diff --git a/docs/evidence/lead-form-tablet-squeeze-2026-08-10.md b/docs/evidence/lead-form-tablet-squeeze-2026-08-10.md
new file mode 100644
index 00000000..0744cb16
--- /dev/null
+++ b/docs/evidence/lead-form-tablet-squeeze-2026-08-10.md
@@ -0,0 +1,82 @@
+# Homepage lead form — tablet squeeze fix, measured evidence
+
+Date: 2026-08-10
+Scope: `public/index.css` — the homepage intake form (`form.lead.two`) squeezed the domain and email inputs to ~100px at tablet width. The two-field form shares the `.lead` class with the checks-section header, whose `gap:70px / justify-content:space-between / align-items:flex-end` treatment leaked onto the form (via the `.lead` selector) and, together with the nowrap submit button, starved the inputs of width below desktop sizes.
+This receipt records real browser-layout measurements. It is behavior evidence, not a source check, and it does not claim anything about the hosted/live deployment.
+
+## What was measured
+
+A headless Chromium rendered the homepage from a local static copy of `public/` (Google Fonts loaded over the network as in production; `document.fonts.ready` waited on, Karla confirmed loaded). For each viewport we recorded `getBoundingClientRect()` width for the form, both inputs, and the submit button, plus the computed `flex-direction`/`gap`, whether the "yourwebsite.com" placeholder's measured text width fits inside the first input's visible text area (input width minus its 42px horizontal padding), and whether `document.documentElement.scrollWidth == clientWidth` (no horizontal overflow).
+
+The "unfixed" variant is the exact pre-change tree: same `public/`, but `index.css` (and its HTML/JS dependencies) replaced by the `origin/main` (HEAD) versions, so both states are measured with one method in one session.
+
+## Environment
+
+- Node v22.23.1, Playwright 1.62.1, bundled Chromium headless, Linux.
+- Viewports: 390x900 and 761/768/800/834/900/901/1024/1280 wide x900 (deviceScaleFactor 1).
+- Wait: `networkidle`, then `document.fonts.ready`; Karla `document.fonts.check("16px Karla")` confirmed true in every run.
+
+## Results — input widths (px), unfixed → fixed
+
+| Viewport | layout (**unfixed**) | i1/i2 (**unfixed**) | layout (**fixed**) | i1/i2 (**fixed**) |
+|---|---|---|---|---|
+| 390 | column, gap 18px | 332 / 332 | column, gap 18px | 332 / 332 (identical) |
+| 761 | row, gap 70px | **99 / 98** | column, gap 18px | 623 / 623 |
+| 768 | row, gap 70px | **103 / 102** | column, gap 18px | 630 / 630 |
+| 800 | row, gap 70px | **119 / 118** | column, gap 18px | 662 / 662 |
+| 834 | row, gap 70px | **136 / 135** | column, gap 18px | 696 / 696 |
+| 900 | row, gap 70px | **139 / 138** | column, gap 18px | 762 / 762 |
+| 901 | row, gap 70px | **139 / 138** | row, no gap | 209 / 208 |
+| 1024 | row, gap 70px | **139 / 138** | row, no gap | 209 / 208 |
+| 1280 | row, gap 70px | **139 / 138** | row, no gap | 209 / 208 |
+
+Submit button: 294px (nowrap) in every row layout, unfixed and fixed. The unfixed "yourwebsite.com" placeholder text (~140px at 16px Karla) does **not** fit the first input's visible area at any viewport ≥761px — including desktop, where the leaked 70px gaps cap the inputs at 139px (97px visible). After the fix the placeholder fits at every width: 167px visible in the 901px+ one-line layout, full-width in the stacked layout below.
+
+## What changed
+
+1. The checks-header rules `.lead{...}` (desktop and mobile) are scoped to `.checks .lead`, so the shared class can no longer leak its `gap:70px / space-between / flex-end` treatment onto the lead form.
+2. The two-field form's stacked breakpoint moves from `max-width:760px` to its own `max-width:900px`, with an explicit `gap:18px` (previously inherited from the leaked `.lead` mobile rule), so tablet widths (761–900px) stack full-width instead of squeezing.
+
+## Deliberate desktop change
+
+At ≥901px the one-line form loses the accidental 70px gaps: inputs grow 139 → 208px and the placeholders render fully. This matches the identical two-field form on /audit (shared.css), which never had the `.lead` leak and rendered 264px inputs at 1024px. The submit button (294px) and the pill styling are unchanged. At ≤390px the stacked mobile layout is pixel-identical before and after (332/332 inputs, gap 18px).
