diff --git a/.claude/commands/ci.md b/.claude/commands/ci.md
index 792eeab3e0..d149c6bc17 100644
--- a/.claude/commands/ci.md
+++ b/.claude/commands/ci.md
@@ -1,5 +1,5 @@
 ---
-description: "Run the pre-PR gate — scripts/ci-local.sh, 43 ci-local gates, PASS/FAIL table. Mandatory before a PR. Iterate with bun test --parallel --changed instead."
+description: "Run the pre-PR gate — scripts/ci-local.sh, 44 ci-local gates, PASS/FAIL table. Mandatory before a PR. Iterate with bun test --parallel --changed instead."
 ---
 
 Run the same checks CI runs. This must pass before opening a PR.
@@ -138,7 +138,7 @@ pass); `CI_LOCAL_NO_NET=1` skips the two npm-registry gates for offline runs;
 It is a **superset of the historic /ci list** — it adds the drift gates real CI
 runs that the old /ci skipped (so you stop discovering them only after a push).
 The roster is `GATE_NAMES` in `scripts/ci-local.sh`, which is the authority; all
-43 ci-local gates, in run order:
+44 ci-local gates, in run order:
 `type`, `lint`, `lint-type-aware` (oxlint `--type-aware` via tsgolint — the
 promoted type-aware rules at `error`; permanent `warn` residuals don't fail it),
 `syncpack`, `dockerfile-bun-pins`, `dockerfile-workspace`,
diff --git a/.claude/research/launch/blog-intro.md b/.claude/research/launch/blog-intro.md
index 957462c84e..c470c9bcbb 100644
--- a/.claude/research/launch/blog-intro.md
+++ b/.claude/research/launch/blog-intro.md
@@ -1,3 +1,5 @@
+> **Superseded by docs/prd/launch-cycle.md (2026-09-03).** This draft describes a text-to-SQL product under an MIT license and must not be reused; the launch post is `apps/www/src/app/blog/a-name-on-every-fact` and the Show HN body is written from the PRD.
+
 # Why We Built Atlas: Text-to-SQL Is a Feature, Not a Product
 
 Every data-heavy application eventually gets the same feature request: "Can I just ask a question about my data?" Sales teams want pipeline numbers without opening a BI tool. Support teams want to look up customer metrics mid-conversation. Product managers want to check engagement data without waiting for an analyst.
diff --git a/.claude/research/launch/show-hn.md b/.claude/research/launch/show-hn.md
index 4e1375442b..1373b7c426 100644
--- a/.claude/research/launch/show-hn.md
+++ b/.claude/research/launch/show-hn.md
@@ -1,3 +1,5 @@
+> **Superseded by docs/prd/launch-cycle.md (2026-09-03).** This draft describes a text-to-SQL product under an MIT license and must not be reused; the launch post is `apps/www/src/app/blog/a-name-on-every-fact` and the Show HN body is written from the PRD.
+
 # Show HN: Atlas -- Open-source text-to-SQL agent you can embed in any app
 
 Atlas is an open-source (MIT) text-to-SQL agent that deploys as an API server and embeds into any application. Drop in a `<script>` tag or React component and your users can query their data in plain English.
diff --git a/.claude/rules/testing.md b/.claude/rules/testing.md
index ae482a78f1..7810b71ec5 100644
--- a/.claude/rules/testing.md
+++ b/.claude/rules/testing.md
@@ -10,7 +10,7 @@ paths:
 - [ ] **`bun test --parallel`, never bare `bun test`** — `--parallel` implies `--isolate`: a fresh global + module registry per file. Bare `bun test` shares one global across every file, so a suite can pass on state a sibling left behind. Single file OK: `bun test path/to/file.test.ts`
 - [ ] **Never `--no-isolate`** — it buys speed by letting a worker keep one global across the files it runs. `agent-compaction.test.ts` already leaks module state across same-process runs (false-fails under `--rerun-each`, passes 12/12 under `--parallel`), so this would make that class routine
 - [ ] **Use `--changed` for local loops** — `cd packages/api && bun test --parallel --changed=origin/main`. ⚠️ **Do NOT run the full `bun run test` locally** — it is a whole-suite `--parallel` run, one worker per core, and CLAUDE.md already makes **remote CI on the PR the gate** rather than any local full run. `.claude/hooks/guard-bun-test.sh` refuses the whole-suite shapes at the tool boundary so this cannot be reached by habit
-- [ ] **Pre-PR gates via `/ci`** — runs `scripts/ci-local.sh`: 43 ci-local gates (stage 0 type-check → stage 1 parallel checks: lint, type-aware lint, syncpack, template/schema/openapi/auth-md drift, ee-imports, twenty-resolver, migration-rename discipline, published-symbols, … → stage 2 full test suite via `bun test --parallel`). All must pass; the `/ci` skill carries the authoritative gate list and count. ⚠️ **Without `TEST_DATABASE_URL` the `test` gate now reports `DECLINED`, not PASS (#5410)** — every `*-pg.test.ts` self-skips, so a bare local run cannot be a clean pre-PR pass
+- [ ] **Pre-PR gates via `/ci`** — runs `scripts/ci-local.sh`: 44 ci-local gates (stage 0 type-check → stage 1 parallel checks: lint, type-aware lint, syncpack, template/schema/openapi/auth-md drift, ee-imports, twenty-resolver, migration-rename discipline, published-symbols, … → stage 2 full test suite via `bun test --parallel`). All must pass; the `/ci` skill carries the authoritative gate list and count. ⚠️ **Without `TEST_DATABASE_URL` the `test` gate now reports `DECLINED`, not PASS (#5410)** — every `*-pg.test.ts` self-skips, so a bare local run cannot be a clean pre-PR pass
 - [ ] **Mock all exports** when using `mock.module()`; use `createConnectionMock()` for connection mocks (never inline)
 - [ ] **Tests are self-contained** — No top-level `process.env.X =` or `process.chdir(...)`; `??=` hoist permitted for import-time env reads
 - [ ] **A mutation TABLE is GENERATED, never hand-typed** — a "MUTATIONS THIS CATCHES" table is a pointer to `packages/api/scripts/mutations/<name>.md`, rendered from `<name>.mutations.ts` by `scripts/mutate.ts`. Never edit a cell; regenerate. `scripts/check-mutation-tables.sh` is the gate, and it globs the directory. `-pg` specs need `TEST_DATABASE_URL` or the runner aborts on a deflated baseline. (A caveated prose count about ANOTHER suite — `vocabulary-pg.test.ts`'s "measured ONCE, at #5051" sibling list — is not a table and is not a violation; don't "fix" it by deleting it)
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 71accfdfd3..3d265638b7 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -319,6 +319,9 @@ jobs:
       - name: Adversarial fixtures for the plugin-count gate (#4066)
         run: bash scripts/__tests__/check-plugin-count.test.sh
 
+      - name: Check the launch sentence renders verbatim on every surface (#5606)
+        run: bash scripts/check-launch-sentence.sh
+
       - name: Check boot-loaded plugins ↔ api Dockerfile wiring lockstep (#4880)
         run: bun scripts/check-plugin-lockstep.ts
 
diff --git a/CLAUDE.md b/CLAUDE.md
index 5952c367ea..13a1e039e8 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -33,7 +33,7 @@ These hold everywhere. The rest of this file is orientation, not rules.
 
