diff --git a/package.json b/package.json
index 383eadf..07e8017 100644
--- a/package.json
+++ b/package.json
@@ -29,7 +29,8 @@
     "format:check": "prettier --check src test scripts",
     "typecheck": "tsc --noEmit",
     "fixtures:record": "node scripts/record-fixtures.mjs",
-    "gen:models": "node scripts/gen-models.mjs"
+    "gen:models": "node scripts/gen-models.mjs",
+    "e2e": "node scripts/e2e.mjs"
   },
   "dependencies": {
     "hono": "^4.0.0",
diff --git a/scripts/e2e.mjs b/scripts/e2e.mjs
new file mode 100644
index 0000000..4f13dc3
--- /dev/null
+++ b/scripts/e2e.mjs
@@ -0,0 +1,177 @@
+#!/usr/bin/env node
+/**
+ * End-to-end API tests against a live deployment.
+ *
+ *   node scripts/e2e.mjs                          # https://api.interscript.org
+ *   E2E_BASE_URL=https://interscript-api-staging.interscript.workers.dev node scripts/e2e.mjs
+ *
+ * Zero dependencies; exits non-zero on any failure. The first /v1/infer
+ * call may take ~60s (inference-service cold start).
+ */
+const BASE = process.env.E2E_BASE_URL ?? "https://api.interscript.org"
+
+let passed = 0
+const failures = []
+
+function check(name, cond, detail = "") {
+  if (cond) {
+    passed++
+    console.log(`  ok ${name}`)
+  } else {
+    failures.push(`${name}${detail ? ` — ${detail}` : ""}`)
+    console.log(`FAIL ${name}${detail ? ` — ${detail}` : ""}`)
+  }
+}
+
+async function get(path) {
+  const res = await fetch(`${BASE}${path}`)
+  const text = await res.text()
+  let body
+  try {
+    body = JSON.parse(text)
+  } catch {
+    body = text
+  }
+  return { status: res.status, body, headers: res.headers }
+}
+
+async function post(path, body) {
+  const res = await fetch(`${BASE}${path}`, {
+    method: "POST",
+    headers: { "content-type": "application/json" },
+    body: JSON.stringify(body),
+  })
+  const text = await res.text()
+  let parsed
+  try {
+    parsed = JSON.parse(text)
+  } catch {
+    parsed = text
+  }
+  return { status: res.status, body: parsed }
+}
+
+console.log(`E2E against ${BASE}\n`)
+
+// --- service surface ---
+{
+  const res = await get("/")
+  check("GET / 200", res.status === 200)
+  check("GET / lists endpoints", Array.isArray(res.body.endpoints) && res.body.endpoints.includes("POST /v1/infer"))
+}
+{
+  const res = await get("/openapi.json")
+  check("GET /openapi.json 200", res.status === 200)
+  check("openapi 3.1", res.body.openapi === "3.1.0")
+  const paths = Object.keys(res.body.paths ?? {})
+  for (const p of ["/v1/info", "/v1/maps", "/v1/maps/{code}", "/v1/transliterate", "/v1/detect", "/v1/models", "/v1/models/{id}", "/v1/infer"]) {
+    check(`openapi has ${p}`, paths.includes(p))
+  }
+}
+{
+  const res = await get("/v1/info")
+  check("GET /v1/info 200", res.status === 200)
+  check("287 map systems", res.body.maps === 287, `got ${res.body.maps}`)
+  check("model index served", res.body.models >= 9, `got ${res.body.models}`)
+}
+
+// --- maps ---
+{
+  const res = await get("/v1/maps")
+  check("GET /v1/maps 200", res.status === 200)
+  check("maps count consistent", res.body.count === res.body.maps.length)
+  check("known system present", res.body.maps.includes("alalc-kat-Geor-Latn-1997"))
+}
+{
+  const ok = await get("/v1/maps/alalc-kat-Geor-Latn-1997")
+  check("GET /v1/maps/{code} 200", ok.status === 200 && typeof ok.body === "object")
+  const missing = await get("/v1/maps/does-not-exist")
+  check("unknown map 404", missing.status === 404)
+  check("map 404 error shape", missing.body?.error?.code === "map_not_found")
+}
+
+// --- transliteration (byte-parity with the Ruby API fixture) ---
+{
+  const res = await post("/v1/transliterate", { system: "alalc-kat-Geor-Latn-1997", input: "ქართული" })
+  check("transliterate 200", res.status === 200)