+
+## Checks header neutrality
+
+The scoped `.checks .lead` selector leaves the "What we actually look at" header exactly as before: 1024px `space-between / gap 70px / flex-end` row; 390px column with `gap 18px`. The audit page is untouched (its form was never squeezed: shared.css has no `.lead` class rule and its shorter button leaves 224–264px inputs; measured pre-fix).
+
+## No horizontal overflow introduced
+
+At every viewport in both runs, `document.documentElement.scrollWidth == clientWidth` (761px: 761; 768px: 768; 901px: 901; …). The stacked form is `max-width:100%` inside the existing 60px `.wrap` padding.
+
+## Not CI proof
+
+This is local static-server proof, not CI proof and not a hosted/live claim. The repo's CI (`npm test`) has no browser dependency, so it runs only the source-string guards in `scripts/check-site.mjs` (which now pin the lead-form rules and this receipt's anchors) plus the worker/UI contract tests; a browser-layout regression can therefore still ship if the checked-in CSS drifts and CI stays green. The live tinystudio.io deployment was not measured here; a deployed page could differ (CDN cache, different asset versions). To be CI-proof this measurement would need a browser step added to CI (not done — out of scope).
+
+## Exact verification method (reproduce)
+
+1. Copy `public/` to `fixed-site/` and serve it on 127.0.0.1:8131 (`python3 -m http.server 8131 --directory fixed-site`). Copy it again to `unfixed-site/` and overwrite `index.css` (and `index.html`, `index.js`, `fonts.js` for a clean tree) with their `origin/main` versions (`git show origin/main:public/index.css > unfixed-site/index.css`), then serve it on 127.0.0.1:8132.
+2. Run this per state and viewport (requires `npm i -D playwright && npx playwright install chromium` in any project):
+
+```js
+import { chromium } from "playwright";
+const browser = await chromium.launch();
+const page = await browser.newPage();
+await page.setViewportSize({ width: 768, height: 900 });
+await page.goto("http://127.0.0.1:8131/");   // or :8132 for unfixed
+await page.waitForLoadState("networkidle");
+await page.evaluate(() => document.fonts.ready);
+const m = await page.evaluate(() => {
+  const form = document.querySelector("form.two");
+  const inputs = [...form.querySelectorAll("input")];
+  const r = (el) => Math.round(el.getBoundingClientRect().width);
+  return {
+    i1: r(inputs[0]), i2: r(inputs[1]), btn: r(form.querySelector("button")),
+    direction: getComputedStyle(form).flexDirection, gap: getComputedStyle(form).gap,
+    scrollOK: document.documentElement.scrollWidth === document.documentElement.clientWidth
+  };
+});
+console.log(m);
+```
+
+Unfixed at 768px: `{ i1: 103, i2: 102, btn: 294, direction: "row", gap: "70px", scrollOK: true }`. Fixed at 768px: `{ i1: 630, i2: 630, btn: 630, direction: "column", gap: "18px", scrollOK: true }`. Fixed at 1024px: `{ i1: 209, i2: 208, btn: 294, direction: "row", gap: "normal", scrollOK: true }`.
diff --git a/public/index.css b/public/index.css
index 8726a58f..e644cc8b 100644
--- a/public/index.css
+++ b/public/index.css
@@ -103,9 +103,12 @@
 
   /* ── checks ────────────────────────────────────── */
   .checks{margin-top:124px;padding-top:70px;border-top:1px solid var(--line)}
-  .lead{display:flex;justify-content:space-between;align-items:flex-end;gap:70px}
+  /* scoped to .checks: the homepage lead form shares the .lead class, and an
+     unscoped gap/space-between/flex-end treatment leaked onto the form,
+     squeezing the domain+email inputs to ~100px at tablet widths */
+  .checks .lead{display:flex;justify-content:space-between;align-items:flex-end;gap:70px}
   .checks h2{font-family:'Fraunces',serif;font-weight:200;font-size:47px;letter-spacing:-.028em;max-width:12ch;line-height:1.08}
-  .lead p{max-width:38ch;font-size:15.5px;line-height:1.75;color:var(--muted)}
+  .checks .lead p{max-width:38ch;font-size:15.5px;line-height:1.75;color:var(--muted)}
   .checkgrid{margin-top:56px;display:grid;grid-template-columns:repeat(4,1fr)}
   .check{padding-right:34px;border-right:1px solid var(--hair)}
   .check:last-child{border-right:0;padding-right:0}
@@ -169,13 +172,16 @@
   .navlinks a:hover{color:var(--ink)}
   .navcta{border:1px solid var(--line);border-radius:999px;padding:9px 17px;color:var(--ink)!important}
   .navcta:hover{border-color:var(--brass);background:rgba(255,253,248,.7)}