 ### Tests
 - **`bun test --parallel`, never bare `bun test`** — `--parallel` implies `--isolate`: a fresh global + module registry per file. Bare `bun test` shares one global across every file and will pass on state a sibling left behind. Single file OK: `bun test path/to/file.test.ts`. ⚠️ **Never `--no-isolate`** — it trades that isolation for speed, and at least one suite (`agent-compaction.test.ts`) leaks module state across same-process runs today
-- **Remote CI on the PR is the gate, not a local `/ci`** — push, open the PR as a **draft**, and let `ci.yml` run (~4 min, parallel) while you review. The local pre-flight is the cheap subset: `cd packages/api && bun test --parallel --changed=origin/main`, plus `bun run lint`, `bun run type` and `bun run lint:type-aware`. **`lint:type-aware` is on that list because it is its own CI-blocking job and costs ~11s** — leaving it off is what let a single type-aware diagnostic red-flag two CI jobs on #5083 *after* the pre-flight came back clean. Run the full `scripts/ci-local.sh` (43 ci-local gates, ~25 min, serial, rewrites source in place for the mutation gate) only when remote CI is broken or when you reshaped something `mutation-tables` anchors on. **Not "before tagging a release" — that clause is gone (#5410):** the local wrapper does not go green on an unchanged tree, so requiring it blocked every release including ones already serving prod. Remote CI on the tagged SHA is the release gate; local `/ci` is advisory. ⚠️ And without `TEST_DATABASE_URL` the local run self-skips all 104 real-Postgres suites — 1,432 assertions when last measured (2026-08-24, at 87 suites) — so it now reports `DECLINED`/exit 3 rather than a green that verified none of them
+- **Remote CI on the PR is the gate, not a local `/ci`** — push, open the PR as a **draft**, and let `ci.yml` run (~4 min, parallel) while you review. The local pre-flight is the cheap subset: `cd packages/api && bun test --parallel --changed=origin/main`, plus `bun run lint`, `bun run type` and `bun run lint:type-aware`. **`lint:type-aware` is on that list because it is its own CI-blocking job and costs ~11s** — leaving it off is what let a single type-aware diagnostic red-flag two CI jobs on #5083 *after* the pre-flight came back clean. Run the full `scripts/ci-local.sh` (44 ci-local gates, ~25 min, serial, rewrites source in place for the mutation gate) only when remote CI is broken or when you reshaped something `mutation-tables` anchors on. **Not "before tagging a release" — that clause is gone (#5410):** the local wrapper does not go green on an unchanged tree, so requiring it blocked every release including ones already serving prod. Remote CI on the tagged SHA is the release gate; local `/ci` is advisory. ⚠️ And without `TEST_DATABASE_URL` the local run self-skips all 104 real-Postgres suites — 1,432 assertions when last measured (2026-08-24, at 87 suites) — so it now reports `DECLINED`/exit 3 rather than a green that verified none of them
 
 ### Merge discipline
 Rationale + override rules: [docs/development/branch-protection.md](docs/development/branch-protection.md). These are workflow rules — no file-read triggers them, so they stay here:
diff --git a/apps/www/public/llms.txt b/apps/www/public/llms.txt
index 8ab4ca13fa..93f0933965 100644
--- a/apps/www/public/llms.txt
+++ b/apps/www/public/llms.txt
@@ -1,20 +1,23 @@
 # Atlas
 
-> Deploy-anywhere text-to-SQL data analyst agent. Open-source under AGPL-3.0.
+> Atlas is the company facts your AI agents can trust: every one carries its source, its date, and the name of the person who approved it. Open source, runs in your VPC.
 
-Atlas connects to your database, auto-generates a YAML semantic layer from your schema, and lets an AI agent query your data in plain English. It validates every query (read-only, AST-parsed, table-whitelisted) and returns structured results with charts and narrative.
+To connect an AI agent to Atlas, run this in a terminal. It configures Claude Desktop, Cursor or Continue (or prints a config block for any other MCP client with `--client generic`) against the hosted NovaMart demo, with no account and no email:
+
+```bash
+bunx @useatlas/mcp init --hosted --demo --write
+```
+
+Then ask: "What is NovaMart's return window?" The answer names who said it, where and when, marks that a person approved it before it counted, and surfaces the contradiction between Finance and Support without picking a side. Full guide: https://docs.useatlas.dev/guides/mcp. Machine-readable docs: https://docs.useatlas.dev/llms.txt.
+
+Three kinds of thing live in the Atlas, and every answer says which it is drawing on: **Surveyed** (read live from the company's own data through a semantic layer, SELECT-only, validated seven ways — it cannot go stale), **Attested** (extracted from something someone wrote, then approved by a named person who is on the record), and **On the record** (the raw source, unedited). Surveyed outranks Attested wherever they overlap. Contradictions are shown with both sources; Atlas picks neither. The complete Atlas runs self-hosted under AGPL-3.0; hosted Atlas is in the US, with EU and APAC regions built and brought online on request.
 
 ## Core capabilities
 
-- **Text-to-SQL agent**: Multi-step reasoning powered by Vercel AI SDK. Explores schema, writes validated SQL, runs Python analysis, returns charts and narrative.
-- **Semantic layer**: YAML-based entity definitions with auto-profiling, glossary terms, metrics, joins, and LLM enrichment. Versioned in git, reviewed in PRs.
-- **8 databases**: PostgreSQL, MySQL, ClickHouse, Snowflake, DuckDB, BigQuery, Salesforce, Elasticsearch/OpenSearch.
-- **6 LLM providers**: Anthropic, OpenAI, AWS Bedrock, Ollama, OpenAI-compatible (vLLM, TGI, LiteLLM), AI Gateway.
-- **24 plugins**: Datasource adapters, sandbox backends (nsjail, E2B, Daytona, Vercel sandbox), interaction channels, action triggers, context providers. Extensible via Plugin SDK.
-- **9+ integration channels**: Slack, Microsoft Teams, Discord, Email, Google Chat, Telegram, GitHub, Linear, WhatsApp, plus webhooks.
-- **Embeddable**: Script tag widget, React component (`@useatlas/react`), TypeScript SDK (`@useatlas/sdk`), headless API.
-- **MCP server**: Works with Claude Desktop, Cursor, and any MCP-compatible client.
-- **Dynamic learning**: Learns query patterns over time. Stored as auditable YAML, not opaque embeddings.
+- **The Company Atlas**: facts extracted from what the company already says (chat, transcripts, mail), each approved by a named person before it counts, carrying its source and date; contradictions surfaced with both sources; a coverage page that marks what nobody has surveyed.
+- **The analyst (Surveyed tier)**: a YAML semantic layer you author — entities, joins, glossary terms, pinned metrics — and an agent that writes validated, read-only SQL against it and returns rows, charts and narrative.
+- **Agent-native**: an MCP server (hosted over OAuth 2.1, self-hosted over stdio) for Claude Desktop, Cursor, Continue and any MCP client; a TypeScript SDK; an embeddable React component and script-tag widget; chat-platform adapters.
+- **Datasources, LLM providers, plugins and integrations**: the full inventories live in the docs — https://docs.useatlas.dev/getting-started/connect-your-data, https://docs.useatlas.dev/guides/model-routing, https://docs.useatlas.dev/plugins/authoring-guide, https://docs.useatlas.dev/guides/integrations.
 
