diff --git a/internal/check-link-rot-sample.mjs b/internal/check-link-rot-sample.mjs
index d667c8d6d..4e389454a 100644
--- a/internal/check-link-rot-sample.mjs
+++ b/internal/check-link-rot-sample.mjs
@@ -6,24 +6,48 @@
  * keep offering it indefinitely, still labelled "link checked". Clause L of the coverage gate
  * guards the freshness of the RECORD; it cannot notice that the web moved underneath it.
  *
- * WHY A SAMPLE. 8,285 verified links is a lot of traffic to aim at seven ministries for a
- * maintenance check. A stratified sample — up to 25 per host, taken at a fixed stride so the
+ * WHY A SAMPLE. 8,394 verified links is a lot of traffic to aim at a handful of ministries for
+ * a maintenance check. A stratified sample — up to 25 per host, taken at a fixed stride so the
  * same run is repeatable — is enough to tell "nothing is rotting" from "a publisher has
  * reorganised", which is the decision this check exists to inform. A full re-verify is
  * scripts/verify-gospel-links.mjs and is the right tool once this finds something.
  *
- * ONLY HOSTS WHERE A STATUS MEANS SOMETHING — the part worth keeping.
- * Four of the eleven hosts NEVER 404:
- *   - jesusfilm.org and live.bible.is return an identical SPA shell for any path;
- *   - knowgod.com does the same;
- *   - fivefish.org answers a nonsense path with a plausible geo-located page, HTTP 200.
- * Re-checking those by status would report perfect health forever, which is worse than not
- * checking them at all, because it would look like evidence. They are excluded here and are
- * verified by their publisher's API or by page title in the main verifier instead.
+ * ─────────────────────────────────────────────────────────────────────────────────────────
+ * WHAT THIS FILE USED TO SAY, AND WHY IT WAS WRONG. The first version excluded four hosts on
+ * the grounds that they "NEVER 404" and so could not be re-checked by status. That belief was
+ * inherited, not measured, and it quietly exempted 5,760 of 8,394 shipped links — 69% of the
+ * corpus, including the two largest hosts on the map. Measured on 2026-09-25, on the exact
+ * URL SHAPE the map actually ships:
  *
- * IT IS SHOWN TO BE ABLE TO FAIL. Three URLs that must not exist are requested in band every
- * run. If any of them answers 2xx the check reports that it proved nothing and exits non-zero
- * rather than reporting a pass — the same rule applied to every other gate in this bundle.
+ *   jesusfilm.org   /watch/jesus/<slug>   real 200 · three different bogus slugs ALL 404,
+ *                   by HEAD as well as GET. It does 404. The exclusion was simply false, and
+ *                   2,049 links sat outside every check because of it.
+ *   live.bible.is   /bible/<filesetId>    always 200, but a bogus fileset returns an empty
+ *                   SPA shell of 7,860–7,865 bytes while real filesets return 13.5–85 KB.
+ *                   Status is useless; SIZE separates them cleanly, at better than 1.7×.
+ *                   This is not a new trick — scripts/fetch-fcbh-index.mjs already builds the
+ *                   index this way and records `shellBytes` for it. Only this check had not
+ *                   learned it. 2,636 links.
+ *   fivefish.org    /<iso639-3>           always 200; found and not-found differ by page
+ *                   TITLE, which is how scripts/fetch-grn-index.mjs classifies them. Real,
+ *                   but their robots.txt asks Crawl-delay: 10, so a 25-link sample is four
+ *                   minutes of polite waiting and belongs in its own slower band, not here.
+ *                   982 links, still unchecked, deliberately and on the record.
+ *   knowgod.com     /<path>               byte-identical 3,864-byte "Know God" page for a real
+ *                   path and a nonsense one. This one really is unverifiable by fetching it.
+ *                   94 links, and saying so is the honest answer.
+ *
+ * The lesson is the one this repo keeps relearning: a claim about the outside world decays,
+ * and an exclusion is a claim. Each host below carries the measurement that justifies its
+ * band, so the next person can re-measure rather than re-inherit.
+ * ─────────────────────────────────────────────────────────────────────────────────────────
+ *
+ * IT IS SHOWN TO BE ABLE TO FAIL. Every band runs its controls in band, every run:
+ *   - status band: URLs that must not exist, one per host shape. Any 2xx and the check
+ *     reports that it proved nothing and exits non-zero rather than reporting a pass.
+ *   - shell band: a bogus fileset must read as shell (or the check cannot see death) AND a
+ *     fixed real fileset must read as substantive (or the check would call the whole host
+ *     dead the day its page shape changes).
  */
 import { readFileSync } from 'node:fs'
 import { resolve, dirname } from 'node:path'
@@ -36,34 +60,61 @@ const TABLE = resolve(REPO, 'data/tables/gospel-link-verification.json')
 const UA =
   'Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/125.0.0.0 Safari/537.36'
 