+  check("transliterate Ruby parity", res.body.output === "kʻartʻuli", `got ${JSON.stringify(res.body.output)}`)
+  const missing = await post("/v1/transliterate", { system: "nope-xxx-Xxx-Xxx-0000", input: "x" })
+  check("transliterate unknown 404", missing.status === 404)
+  const tooLong = await post("/v1/transliterate", { system: "alalc-kat-Geor-Latn-1997", input: "a".repeat(1_000_001) })
+  check("transliterate 1MB limit 413", tooLong.status === 413)
+  const bad = await post("/v1/transliterate", { system: 42 })
+  check("transliterate bad body 400", bad.status === 400)
+}
+
+// --- detect ---
+{
+  const res = await post("/v1/detect", { input: "ქართული", output: "kʻartʻuli" })
+  check("detect 200", res.status === 200)
+  check("detect ranks a Georgian system first", (res.body.results ?? [])[0]?.mapName?.includes("kat"), JSON.stringify((res.body.results ?? []).slice(0, 2)))
+}
+
+// --- models ---
+{
+  const res = await get("/v1/models")
+  check("GET /v1/models 200", res.status === 200)
+  check("models count consistent", res.body.count === res.body.models.length)
+  const ids = res.body.models.map((m) => m.id)
+  check("hebrew model listed", ids.includes("heb-diac-1.0"))
+  check("urdu g2p listed", ids.includes("urd-g2p-1.0"))
+}
+{
+  const ok = await get("/v1/models/heb-diac-1.0")
+  check("GET /v1/models/{id} 200", ok.status === 200 && ok.body.task === "diacritization")
+  check("model entry has metrics", Array.isArray(ok.body.metrics) && ok.body.metrics.length > 0)
+  const missing = await get("/v1/models/nope-9.9")
+  check("unknown model 404", missing.status === 404)
+}
+
+// --- neural inference (live models) ---
+{
+  const res = await post("/v1/infer", { model: "heb-diac-small-1.0", input: "שלום עליכם" })
+  check("infer hebrew 200", res.status === 200, JSON.stringify(res.body).slice(0, 120))
+  check("infer hebrew adds nikud", typeof res.body.output === "string" && res.body.output.includes("ָ"), `got ${JSON.stringify(res.body.output)}`)
+}
+{
+  const res = await post("/v1/infer", { model: "urd-g2p-1.0", input: "اردو" })
+  check("infer urdu 200", res.status === 200)
+  check("infer urdu IPA output", typeof res.body.output === "string" && res.body.output.length > 0 && /[ɑɐəɪʊuː]/.test(res.body.output), `got ${JSON.stringify(res.body.output)}`)
+}
+{
+  const missing = await post("/v1/infer", { model: "nope-9.9", input: "x" })
+  check("infer unknown model 404", missing.status === 404)
+  const bad = await post("/v1/infer", { model: "heb-diac-1.0" })
+  check("infer missing input 400", bad.status === 400)
+}
+
+// --- GraphQL back-compat ---
+{
+  const q = async (query) => {
+    const res = await fetch(`${BASE}/graphql`, {
+      method: "POST",
+      headers: { "content-type": "application/json" },
+      body: JSON.stringify({ query }),
+    })
+    return res.json()
+  }
+  const info = await q("{ info }")
+  check("graphql info", JSON.parse(info.data.info).version.startsWith("3."))
+  const codes = await q("{ systemCodes }")
+  check("graphql systemCodes 287", codes.data.systemCodes.length === 287, `got ${codes.data.systemCodes.length}`)
+  const tr = await q('{ transliterate(systemCode: "alalc-kat-Geor-Latn-1997", input: "ქართული") }')
+  check("graphql transliterate parity", tr.data.transliterate === "kʻartʻuli")
+}
+
+// --- CORS ---
+{
+  const res = await fetch(`${BASE}/v1/info`, { method: "OPTIONS" })
+  check("CORS preflight ok", res.status === 204 || res.ok, `status ${res.status}`)
+  check("CORS allow-origin", (res.headers.get("access-control-allow-origin") ?? "") === "*")
+}
+
+console.log(`\n${passed} passed, ${failures.length} failed`)
+if (failures.length) {
+  for (const f of failures) console.log(`FAIL ${f}`)
+  process.exit(1)
+}