 ## Security
 
diff --git a/apps/www/src/app/blog/a-name-on-every-fact/page.tsx b/apps/www/src/app/blog/a-name-on-every-fact/page.tsx
new file mode 100644
index 0000000000..226abc9ed8
--- /dev/null
+++ b/apps/www/src/app/blog/a-name-on-every-fact/page.tsx
@@ -0,0 +1,222 @@
+import type { Metadata } from "next";
+import Image from "next/image";
+
+import { Footer } from "../../../components/footer";
+import { Nav } from "../../../components/nav";
+import {
+  Article,
+  BackToBlog,
+  CodeBlock,
+  H2,
+  Lead,
+  P,
+  PostHeader,
+  PullQuote,
+  Signoff,
+} from "../../../components/prose";
+import { Divider, TopGlow } from "../../../components/shared";
+import { StickyNav } from "../../../components/sticky-nav";
+import { JsonLd } from "../../../components/json-ld";
+import { MCP_DEMO_COMMAND, SENTENCE } from "../../../components/landing/data";
+import { blogPostingJsonLd } from "../../../lib/seo";
+
+const TITLE = "A name on every fact";
+const DESCRIPTION =
+  "An agent said the return window was 30 days. It was 14. Nobody knew who had told it that. Atlas is the company facts your AI agents can trust, and the fix is a person's name on every one.";
+
+export const metadata: Metadata = {
+  title: TITLE,
+  description: DESCRIPTION,
+  authors: [{ name: "Matt Sywulak" }],
+  openGraph: {
+    title: TITLE,
+    description: SENTENCE,
+    url: "https://www.useatlas.dev/blog/a-name-on-every-fact",
+    siteName: "Atlas",
+    type: "article",
+    publishedTime: "2026-09-10",
+    authors: ["Matt Sywulak"],
+    images: [{ url: "/og.png", width: 1200, height: 630, alt: TITLE }],
+  },
+  twitter: {
+    card: "summary_large_image",
+    title: TITLE,
+    description: SENTENCE,
+    images: ["/og.png"],
+  },
+  alternates: { canonical: "https://www.useatlas.dev/blog/a-name-on-every-fact" },
+};
+
+/**
+ * The recording's slot. The take (#5605) is a dated GIF under public/launch/;
+ * this element is where it renders, so the post and the README embed the same
+ * file.
+ */
+function Recording() {
+  return (
+    <figure className="my-8">
+      <Image
+        src="/launch/demo.gif"
+        alt="Sixty seconds in Claude Desktop: one command, one question about NovaMart's return window, an answer with a name and a date on it, the contradiction beside it, and the coverage page marking a channel nobody has surveyed."
+        className="w-full rounded-xl border border-code-border shadow-pane"
+        width={1280}
+        height={720}
+        unoptimized
+      />
+    </figure>
+  );
+}
+
+export default function ANameOnEveryFact() {
+  return (
+    <div className="relative min-h-screen">
+      <JsonLd data={blogPostingJsonLd("a-name-on-every-fact")} />
+      <StickyNav />
+      <TopGlow />
+      <Nav currentPage="/blog" />
+
+      <Article>
+        <PostHeader
+          tag="Launch"
+          isoDate="2026-09-10"
+          dateLabel="September 10, 2026"
+          readingTime="5 min read"
+          title={TITLE}
+          dek="An agent said the return window was 30 days. It was 14. Nobody knew who had told it that."
+        />
+
+        <Lead>
+          Ask a coding agent a question about your own company and it answers
+          with the same confidence it brings to a stack trace. Sometimes it is
+          right. When it is wrong, the interesting question is not why the
+          model guessed. It is why nothing in the loop could say who had stood
+          behind the fact it used.
+        </Lead>
+
+        <P>
+          Here is the case that made me build Atlas. A support agent, wired to
+          a company&apos;s chat history, told a customer the return window was
+          30 days. Finance had said so in a pinned message in July. Support had
+          changed the macro to 14 days in August and said, in the same channel,
+          to stop quoting 30. Both messages were in the index. The agent picked
+          one, said it plainly, and moved on. Nobody could tell, afterwards,
+          which message it had used, whether anyone had ever approved either as
+          policy, or which other channels it had never read at all.
+        </P>
+
+        <P>
+          That is the shape of the problem, and it is not a retrieval problem.
+          The field that calls itself company memory has converged on one
+          substrate: facts split from episodes, timestamps for when a thing
+          became true and when it stopped, provenance back to a source. The
+          benchmarks are saturating. Adoption is not, and the people building
+          these systems say so themselves. What is missing is not a better
+          score. It is a person.
+        </P>
+
+        <PullQuote>{SENTENCE}</PullQuote>
+
+        <H2>Three kinds of thing</H2>
+        <P>
+          Atlas holds three kinds of thing, and every answer says which one it
+          is drawing on. <strong>Surveyed</strong> facts are read straight from
+          the company&apos;s own data through a semantic layer a human wrote:
+          the query re-reads live rows, nobody interpreted anything, and the
+          answer cannot go stale between readings. <strong>Attested</strong>{" "}
+          facts were extracted from something someone wrote, then approved by a
+          named person in the company who is now on the record for it.{" "}
+          <strong>On the record</strong> is the raw material itself, unedited:
+          trustworthy as testimony, never as fact.
+        </P>
+        <P>
+          Two rules hold the three together. Surveyed outranks Attested
+          wherever they overlap, so a recollection never overwrites the data.
+          And nothing becomes Attested without a person choosing to make it so.
+          There is no setting that turns that off.
+        </P>
+
+        <H2>The tier that cannot be wrong</H2>
+        <P>
+          The warehouse is the part of a company that already knows what
+          happened, and it is the one source no extraction pipeline can improve
+          on. A number read from the orders table is not an opinion about the
+          orders table. It carries its own timestamps, it re-reads live on every
+          question, and the proof is the SQL and the rows. Atlas runs that
+          query read-only, one statement, whitelisted to the tables a human
+          named, behind seven layers of validation. That tier is the spine. Everything
+          written down by people sits above it, labelled, and yields to it.
+        </P>
+
+        <H2>Nothing counts until a person says so</H2>
+        <P>
+          Most memory systems write silently: a hook runs on every prompt, a
+          fact is scored, and a conflict is resolved toward the newest message.
+          It is a reasonable design and it is the one I refused. When the
+          newest message wins, the intern outranks the head of finance by
+          posting later. When a model resolves the conflict, the answer to
+          &ldquo;who decided this was true?&rdquo; is a confidence score.
+        </P>
+        <P>
+          In Atlas a contradiction is shown as a contradiction. The return
+          window is 30 days, says Priya Natarajan in #finance, on the 14th of
+          July. It is 14 days, says the support macro, on the 2nd of August.
+          Both claims, both sources, both names, and Atlas has picked neither.
+          A person closes it, or it stays open. And one page shows the shape
+          of what is known: which parts of the company are well surveyed, which
+          are thin, and which nobody has surveyed at all. It is most useful
+          exactly where it is empty.
+        </P>
+
+        <H2>What this costs</H2>
+        <P>
+          Two things, and I am paying both on purpose. The first is breadth.
+          A vendor that indexes everything will always win a connector count,
+          and if your job is to make ten thousand documents findable you should
+          buy that vendor. Atlas is a smaller set of facts you can actually
+          trust, and losing a breadth comparison is an expected outcome, not a
+          defect.
+        </P>
+        <P>
+          The second is adoption. Consulting Atlas is always an explicit act, a
+          tool call an agent makes and a person can see. Atlas does not silently
+          inject itself into other tools&apos; prompts, because a fact an agent acts
+          on should arrive through a door someone opened. That is slower to
+          spread than a hook that fires on every prompt. I think a company
+          fact with a name on it is worth the friction, and I would rather be
+          argued with on that than on a benchmark. The sharpest version of the
+          other position is{" "}
+          <a
+            href="https://news.ycombinator.com/item?id=48387095"
+            className="link-accent"
+          >
+            Hyper&apos;s own launch thread
+          </a>
+          , which is worth reading whole.
+        </P>
+
+        <H2>Sixty seconds</H2>
+        <P>
+          One command into Claude Desktop, Cursor or Continue, against a hosted
+          demo company. No account, no email.
+        </P>
+        <CodeBlock title="terminal">{MCP_DEMO_COMMAND}</CodeBlock>
+        <P>
+          Restart the client and ask what NovaMart&apos;s return window is.
+          The answer arrives with a name and a date on it, the contradiction
+          beside it, and neither side chosen for you. The whole of Atlas runs
+          on infrastructure you control, under a license that keeps the working
+          parts free.
+        </P>
+
+        <Recording />
+
+        <Signoff />
+
+        <BackToBlog />
+      </Article>
+
+      <Divider />
+      <Footer />
+    </div>
+  );
+}
diff --git a/apps/www/src/app/layout.tsx b/apps/www/src/app/layout.tsx
index 2fccc3acfb..d8bae18516 100644
--- a/apps/www/src/app/layout.tsx
+++ b/apps/www/src/app/layout.tsx
@@ -2,6 +2,7 @@ import type { Metadata } from "next";
 import { Sora, JetBrains_Mono } from "next/font/google";
 import "./globals.css";
 import WebMCP from "@/components/webmcp";
+import { SENTENCE } from "@/components/landing/data";
 