-/** Hosts whose HTTP status is real evidence. Everything else lies; see the header. */
+/** Hosts whose HTTP status is real evidence. Measured, not assumed — see the header. */
 const STATUS_MEANINGFUL = new Set([
   'ebible.org', 'alkitab.pw', 'png.bible', 'git.door43.org',
   'scriptureearth.org', 'zume.training', 'arc.gt',
+  'jesusfilm.org', 'www.jesusfilm.org',
 ])
 
+/** Hosts that answer 200 for anything but whose EMPTY page has a measurable size. */
+const SHELL_MEASURED = {
+  'live.bible.is': {
+    /* Measured at run time, never hard-coded: the shell grows when they ship a bundle. */
+    shellProbe: 'https://live.bible.is/bible/ZZZQQQ',
+    /* Must come back substantive, or the size rule has stopped separating anything. */
+    livingControl: 'https://live.bible.is/bible/ENGKJV',
+    /* 7,860 shell vs 13,547 smallest real page on 2026-09-25 — 1.72×. 1.25 is slack. */
+    aliveRatio: 1.25,
+  },
+}
+
+/** Hosts left out, with the reason, so the exclusion stays a measurement and not a habit. */
+const UNCHECKABLE = {
+  'fivefish.org': 'found/not-found differ by title only, and Crawl-delay: 10 needs its own band',
+  'knowgod.com': 'byte-identical page for a real and a nonsense path — unverifiable by fetch',
+}
+
 /** Must not exist. If one of these passes, the check cannot see death and says so. */
 const CONTROLS = [
   'https://ebible.org/zzz-no-such-translation-xyzzy/',
   'https://scriptureearth.org/data/zzz/PDF/nope.pdf',
   'https://git.door43.org/unfoldingWord/no-such-repo-xyzzy',
+  /* The shipped shape, not a plausible-looking one: /watch/jesus/<language-slug>. */
+  'https://jesusfilm.org/watch/jesus/zzz-no-such-language-xyzzy',
 ]
 
 const PER_HOST = Number(process.env.ROT_PER_HOST || 25)
 
-async function status(url) {
+async function get(url, method = 'GET') {
   try {
     const r = await fetch(url, {
+      method,
       headers: { 'user-agent': UA },
       redirect: 'follow',
       signal: AbortSignal.timeout(20000),
     })
-    return r.status
+    const body = method === 'GET' ? await r.text() : ''
+    return { status: r.status, len: body.length }
   } catch {
-    return 0
+    return { status: 0, len: 0 }
   }
 }
 
+/* A 808 KB film page times 25 is 20 MB of somebody else's bandwidth to learn one number that
+   HEAD already carries. Status hosts are asked with HEAD; the shell band needs the body. */
+const status = (url) => get(url, 'HEAD').then((r) => r.status)
+
 let table
 try { table = JSON.parse(readFileSync(TABLE, 'utf8')) } catch {
   console.error('LINK-ROT: NOT MEASURED — no verification table to sample.')
@@ -75,10 +126,11 @@ const verified = Object.entries(table.links || {})
   .map(([u]) => u)
 