-  /* two-field lead form — website + email */
+  /* two-field lead form — website + email, still one line on desktop */
   form.two{max-width:720px;padding:7px 7px 7px 0;overflow:hidden}
   form.two input{flex:1 1 0;min-width:0;padding:12px 18px 12px 24px}
   form.two input:first-of-type{border-right:1px solid var(--line)}
   form.two button{flex:0 0 auto}
-  @media (max-width:760px){
-    form.two{flex-direction:column;align-items:stretch;border-radius:20px;padding:8px;max-width:100%}
+  /* Stack below 900px: with the nowrap submit button, a one-line row below
+     tablet width squeezes the inputs to ~100px. The gap is explicit here —
+     it used to arrive via the .lead class leak (now scoped to .checks). */
+  @media (max-width:900px){
+    form.two{flex-direction:column;align-items:stretch;border-radius:20px;padding:8px;max-width:100%;gap:18px}
     form.two input{padding:14px 18px}
     form.two input:first-of-type{border-right:0;border-bottom:1px solid var(--line)}
     form.two button{width:100%}
@@ -235,7 +241,7 @@
   .finding .stat{font-size:56px}
   .spec{grid-template-columns:1fr;gap:34px}
   .specbody h2{font-size:34px}
-  .lead{flex-direction:column;align-items:flex-start;gap:18px}
+  .checks .lead{flex-direction:column;align-items:flex-start;gap:18px}
   .checks h2{font-size:36px}
   .checkgrid{grid-template-columns:1fr;gap:26px}
   .check{padding-right:0;padding-bottom:26px;border-right:0;border-bottom:1px solid var(--hair)}
diff --git a/scripts/check-site.mjs b/scripts/check-site.mjs
index ef40c8c5..b9c638ee 100644
--- a/scripts/check-site.mjs
+++ b/scripts/check-site.mjs
@@ -445,6 +445,43 @@ for (const anchor of ["unfixed", "390x844", "44px", "not CI proof", "Exact verif
   }
 }
 
+// Lead-form tablet squeeze guard (review finding on the homepage intake form):
+// the homepage form shares the .lead class with the checks-section header,
+// whose gap:70px / space-between / flex-end treatment leaked onto it and,
+// together with the nowrap submit button, squeezed the domain and email
+// inputs to ~100px at tablet width (measured 99/98px at 761px, 103/102px at
+// 768px, and never above 139px — the "yourwebsite.com" placeholder clipped —
+// at any width under the fix). The fix scopes the checks-header rules to
+// .checks .lead and gives the two-field form its own 900px stacking
+// breakpoint, so the one-line layout keeps >=208px inputs and anything
+// narrower goes full-width stacked. These are STATIC SOURCE GUARDS (regex
+// over the served CSS), not behavioral tests: CI has no browser. The
+// behavioral, measured layout proof lives in
+// docs/evidence/lead-form-tablet-squeeze-2026-08-10.md.
+const indexCss = read("public/index.css");
+if (!indexCss.includes(".checks .lead{display:flex;justify-content:space-between;align-items:flex-end;gap:70px}")) {
+  failures.push("Homepage checks header must scope .lead to .checks .lead (the shared class leaks its gap onto the lead form).");
+}
+if (!indexCss.includes("@media (max-width:900px){")) {
+  failures.push("Two-field lead form must carry its own 900px stacking breakpoint (tablet squeeze guard).");
+}
+if (!indexCss.includes("form.two{flex-direction:column;align-items:stretch;border-radius:20px;padding:8px;max-width:100%;gap:18px}")) {
+  failures.push("Two-field lead form must stack below 900px with an explicit gap (tablet squeeze guard).");
+}
+if (indexCss.includes("@media (max-width:760px){\n    form.two{flex-direction:column")) {
+  failures.push("Two-field lead form must not stack at 760px (squeezes the inputs to ~100px at tablet widths).");
+}
+
+// The behavioral lead-form proof (real Chromium measurement, local static
+// copy) is checked in so the static guards above stay distinguishable from it.
+// Existence and section anchors only — this does not re-verify measurements.
+const leadFormReceipt = read("docs/evidence/lead-form-tablet-squeeze-2026-08-10.md");
+for (const anchor of ["unfixed", "761px", "**fixed**", "208px", "not CI proof", "Exact verification method"]) {
+  if (!leadFormReceipt.includes(anchor)) {
+    failures.push(`Lead-form evidence receipt must record the ${JSON.stringify(anchor)} section.`);
+  }
+}
+
 // Render-blocking font stylesheet guards (dogfood finding b8f6046e942a).
 // The Google Fonts css2 stylesheet was fetched render-blocking on every public
 // page — a <link rel="stylesheet"> in the homepage head, and an @import chain