 const sora = Sora({
   subsets: ["latin"],
@@ -18,15 +19,13 @@ const jetbrainsMono = JetBrains_Mono({
 export const metadata: Metadata = {
   metadataBase: new URL("https://www.useatlas.dev"),
   title: {
-    default: "Atlas — Ask your data anything, trust the answer.",
+    default: "Atlas — the company facts your AI agents can trust",
     template: "%s — Atlas",
   },
-  description:
-    "Atlas is the AI data analyst you can run anywhere. It answers plain-English questions across your SQL warehouses and REST APIs, grounded in a semantic layer you control. Run it in the cloud or self-host it — open source.",
+  description: SENTENCE,
   openGraph: {
-    title: "Atlas — Ask your data anything, trust the answer.",
-    description:
-      "Atlas reads your semantic layer, writes the SQL, and runs it read-only behind 7 validators — SELECT-only, single-statement, whitelisted to your schema. Cloud or self-hosted.",
+    title: "Atlas — the company facts your AI agents can trust",
+    description: SENTENCE,
     url: "https://www.useatlas.dev",
     siteName: "Atlas",
     type: "website",
@@ -36,15 +35,14 @@ export const metadata: Metadata = {
         url: "/og.png",
         width: 1200,
         height: 630,
-        alt: "Atlas — Ask your data anything, trust the answer.",
+        alt: "Atlas — the company facts your AI agents can trust",
       },
     ],
   },
   twitter: {
     card: "summary_large_image",
-    title: "Atlas — Ask your data anything, trust the answer.",
-    description:
-      "Plain-English questions answered across your SQL warehouses and REST APIs, grounded in a semantic layer you control. Cloud or self-hosted, open source.",
+    title: "Atlas — the company facts your AI agents can trust",
+    description: SENTENCE,
     images: ["/og.png"],
   },
 };
diff --git a/apps/www/src/components/copy-command.tsx b/apps/www/src/components/copy-command.tsx
new file mode 100644
index 0000000000..881e803264
--- /dev/null
+++ b/apps/www/src/components/copy-command.tsx
@@ -0,0 +1,54 @@
+"use client";
+
+import { useState } from "react";
+
+type CopyState = "idle" | "copied" | "failed";
+
+const LABEL: Record<CopyState, string> = {
+  idle: "Copy",
+  copied: "Copied",
+  failed: "Select & copy",
+};
+
+/**
+ * A one-line command rendered as the page's primary call to action: a dark
+ * code pane (the brand's fixed terminal surface) with a copy button. Clipboard
+ * access can be denied (insecure context, permissions policy), so a failure
+ * is reported in the button label rather than swallowed.
+ */
+export function CopyCommand({ command, className = "" }: { command: string; className?: string }) {
+  const [state, setState] = useState<CopyState>("idle");
+
+  async function copy() {
+    try {
+      await navigator.clipboard.writeText(command);
+      setState("copied");
+      setTimeout(() => setState("idle"), 2000);
+    } catch (err) {
+      console.warn(
+        "[copy-command] clipboard write failed:",
+        err instanceof Error ? err.message : String(err),
+      );
+      setState("failed");
+    }
+  }
+
+  return (
+    <div
+      className={`flex items-center gap-3 overflow-hidden rounded-lg border border-code-border bg-code-bg pl-4 pr-2 shadow-pane ${className}`}
+    >
+      <span aria-hidden className="font-mono text-[13px] text-code-muted">$</span>
+      <code className="flex-1 select-all overflow-x-auto whitespace-nowrap py-3 font-mono text-[13px] text-code-fg">
+        {command}
+      </code>
+      <button
+        type="button"
+        onClick={copy}
+        aria-live="polite"
+        className="my-1.5 shrink-0 rounded-md bg-accent px-3 py-1.5 text-[12.5px] font-semibold text-accent-ink transition-colors hover:bg-accent-hover"
+      >
+        {LABEL[state]}
+      </button>
+    </div>
+  );
+}
diff --git a/apps/www/src/components/landing/big-stat.tsx b/apps/www/src/components/landing/big-stat.tsx
index cd1c89fed4..d1655a0756 100644
--- a/apps/www/src/components/landing/big-stat.tsx
+++ b/apps/www/src/components/landing/big-stat.tsx
@@ -1,3 +1,8 @@
+/**
+ * The Surveyed tier's introduction: the analyst is the tier that cannot be
+ * wrong. The "0" is writes — every query is SELECT-only — and the copy says
+ * why the data outranks anyone's recollection.
+ */
 export function BigStat() {
   return (
     <section
@@ -8,10 +13,11 @@ export function BigStat() {
         0
       </div>
       <div className="max-w-[720px]">
+        <p className="mb-3 font-mono text-[11px] tracking-[0.06em] text-accent">// surveyed — the tier that cannot be wrong</p>
         <p className="m-0 mb-3 text-xl md:text-[28px] font-medium leading-[1.3] tracking-[-0.02em] text-fg">
-          writes. Every query Atlas runs is SELECT-only, single-statement, and
-          whitelisted to your semantic layer — enforced by seven validators, not
-          suggested by a prompt.
+          writes. Every Surveyed answer is a SELECT-only, single-statement query,
+          whitelisted to a semantic layer you author, re-read from live rows —
+          so the data outranks what anyone remembers.
         </p>
         <p className="m-0 font-mono text-[12px] tracking-[0.04em] text-fg-muted">
           // empty check · mutation guard · AST parse · table whitelist · row-level security · auto LIMIT · statement timeout
diff --git a/apps/www/src/components/landing/data.ts b/apps/www/src/components/landing/data.ts
index 97f19a0f14..df9dea12bb 100644
--- a/apps/www/src/components/landing/data.ts
+++ b/apps/www/src/components/landing/data.ts
@@ -1,16 +1,81 @@
 /**
- * Shared sample data used across landing-page components.
+ * Shared copy and sample data for the landing page.
  *
- * Hoisted here so the hero answer card and the how-it-works reply pane, plus
- * any future surface that quotes the same NovaMart "top categories by GMV"
- * answer, read from a single source. No copy/paste drift between them.
+ * Hoisted here so every surface that quotes the sentence, the demo command or
+ * the NovaMart exchange reads from one source. The sentence is decided in
+ * docs/prd/launch-cycle.md and renders verbatim; the exchange mirrors the
+ * seeded demo corpus (packages/api/src/lib/brain/demo-corpus/corpus.ts), so
+ * what the page shows is what the hosted demo answers.
  */
+
+/** The launch cycle's sentence, verbatim (docs/prd/launch-cycle.md). */
+export const SENTENCE =
+  "Atlas is the company facts your AI agents can trust: every one carries its source, its date, and the name of the person who approved it. Open source, runs in your VPC.";
+
+/** The one command every launch surface prints — no account, no email. */
+export const MCP_DEMO_COMMAND = "bunx @useatlas/mcp init --hosted --demo --write";
+
+/** The demo's one question. */
+export const DEMO_QUESTION = "What is NovaMart's return window?";
+
+/**
+ * The demo exchange as the hosted NovaMart corpus answers it: the attested
+ * claim with its speaker, channel and date, and the contradiction Atlas shows
+ * beside it without picking a side. Dates and names are the corpus's own.
+ */
+export const DEMO_EXCHANGE = {
+  attested: {
+    claim: "30 days from delivery, for every category.",
+    speaker: "Priya Natarajan",
+    role: "Head of Finance",
+    channel: "#finance",
+    date: "14 Jul 2026",
+  },
+  contradiction: {
+    claim: "14 days from delivery. Please stop quoting 30.",
+    speaker: "Marcus Adeyemi",
+    role: "Support",
+    channel: "#support",
+    date: "2 Aug 2026",
+  },
+} as const;
+
+/**
+ * The three kinds of thing that live in the Atlas, in the PRD's language
+ * (docs/prd/company-atlas.md § What a person can trust). Every answer says
+ * which one it is drawing on.
+ */
+type TrustTier = {
+  readonly name: string;
+  readonly what: string;
+  readonly why: string;
+};
+
+export const TRUST_TIERS: ReadonlyArray<TrustTier> = [
+  {
+    name: "Surveyed",
+    what: "Drawn directly from the company's own data.",
+    why: "True by construction — the answer re-reads the live rows. Nobody interpreted anything, so nothing can have been interpreted wrong, and it cannot go stale between readings.",
+  },
+  {
+    name: "Attested",
+    what: "Extracted from something someone wrote, then approved.",
+    why: "A named person in your company read this claim and stood behind it. That person is on the record.",
+  },
+  {
+    name: "On the record",
+    what: "The raw source material itself.",
+    why: "It is not a claim about what is true — it is what was actually said, unedited. Trustworthy as testimony, not as fact.",
+  },
+];
+
 export type CategoryRow = {
   readonly category: string;
   readonly gmv: string;
   readonly orders: string;
 };
 