 const byHost = new Map()
+let skipped = 0
 for (const u of verified) {
   let h
   try { h = new URL(u).host } catch { continue }
-  if (!STATUS_MEANINGFUL.has(h)) continue
+  if (!STATUS_MEANINGFUL.has(h) && !SHELL_MEASURED[h]) { skipped++; continue }
   if (!byHost.has(h)) byHost.set(h, [])
   byHost.get(h).push(u)
 }
@@ -86,15 +138,15 @@ for (const u of verified) {
 /* Fixed stride, sorted input: the same sample every run, so a change is a change in the web
    rather than a change in which links we happened to pick. */
 const sample = []
-for (const [, urls] of [...byHost].sort((a, b) => a[0].localeCompare(b[0]))) {
+for (const [host, urls] of [...byHost].sort((a, b) => a[0].localeCompare(b[0]))) {
   urls.sort()
   const step = Math.max(1, Math.floor(urls.length / PER_HOST))
-  const take = []
-  for (let i = 0; i < urls.length && take.length < PER_HOST; i += step) take.push(urls[i])
-  sample.push(...take)
+  for (let i = 0; i < urls.length && sample.filter((s) => s.host === host).length < PER_HOST; i += step) {
+    sample.push({ host, url: urls[i] })
+  }
 }
 
-/* ── controls first ─────────────────────────────────────────────────────────────────────── */
+/* ── controls: the status band ──────────────────────────────────────────────────────────── */
 const ctl = []
 for (const u of CONTROLS) ctl.push([u, await status(u)])
 const blind = ctl.filter(([, s]) => s >= 200 && s < 300)
@@ -105,21 +157,55 @@ if (blind.length) {
   console.error('  trusting any pass it reports.')
   process.exit(1)
 }
-console.log(`LINK-ROT: controls pass — ${ctl.length} impossible URL(s), none answered 2xx.`)
+console.log(`LINK-ROT: status controls pass — ${ctl.length} impossible URL(s), none answered 2xx.`)
 
+/* ── controls: the shell band, both directions ─────────────────────────────────────────── */
+const shells = {}
+for (const [host, cfg] of Object.entries(SHELL_MEASURED)) {
+  if (!byHost.has(host)) continue
+  const empty = await get(cfg.shellProbe)
+  const living = await get(cfg.livingControl)
+  const floor = empty.len * cfg.aliveRatio
+  if (!empty.len) {
+    console.error(`LINK-ROT: PROVED NOTHING — ${host} shell probe returned nothing to measure.`)
+    process.exit(1)
+  }
+  if (living.len <= floor) {
+    console.error(
+      `LINK-ROT: PROVED NOTHING — ${host} answered its KNOWN-GOOD control with ${living.len}b, ` +
+        `no bigger than the ${empty.len}b empty shell.`,
+    )
+    console.error('  The size rule has stopped separating a real page from an empty one, so a')
+    console.error('  pass here would mean nothing and a fail would libel every link on the host.')
+    process.exit(1)
+  }
+  shells[host] = floor
+  console.log(
+    `LINK-ROT: ${host} shell measured at ${empty.len}b; known-good page ${living.len}b ` +
+      `(${(living.len / empty.len).toFixed(1)}×). Alive means over ${Math.round(floor)}b.`,
+  )
+}
+
+/* ── the sample ─────────────────────────────────────────────────────────────────────────── */
 const dead = []
-for (const u of sample) {
-  const s = await status(u)
-  if (!(s >= 200 && s < 300)) dead.push([u, s])
+for (const { host, url } of sample) {
+  if (shells[host] !== undefined) {
+    const r = await get(url)
+    if (r.len <= shells[host]) dead.push([url, `${r.len}b — empty shell`])
+    continue
+  }
+  const s = await status(url)
+  if (!(s >= 200 && s < 300)) dead.push([url, s || 'no response'])
 }
 
 console.log(
   `LINK-ROT: re-checked ${sample.length} verified link(s) across ${byHost.size} host(s) ` +
-    `where a status is meaningful (${verified.length - sample.length} not sampled).`,
+    `(${skipped} link(s) on hosts nothing can check: ` +
+    `${Object.keys(UNCHECKABLE).join(', ')}).`,
 )
 if (dead.length) {
   console.error(`LINK-ROT: FAIL — ${dead.length} link(s) we call "checked" no longer resolve:`)
-  for (const [u, s] of dead.slice(0, 10)) console.error(`  ${s || 'no response'} ${u}`)
+  for (const [u, why] of dead.slice(0, 10)) console.error(`  ${why} ${u}`)
   console.error('  Re-run scripts/verify-gospel-links.mjs to re-test the whole corpus; the')
   console.error('  build drops anything it proves dead.')
   process.exit(1)