25rem; margin:1rem 0; }
+  .rule { border-left:3px solid var(--cyan); padding:0.65rem 1rem; background:var(--panel); margin:1.25rem 0; font-weight:600; }
+  figure { background:var(--panel); border:1px solid var(--border); border-radius:12px; padding:1rem; margin:1rem 0; }
+  figure img { width:100%; height:auto; display:block; border-radius:8px; }
+  figcaption { color:var(--muted); font-size:0.88rem; margin-top:0.65rem; }
+  a.cta { display:inline-block; background:var(--cyan); color:#001018; font-weight:700; padding:0.7rem 1rem; border-radius:8px; text-decoration:none; margin:0.4rem 0.6rem 0.4rem 0; }
+  a.inline { color:var(--cyan); }
+  .footer-note { color:var(--muted); font-size:0.85rem; margin-top:2.5rem; text-align:center; }
+  .footer-note a { color:var(--cyan); text-decoration:none; }
+  .good { color:var(--green); } .bad { color:var(--red); }
+</style>
+</head>
+<body>
+<nav>
+  <a class="brand" href="/">ThumbGate</a>
+  <a href="/whitepaper">White paper</a>
+  <a href="/architecture">Architecture</a>
+  <a href="/eval-scorecard">Scorecard</a>
+  <a href="/evaluations">Evaluations</a>
+  <a href="/case-studies">Case studies</a>
+  <a href="/numbers">Numbers</a>
+  <a href="/pricing">Pricing</a>
+</nav>
+<main class="container">
+  <h1>How we know ThumbGate works</h1>
+  <p class="subtitle">Evaluation white paper · Updated 2026-07-29 · Source: <code>docs/HOW-WE-KNOW-IT-WORKS.md</code></p>
+
+  <div class="rule">Offline goldens + live monitors, not vibes. <code>unsafeActionRate</code> must stay <span class="good">0</span>.</div>
+
+  <p>ThumbGate is a local-first PreToolUse enforcement layer: feedback becomes lessons, repeated failures become gates, and risky tool calls are checked before execution. This page is the buyer-readable map from that claim to evidence.</p>
+
+  <figure>
+    <img src="/assets/diagrams/thumbgate-architecture.png" alt="ThumbGate system architecture diagram" width="1200" height="800" loading="eager">
+    <figcaption>System map: agents, hooks, gate engine, governance, security, feedback promotion. Full gallery on <a class="inline" href="/architecture">/architecture</a>.</figcaption>
+  </figure>
+
+  <h2>1. Golden evaluation dataset</h2>
+  <p>Committed suites encode expected allow / deny / warn decisions and reference answers — not aspirational prose.</p>
+  <table>
+    <tr><th>Pack</th><th>Path</th></tr>
+    <tr><td>ThumbGate Bench</td><td><code>bench/thumbgate-bench.json</code></td></tr>
+    <tr><td>Agent safety eval</td><td><code>config/evals/agent-safety-eval.json</code></td></tr>
+    <tr><td>Prompt eval</td><td><code>bench/prompt-eval-suite.json</code></td></tr>
+    <tr><td>Observability / grounding</td><td><code>bench/observability-eval-suite.json</code></td></tr>
+    <tr><td>ProgramBench smoke</td><td><code>bench/programbench-smoke.json</code></td></tr>
+    <tr><td>Shell golden tests</td><td><code>verification/golden_tests/**</code></td></tr>
+  </table>
+  <p>Live failures feed offline goldens through <code>scripts/llm-behavior-monitor.js</code> promotion candidates.</p>
+
+  <h2>2. Offline regression tests</h2>
+  <pre>npm test
+npm run test:coverage
+npm run prove:adapters
+npm run prove:automation
+npm run self-heal:check</pre>
+  <p>Observed results land in <a class="inline" href="https://github.com/IgorGanapolsky/ThumbGate/blob/main/docs/VERIFICATION_EVIDENCE.md">docs/VERIFICATION_EVIDENCE.md</a>. Releases also require Changesets, SemVer, and version sync.</p>
+
+  <h2>3. Tool-call correctness</h2>
+  <p>ThumbGate Bench measures whether the gate decides correctly on the actual tool call.</p>
+  <table>
+    <tr><th>Metric</th><th>Intent</th></tr>
+    <tr><td><code>taskSuccessRate</code></td><td>Decision matches expected</td></tr>
+    <tr><td><code>unsafeActionRate</code></td><td class="good">Must stay 0</td></tr>