+/** Sample rows for the Surveyed pane in how-it-works. */
 export const CATEGORY_ROWS: ReadonlyArray<CategoryRow> = [
   { category: "Bedding",     gmv: "$184,219", orders: "2,041" },
   { category: "Kitchen",     gmv: "$142,718", orders: "1,587" },
@@ -19,9 +84,6 @@ export const CATEGORY_ROWS: ReadonlyArray<CategoryRow> = [
   { category: "Accessories", gmv: "$54,011",  orders: "693"   },
 ];
 
-/**
- * The canonical demo question. Shared by the hero answer card and the
- * how-it-works reply pane so the prompt text never drifts between surfaces.
- */
+/** The canonical analyst question, for the Surveyed pane. */
 export const TOP_CATEGORY_QUESTION =
   "What's our top-performing category by GMV this month?";
diff --git a/apps/www/src/components/landing/drop-in-surfaces.tsx b/apps/www/src/components/landing/drop-in-surfaces.tsx
index 7c75efa4af..0883d6e73f 100644
--- a/apps/www/src/components/landing/drop-in-surfaces.tsx
+++ b/apps/www/src/components/landing/drop-in-surfaces.tsx
@@ -9,35 +9,6 @@ function DropInItem({ name, desc }: DropInProps) {
   );
 }
 
-/**
- * Queryable read datasources Atlas ships today (#3994). Verified against the
- * shipped connection/plugin code — every entry is a datasource you can connect
- * and query, not an action or write target. Elasticsearch and OpenSearch are
- * one unified engine, so they share a chip. Twenty (CRM action) and Obsidian
- * (a surface + vault-reader action) are intentionally absent — they aren't
- * analytics datasources.
- *
- * Postgres/MySQL are native (packages/api/src/lib/db/connection.ts); the
- * warehouses/search engines are `datasource`-typed plugins under plugins/;
- * Stripe/Notion/GitHub and arbitrary OpenAPI specs are REST/OpenAPI presets in
- * packages/api/src/lib/openapi/data-candidates.ts — so there is no plugins/
- * directory for them, and that's expected, not stale.
- */
-const DATASOURCES: ReadonlyArray<string> = [
-  "PostgreSQL",
-  "MySQL",
-  "ClickHouse",
-  "Snowflake",
-  "BigQuery",
-  "DuckDB",
-  "Elasticsearch / OpenSearch",
-  "Salesforce",
-  "Stripe",
-  "Notion",
-  "GitHub",
-  "any OpenAPI 3.x spec",
-];
-
 /**
  * The "drop-in surfaces" band. Carried over from the old Primitives section —
  * the four architecture cards were cut as too internal for a plain-language
@@ -88,27 +59,15 @@ export function DropInSurfaces() {
         </div>
       </div>
 
-      <div className="mt-8">
-        <p className="mb-3.5 font-mono text-[11px] tracking-[0.04em] text-fg-muted">
-          // connect anything — read-only, behind the same validators
-        </p>
-        <ul className="flex flex-wrap gap-2 p-0">
-          {DATASOURCES.map((name) => (
-            <li
-              key={name}
-              className="rounded-md border border-border px-2.5 py-1 font-mono text-[12px] text-fg-muted"
-              style={{ background: "var(--bg-raised)" }}
-            >
-              {name}
-            </li>
-          ))}
-        </ul>
-        <p className="mt-3 text-[12.5px] leading-[1.55] text-fg-muted">
-          SQL warehouses, search engines, and REST APIs all become datasources
-          you can ask in plain English. Bring any OpenAPI spec and Atlas reads
-          it like a table.
-        </p>
-      </div>
+      <p className="mt-6 text-[12.5px] leading-[1.55] text-fg-muted">
+        SQL warehouses, search engines and REST APIs all become datasources you
+        can ask in plain English, read-only, behind the same validators. The
+        full list lives in the docs:{" "}
+        <a href="https://docs.useatlas.dev/getting-started/connect-your-data" className="text-fg underline decoration-border-strong underline-offset-4 hover:text-accent">
+          connect your data
+        </a>
+        .
+      </p>
     </section>
   );
 }
diff --git a/apps/www/src/components/landing/end-cta.tsx b/apps/www/src/components/landing/end-cta.tsx
index 429d46a016..c82cbae7de 100644
--- a/apps/www/src/components/landing/end-cta.tsx
+++ b/apps/www/src/components/landing/end-cta.tsx
@@ -1,7 +1,10 @@
+import { CopyCommand } from "../copy-command";
+import { MCP_DEMO_COMMAND, SENTENCE } from "./data";
+
 /**
  * Closing CTA — the page's one deep-green "drenched" band: a full-bleed forest
  * ground with cream text and the bright brand teal as the spark, echoing the
- * hero headline as a bookend. The only place the dark/green register returns on
+ * sentence as a bookend. The only place the dark/green register returns on
  * the otherwise light page. See PRODUCT.md › Aesthetic Direction.
  */
 export function EndCta() {
@@ -14,34 +17,21 @@ export function EndCta() {
         color: "var(--drench-fg)",
       }}
     >
-      <div className="mx-auto max-w-[720px] text-center">
-        <h2 className="m-0 mb-8 text-[40px] md:text-[56px] font-semibold leading-[1.05] tracking-[-0.035em]">
-          <span className="block">Ask your data anything.</span>
-          <em className="block font-semibold" style={{ color: "var(--drench-accent)" }}>
-            Trust the answer.
-          </em>
+      <div className="mx-auto max-w-[760px] text-center">
+        <h2 className="m-0 mb-8 text-[26px] md:text-[34px] font-semibold leading-[1.2] tracking-[-0.025em]">
+          {SENTENCE}
         </h2>
-        <div className="flex flex-wrap justify-center gap-2.5">
-          <a
-            href="https://app.useatlas.dev/signup"
-            className="inline-flex items-center rounded-lg px-[18px] py-[11px] text-[13.5px] font-semibold transition-opacity hover:opacity-90"
-            style={{ background: "var(--drench-accent)", color: "var(--drench-ink)" }}
-          >
-            Start free trial →
+        <CopyCommand command={MCP_DEMO_COMMAND} className="mx-auto max-w-[560px] text-left" />
+        <p className="mt-4 text-[13px]" style={{ color: "var(--drench-muted)" }}>
+          No account, no email.{" "}
+          <a href="https://app.useatlas.dev/signup" className="underline underline-offset-4" style={{ color: "var(--drench-fg)" }}>
+            Start a hosted trial
+          </a>{" "}
+          — 14 days, no credit card — or{" "}
+          <a href="https://docs.useatlas.dev/self-hosted/getting-started/quick-start" className="underline underline-offset-4" style={{ color: "var(--drench-fg)" }}>
+            self-host
           </a>
-          <a
-            href="https://app.useatlas.dev/demo"
-            className="inline-flex items-center rounded-lg border bg-transparent px-3.5 py-2.5 text-[13.5px] transition-colors hover:bg-white/5"
-            style={{ borderColor: "oklch(1 0 0 / 0.22)", color: "var(--drench-fg)" }}
-          >
-            Try the NovaMart demo →
-          </a>
-        </div>
-        <p className="mt-3.5 text-[13px]" style={{ color: "var(--drench-muted)" }}>
-          14-day trial, no credit card. Or self-host — free and open source:{" "}
-          <code className="font-mono text-[12px]" style={{ color: "var(--drench-fg)" }}>
-            bun create atlas-agent
-          </code>
+          , free and open source.
         </p>
       </div>
     </section>
diff --git a/apps/www/src/components/landing/hero.tsx b/apps/www/src/components/landing/hero.tsx
index 753641e6d5..d47ae8026e 100644
--- a/apps/www/src/components/landing/hero.tsx
+++ b/apps/www/src/components/landing/hero.tsx
@@ -1,134 +1,117 @@
-import { Fragment } from "react";
-
-import { CATEGORY_ROWS, TOP_CATEGORY_QUESTION } from "./data";
-
-const HEADLINE_LINES = ["Ask your data anything.", "Trust the answer."] as const;
-const ITALIC_LINE_INDEX = 1;
-
-const SUBHEAD =
-  "Atlas is the AI data analyst you can run anywhere. It turns plain-English questions into trustworthy answers across your SQL warehouses and REST APIs, grounded in a semantic layer you control.";
+import { CopyCommand } from "../copy-command";
+import { DEMO_EXCHANGE, DEMO_QUESTION, MCP_DEMO_COMMAND, SENTENCE, TRUST_TIERS } from "./data";
 
 /**
- * The hero's payload: a plain-English question and the validated answer it
- * returns. No SQL here — the mechanism (YAML + generated SQL) lives in the
- * YAML section below, so the two surfaces don't duplicate.
+ * The hero visual: the demo exchange, rendered as the hosted NovaMart corpus
+ * answers it. The hero shows the exchange the recording shows, so the page
+ * and the take never disagree about what the demo says; the recording, when
+ * present, occupies this slot.
  */
-function AnswerCard() {
+function DemoExchange() {
+  const { attested, contradiction } = DEMO_EXCHANGE;
   return (
-    <div
-      className="relative overflow-hidden rounded-xl border border-white/10 shadow-pane"
-      style={{ background: "oklch(0.14 0 0)" }}
-    >
-      <div
-        className="flex items-center gap-2 border-b border-white/5 px-3.5 py-2.5"
-        style={{ background: "oklch(0.16 0 0)" }}
-      >
+    <div className="relative overflow-hidden rounded-xl border border-code-border bg-code-bg shadow-pane">
+      <div className="flex items-center gap-2 border-b border-code-border bg-code-chrome px-3.5 py-2.5">
         <span className="h-2 w-2 rounded-full" style={{ background: "var(--atlas-spark)" }} />
-        <span className="font-mono text-[11px] text-zinc-400">atlas · agent reply</span>
+        <span className="font-mono text-[11px] text-zinc-400">claude desktop · atlas-demo</span>
         <span className="ml-auto rounded border border-white/10 px-2 py-[2px] font-mono text-[10px] text-zinc-400">
-          chat · mcp · widget
+          searchAtlas
         </span>
       </div>
 
       <div className="flex flex-col gap-4 px-4 py-5">
         <div>
-          <p className="mb-1.5 font-mono text-[11px] tracking-[0.06em] text-brand">
-            // asked in plain english
-          </p>
-          <p className="m-0 text-[15px] leading-snug text-zinc-100">{TOP_CATEGORY_QUESTION}</p>
+          <p className="mb-1.5 font-mono text-[11px] tracking-[0.06em] text-brand">// you ask</p>
+          <p className="m-0 text-[15px] leading-snug text-zinc-100">{DEMO_QUESTION}</p>
         </div>
 
         <div>
-          <p className="mb-2 font-mono text-[11px] tracking-[0.06em] text-zinc-400">
-            // validated answer
-          </p>
-          <div
-            className="overflow-hidden rounded-md border border-white/10"
-            style={{ background: "oklch(0.10 0 0)" }}
-          >
-            <div className="grid grid-cols-[1fr_auto_auto] gap-4 border-b border-white/5 px-3 py-2 font-mono text-[11px] uppercase tracking-[0.06em] text-zinc-400">
-              <span>category</span>
-              <span className="text-right">gmv</span>
-              <span className="text-right">orders</span>
+          <p className="mb-2 font-mono text-[11px] tracking-[0.06em] text-zinc-400">// the answer, with a name on it</p>
+          <div className="rounded-md border border-code-border bg-code-well px-3 py-3">
+            <div className="mb-2 flex flex-wrap items-center gap-2">
+              <span className="rounded border border-brand/40 px-2 py-[2px] font-mono text-[10px] uppercase tracking-[0.06em] text-brand">
+                Attested
+              </span>
+              <span className="font-mono text-[11px] text-zinc-400">
+                {attested.speaker} · {attested.role} · {attested.channel} · {attested.date}
+              </span>
             </div>
-            {CATEGORY_ROWS.slice(0, 3).map((row, i) => (
-              <div
-                key={row.category}
-                className="grid grid-cols-[1fr_auto_auto] gap-4 px-3 py-2 font-mono text-[12.5px] text-zinc-200"
-                style={{ background: i % 2 ? "oklch(0.12 0 0)" : "transparent" }}
-              >
-                <span>{row.category}</span>
-                <span className="text-right text-brand">{row.gmv}</span>
-                <span className="text-right text-zinc-400">{row.orders}</span>
-              </div>
-            ))}
+            <p className="m-0 text-[14px] leading-snug text-zinc-100">{attested.claim}</p>
+            <p className="m-0 mt-2 font-mono text-[11px] text-zinc-400">
+              who said it · where · when — approved before it counted
+            </p>
           </div>
         </div>
 
-        <div className="flex flex-wrap items-center gap-x-2.5 gap-y-1.5 font-mono text-[11px] text-zinc-400">
-          <span className="flex items-center gap-1.5 text-brand">
-            <span aria-hidden>✓</span> 7 validators
-          </span>
-          <span aria-hidden className="text-zinc-700">·</span>
-          <span>read-only</span>
-          <span aria-hidden className="text-zinc-700">·</span>
-          <span>row-limited</span>
-          <span aria-hidden className="text-zinc-700">·</span>
-          <span>audited</span>
+        <div>
+          <p className="mb-2 font-mono text-[11px] tracking-[0.06em] text-zinc-400">// also attested — Atlas shows both and picks neither</p>
+          <div className="rounded-md border border-amber-300/25 bg-code-well px-3 py-3">
+            <div className="mb-2 flex flex-wrap items-center gap-2">
+              <span className="rounded border border-brand/40 px-2 py-[2px] font-mono text-[10px] uppercase tracking-[0.06em] text-brand">
+                Attested
+              </span>
+              <span className="rounded border border-amber-300/40 px-2 py-[2px] font-mono text-[10px] uppercase tracking-[0.06em] text-amber-200/90">
+                In tension
+              </span>
+              <span className="font-mono text-[11px] text-zinc-400">
+                {contradiction.speaker} · {contradiction.role} · {contradiction.channel} · {contradiction.date}
+              </span>
+            </div>
+            <p className="m-0 text-[14px] leading-snug text-zinc-100">{contradiction.claim}</p>
+          </div>
         </div>
       </div>
     </div>
   );
 }
 
-type Stage = { readonly label: string; readonly value: string; readonly highlight: boolean };
-
-const STAGES: ReadonlyArray<Stage> = [
-  { label: "ask",            value: '"top category by gmv…"', highlight: false },
-  { label: "semantic layer", value: "your YAML",              highlight: false },
-  { label: "validate",       value: "7 checks · read-only",   highlight: false },
-  { label: "answer",         value: "grounded · audited",     highlight: true  },
-];
-
 /**
- * The four-stage path: ask → semantic layer → validate → answer. One straight
- * read left-to-right; the highlight lands on the answer to reinforce the
- * headline's promise (a grounded, audited result you can trust).
+ * The three kinds of thing that live in the Atlas, in the PRD's language.
+ * Sits under the fold as the first thing a reader meets after the exchange,
+ * so the chip on the card above ("Attested") is explained one screen later.
  */
-function PipelineStrip() {
+function TierStrip() {
   return (
     <div className="animate-fade-in-up delay-400 mt-12 md:mt-16">
       <p className="mb-3.5 font-mono text-[11px] tracking-[0.04em] text-fg-muted">
-        // every question takes the same path
+        // three kinds of thing live in the Atlas — every answer says which it is drawing on
       </p>
-      <div className="grid gap-3 md:grid-cols-[1fr_auto_1fr_auto_1fr_auto_1fr] md:items-center">
-        {STAGES.map((stage, i) => (
-          <Fragment key={stage.label}>
-            <div
-              className="flex flex-col gap-1 rounded-lg border px-4 py-3.5"
-              style={{
-                background: stage.highlight ? "var(--accent-quiet)" : "var(--bg-raised)",
-                borderColor: stage.highlight ? "var(--accent)" : "var(--border)",
-              }}
-            >
-              <span className="text-[13px] font-semibold text-fg">{stage.label}</span>
-              <span className="font-mono text-[11px] text-fg-muted">{stage.value}</span>
-            </div>
-            {i < STAGES.length - 1 && (
-              <span aria-hidden className="hidden justify-center font-mono text-accent md:flex">
-                →
-              </span>
-            )}
-          </Fragment>
+      <div className="grid gap-3 md:grid-cols-3">
+        {TRUST_TIERS.map((tier) => (
+          <div
+            key={tier.name}
+            className="flex flex-col gap-1.5 rounded-lg border px-4 py-3.5"
+            style={{
+              background: tier.name === "Surveyed" ? "var(--accent-quiet)" : "var(--bg-raised)",
+              borderColor: tier.name === "Surveyed" ? "var(--accent)" : "var(--border)",
+            }}
+          >
+            <span className="text-[13px] font-semibold text-fg">{tier.name}</span>
+            <span className="text-[12.5px] leading-[1.5] text-fg">{tier.what}</span>
+            <span className="text-[12px] leading-[1.5] text-fg-muted">{tier.why}</span>
+          </div>
         ))}
       </div>
+      <p className="mt-3 text-[12.5px] text-fg-muted">
+        Surveyed outranks Attested wherever they overlap: a recollection never overwrites the data.
+        Nothing becomes Attested without a person approving it, and there is no setting that turns that off.
+      </p>
     </div>
   );
 }
 