+    <tr><td><code>capabilityRate</code></td><td>Safe work still allowed</td></tr>
+    <tr><td><code>falseBlockRate</code></td><td>No “block everything” cheat</td></tr>
+    <tr><td><code>replayStability</code></td><td>Same suite → same decisions</td></tr>
+  </table>
+  <p><a class="cta" href="/eval-scorecard">Open the live scorecard</a> <a class="inline" href="https://github.com/IgorGanapolsky/ThumbGate/blob/main/docs/THUMBGATE_BENCH.md">Bench methodology →</a></p>
+
+  <h2>4. Latency</h2>
+  <figure>
+    <img src="/assets/diagrams/pre-action-gate-loop.svg" alt="Pre-action gate loop" width="900" height="360" loading="lazy">
+    <figcaption>Enforcement is local: deterministic match first, optional local embeddings, no cloud model on the critical path.</figcaption>
+  </figure>
+  <ul>
+    <li>No LLM on block/allow for the PreToolUse decision</li>
+    <li>Routing budgets: <code>lowLatencyBudgetMs: 300</code> for the fast classical lane</li>
+    <li>High-risk ambiguity can stop for human review instead of spending tokens</li>
+  </ul>
+
+  <h2>5. Cost</h2>
+  <ul>
+    <li>A blocked tool call does not spend the model round-trip</li>
+    <li><code>npx thumbgate cost</code> prints conservative $ saved from recorded blocks</li>
+    <li>Budget ledger + tokenomics guardrails prevent runaway spend</li>
+    <li>Methodology: <a class="inline" href="/agents-cost-savings">/agents-cost-savings</a> · inventory: <a class="inline" href="/numbers">/numbers</a></li>
+  </ul>
+
+  <h2>6. Human review</h2>
+  <ul>
+    <li>Classifier route <code>human_review</code> for credentials, customer data, regulated workflows, payments</li>
+    <li>Rubrics require verification evidence before “done”</li>
+    <li>Protected actions need scoped approval; break-glass is short-lived and narrow</li>
+    <li>Outbound social replies stay draft-only until a human publishes</li>
+  </ul>
+
+  <h2>7. Production monitoring</h2>
+  <figure>
+    <img src="/assets/diagrams/feedback-pipeline.png" alt="Feedback pipeline sequence diagram" width="1200" height="700" loading="lazy">
+    <figcaption>Capture → promote → block the repeat. Production monitors exist to catch silence (fail-open hooks) as well as loud breakage.</figcaption>
+  </figure>
+  <ul>
+    <li>Deploy health: Railway <code>/health</code> version + dashboard smoke</li>
+    <li><code>self-heal:check</code> — budget, tests, prove lanes must be HEALTHY</li>
+    <li>Gate stats + LLM behavior rates (wrong-tool, malformed, drift)</li>
+    <li>Published-artifact evasion jobs and silent-gate canaries</li>
+  </ul>
+
+  <h2>20-minute buyer audit</h2>
+  <pre>git clone https://github.com/IgorGanapolsky/ThumbGate
+cd ThumbGate &amp;&amp; npm ci
+npm test
+npm run prove:adapters &amp;&amp; npm run prove:automation
+npm run self-heal:check
+npm run thumbgate:bench -- --json
+curl -s https://thumbgate-production.up.railway.app/health</pre>
+
+  <h2>What this paper is not</h2>
+  <div class="panel">
+    <ul>
+      <li>Not external customer revenue proof — see commercial truth in the repo</li>
+      <li>Not a claim that free installs hard-block every risky command by default</li>
+      <li>Not a substitute for human review on payments, credentials, or regulated actions</li>
+    </ul>
+  </div>
+
+  <p>
+    <a class="cta" href="/case-studies">Case studies</a>
+    <a class="cta" href="/evaluations">ML evaluations</a>
+    <a class="cta" href="/architecture">Diagram gallery</a>
+  </p>
+
+  <p class="footer-note">
+    Longform source: <a href="https://github.com/IgorGanapolsky/ThumbGate/blob/main/docs/HOW-WE-KNOW-IT-WORKS.md">docs/HOW-WE-KNOW-IT-WORKS.md</a>.
+    Built by <a href="https://github.com/IgorGanapolsky">Igor Ganapolsky</a>.
+  </p>
+</main>
+</body>
+</html>