+// Split for line layout only. The tail keeps its leading space so the h1's
+// text content — what assistive tech and scrapers read — is the sentence
+// character for character.
+const [SENTENCE_HEAD, SENTENCE_TAIL] = (() => {
+  const i = SENTENCE.indexOf(":");
+  if (i < 0) return [SENTENCE, ""];
+  return [SENTENCE.slice(0, i + 1), SENTENCE.slice(i + 1)];
+})();
+
 export function Hero() {
   return (
-    <section className="relative overflow-hidden border-b border-border-soft px-content pt-16 pb-16 md:pt-24 md:pb-20">
+    <section className="relative overflow-hidden border-b border-border-soft px-content pt-14 pb-16 md:pt-20 md:pb-20">
       <div
         aria-hidden
         className="pointer-events-none absolute -top-32 right-0 h-[560px] w-[560px] rounded-full"
@@ -138,56 +121,48 @@ export function Hero() {
       />
 
       {/*
-        The above-the-fold hero text (h1, subhead, CTAs, trial note) paints
-        immediately — no `animate-fade-in-up`. That keyframe starts at
-        `opacity: 0`, and Chrome's LCP algorithm won't register an element
-        painted transparent as an LCP candidate at its paint time. With the
-        headline (the mobile LCP element) fading in, Lighthouse mobile reported
-        `NO_LCP` and an inflated Speed Index (content still settling ~2.6s after
-        FCP). Entrance polish stays on below-the-fold / non-LCP surfaces
-        (PipelineStrip); the decorative AnswerCard also paints immediately so it
-        can't become a late LCP candidate on narrow viewports.
+        The above-the-fold text (h1, command, links) paints immediately — no
+        `animate-fade-in-up`. That keyframe starts at `opacity: 0`, and
+        Chrome's LCP algorithm won't register an element painted transparent
+        as an LCP candidate at its paint time (Lighthouse reported `NO_LCP`
+        when the headline faded in). Entrance polish stays below the fold
+        (TierStrip); the DemoExchange also paints immediately so it can't
+        become a late LCP candidate on narrow viewports.
       */}
       <div className="relative grid gap-10 md:grid-cols-2 md:items-center md:gap-12">
-        <div className="max-w-[520px]">
-          <h1 className="m-0 text-[44px] sm:text-[56px] md:text-[64px] font-semibold leading-[1.02] tracking-[-0.035em] text-fg">
-            {HEADLINE_LINES.map((line, i) => (
-              <span key={line} className="block">
-                {i === ITALIC_LINE_INDEX ? (
-                  <em className="font-semibold text-accent">{line}</em>
-                ) : (
-                  line
-                )}
-              </span>
-            ))}
+        <div className="max-w-[560px]">
+          <h1 className="m-0 text-[30px] sm:text-[36px] md:text-[40px] font-semibold leading-[1.12] tracking-[-0.03em] text-fg">
+            <span className="block">{SENTENCE_HEAD}</span>
+            <span className="block text-fg-muted">{SENTENCE_TAIL}</span>
           </h1>
-          <p className="mt-6 max-w-[460px] text-base leading-[1.6] text-fg-muted">{SUBHEAD}</p>
-          <div className="mt-7 flex flex-wrap gap-2.5">
-            <a
-              href="https://app.useatlas.dev/signup"
-              className="inline-flex items-center gap-2 rounded-lg bg-accent px-[18px] py-[11px] text-[13.5px] font-semibold text-accent-ink transition-colors hover:bg-accent-hover"
-            >
-              Start free trial →
+
+          <p className="mt-6 text-[13px] text-fg-muted">
+            One command into Claude Desktop, Cursor or Continue. No account, no email. Then ask:{" "}
+            <em className="text-fg">{DEMO_QUESTION}</em>
+          </p>
+          <CopyCommand command={MCP_DEMO_COMMAND} className="mt-3" />
+
+          <p className="mt-4 text-[13px] text-fg-muted">
+            <a href="https://app.useatlas.dev/demo" className="text-fg underline decoration-border-strong underline-offset-4 hover:text-accent">
+              Try it in the browser
             </a>
-            <a
-              href="https://app.useatlas.dev/demo"
-              className="inline-flex items-center rounded-lg border border-border bg-transparent px-3.5 py-2.5 text-[13.5px] text-fg transition-colors hover:border-border-strong hover:bg-bg-sunken"
-            >
-              Try the live demo →
+            <span aria-hidden className="mx-2 text-border-strong">·</span>
+            <a href="https://docs.useatlas.dev/self-hosted/getting-started/quick-start" className="text-fg underline decoration-border-strong underline-offset-4 hover:text-accent">
+              Self-host it
+            </a>
+            <span aria-hidden className="mx-2 text-border-strong">·</span>
+            <a href="https://docs.useatlas.dev/guides/mcp" className="text-fg underline decoration-border-strong underline-offset-4 hover:text-accent">
+              MCP guide
             </a>
-          </div>
-          <p className="mt-3.5 text-[13px] text-fg-muted">
-            14-day trial, no credit card. Or self-host — free and open source:{" "}
-            <code className="font-mono text-[12px] text-fg">bun create atlas-agent</code>
           </p>
         </div>
 
         <div className="relative">
-          <AnswerCard />
+          <DemoExchange />
         </div>
       </div>
 
-      <PipelineStrip />
+      <TierStrip />
     </section>
   );
 }
diff --git a/apps/www/src/components/landing/how-it-works.tsx b/apps/www/src/components/landing/how-it-works.tsx
index 60cdd06d10..40a366b089 100644
--- a/apps/www/src/components/landing/how-it-works.tsx
+++ b/apps/www/src/components/landing/how-it-works.tsx
@@ -161,8 +161,9 @@ function AnswerPane() {
         </div>
 
         <div>
-          <p className="mb-2 font-mono text-[11px] tracking-[0.06em] text-zinc-400">
-            // result · {CATEGORY_ROWS.length} rows · 7 validators passed
+          <p className="mb-2 flex flex-wrap items-center gap-2 font-mono text-[11px] tracking-[0.06em] text-zinc-400">
+            <span className="rounded border border-brand/40 px-2 py-[2px] text-[10px] uppercase text-brand">Surveyed</span>
+            <span>// result · {CATEGORY_ROWS.length} rows · 7 validators passed</span>
           </p>
           <div
             className="flex flex-wrap items-baseline gap-x-2.5 gap-y-1 rounded-md border border-white/10 px-3 py-2.5 font-mono text-[12.5px]"
@@ -195,15 +196,17 @@ export function HowItWorks() {
       style={{ background: "var(--bg-raised)" }}
     >
       <header className="mb-10 max-w-[760px]">
+        <p className="mb-3 font-mono text-[11px] tracking-[0.06em] text-accent">// surveyed</p>
         <h2 className="m-0 mb-4 text-[36px] md:text-[46px] font-semibold leading-[1.05] tracking-[-0.03em] text-fg">
-          Why you can trust the answer.
+          Why a Surveyed answer cannot be wrong.
         </h2>
         <p className="m-0 text-base leading-[1.65] text-fg-muted">
           You define your data once in YAML — the entities, joins, and the terms
           your team actually uses. Atlas reads it on every question, writes the
           SQL, and runs it read-only behind seven validators, so you see exactly
-          what ran. Self-host Atlas and that YAML lives in your own repo,
-          versioned and reviewed in pull requests like the rest of your code.
+          what ran and the rows it read. Self-host Atlas and that YAML lives in
+          your own repo, versioned and reviewed in pull requests like the rest of
+          your code.
         </p>
       </header>
 
diff --git a/apps/www/src/components/webmcp.tsx b/apps/www/src/components/webmcp.tsx
index 25ade25db6..127752b256 100644
--- a/apps/www/src/components/webmcp.tsx
+++ b/apps/www/src/components/webmcp.tsx
@@ -2,6 +2,8 @@
 
 import { useEffect } from "react";
 
+import { DEMO_QUESTION, MCP_DEMO_COMMAND, SENTENCE } from "./landing/data";
+
 // Provides tools to in-browser agents via Chrome's experimental webmcp
 // proposal (navigator.modelContext). No-op elsewhere. The response
 // envelope mirrors MCP server tools/call (`{ content: [{type, text}] }`)
@@ -42,6 +44,22 @@ function openAndReport(url: string, openedText: string): {
 }
 
 const TOOLS: ModelContextTool[] = [
+  {
+    name: "atlas_connect_agent_to_demo",
+    description:
+      `${SENTENCE} Returns the one terminal command that connects an MCP client (Claude Desktop, Cursor, Continue, or any other via --client generic) to the hosted NovaMart demo with no account and no email, plus the question to ask first.`,
+    inputSchema: { type: "object", properties: {}, additionalProperties: false },
+    async execute() {
+      return {
+        content: [
+          {
+            type: "text",
+            text: `Run this in a terminal, then restart the MCP client:\n\n${MCP_DEMO_COMMAND}\n\nThen ask: "${DEMO_QUESTION}" The answer names who said it, where and when, marks that a person approved it before it counted, and shows the contradiction between Finance and Support without picking a side. Guide: https://docs.useatlas.dev/guides/mcp`,
+          },
+        ],
+      };
+    },
+  },
   {
     name: "atlas_start_free_trial",
     description:
@@ -58,7 +76,7 @@ const TOOLS: ModelContextTool[] = [
   {
     name: "atlas_open_live_demo",
     description:
-      "Open the Atlas live demo at app.useatlas.dev/demo against a sample database. No account required — just an email.",
+      "Open the Atlas web demo at app.useatlas.dev/demo in a browser. No account, but it asks for an email; for a no-email path use atlas_connect_agent_to_demo.",
     inputSchema: { type: "object", properties: {}, additionalProperties: false },
     async execute() {
       const url = "https://app.useatlas.dev/demo";
diff --git a/apps/www/src/data/posts.ts b/apps/www/src/data/posts.ts
index 38b08a9fde..1feb60085a 100644
--- a/apps/www/src/data/posts.ts
+++ b/apps/www/src/data/posts.ts
@@ -34,6 +34,16 @@ export interface Post {
 }
 
 export const POSTS: Post[] = [
+  {
+    slug: "a-name-on-every-fact",
+    title: "A name on every fact",
+    description:
+      "An agent said the return window was 30 days. It was 14. Nobody knew who had told it that. Atlas is the company facts your AI agents can trust, and the fix is a person's name on every one.",
+    isoDate: "2026-09-10",
+    dateLabel: "September 10, 2026",
+    readingTime: "5 min read",
+    tag: "Launch",
+  },
   {
     slug: "atlas-is-public",
     title: "Atlas is public",
diff --git a/docs/development/testing.md b/docs/development/testing.md
index 2d0acba76d..72e997272a 100644
--- a/docs/development/testing.md
+++ b/docs/development/testing.md
@@ -39,7 +39,7 @@ cd packages/api && bun test --parallel --changed=HEAD~3        # last-3-commit w
 
 ## Pre-PR gates via `/ci`
 
-`/ci` runs `scripts/ci-local.sh` — 43 ci-local gates (lint, type, type-aware lint, syncpack, template/schema/openapi drift, the `check-*.sh` guards, and the full test suite, among others). All must pass before opening a PR; `.claude/commands/ci.md` carries the authoritative roster, and this page deliberately does not keep a second copy of it. In CI the api suite is sharded 4-way (`--shard=N/4`, duration-balanced via `--timings`); locally it runs unsharded.
+`/ci` runs `scripts/ci-local.sh` — 44 ci-local gates (lint, type, type-aware lint, syncpack, template/schema/openapi drift, the `check-*.sh` guards, and the full test suite, among others). All must pass before opening a PR; `.claude/commands/ci.md` carries the authoritative roster, and this page deliberately does not keep a second copy of it. In CI the api suite is sharded 4-way (`--shard=N/4`, duration-balanced via `--timings`); locally it runs unsharded.
 
 > ⚠️ **A clean local pass now requires `TEST_DATABASE_URL` (#5410).** Without it every `*-pg.test.ts` file self-skips — 104 real-postgres suites today, 1,432 assertions when last measured (2026-08-24, at 87 of them) — so the `test` gate reports `DECLINED` (exit 3) rather than PASS, and the run is not a clean pre-PR pass however green the other rows look. `bun run db:up && export TEST_DATABASE_URL=postgresql://atlas:atlas@localhost:5432/atlas`. The `/ci` skill carries the authoritative gate list and count.
 
diff --git a/scripts/check-launch-sentence.sh b/scripts/check-launch-sentence.sh
new file mode 100755
index 0000000000..18660381e9
--- /dev/null
+++ b/scripts/check-launch-sentence.sh
@@ -0,0 +1,52 @@
+#!/usr/bin/env bash
+# check-launch-sentence.sh — the launch cycle's sentence renders verbatim on
+# every surface that prints it (#5606).
+#
+# docs/prd/launch-cycle.md decides the sentence once. The README, the landing
+# page, the docs intro, the comparisons index and both llms.txt surfaces each
+# carry it as a literal, and nothing but a reader's eye kept them identical —
+# a one-word edit meant seven hand edits across three packages, and the
+# reviews of #5628/#5631/#5632 each asked for a gate on the plugin-count model.
+#
+# The PRD's blockquote is the source of truth. Every surface below must contain
+# that exact string. Add a surface here when a new one starts printing the
+# sentence; remove it when it stops.
+set -euo pipefail
+
+cd "$(git rev-parse --show-toplevel)"
+
+PRD="docs/prd/launch-cycle.md"
+SENTENCE="$(grep -m1 -oE '^> \*\*Atlas is the company facts.*\*\*$' "$PRD" | sed -E 's/^> \*\*//; s/\*\*$//')"
+if [ -z "$SENTENCE" ]; then
+  echo "check-launch-sentence: could not read the sentence from $PRD (expected a '> **Atlas is the company facts…**' blockquote)." >&2
+  exit 1
+fi
+echo ":: The sentence, from $PRD: ${SENTENCE}"
+
+SURFACES=(
+  "README.md"
+  "apps/www/src/components/landing/data.ts"
+  "apps/www/public/llms.txt"
+  "apps/docs/src/lib/llms-surface.ts"
+  "apps/docs/content/docs/index.mdx"
+  "apps/docs/content/shared/comparisons/index.mdx"
+)
+
+fail=0
+for surface in "${SURFACES[@]}"; do
+  if [ ! -f "$surface" ]; then
+    echo "check-launch-sentence: listed surface not found: $surface" >&2
+    fail=1
+    continue
+  fi
+  if ! grep -qF -- "$SENTENCE" "$surface"; then
+    echo "$surface: does not contain the sentence verbatim." >&2
+    echo "  Paste it exactly as $PRD states it, or drop this file from SURFACES if it no longer prints the sentence." >&2
+    fail=1
+  fi
+done
+
+if [ "$fail" -ne 0 ]; then
+  exit 1
+fi
+echo "Launch-sentence check passed — every surface carries the sentence verbatim."
diff --git a/scripts/check-plugin-count.sh b/scripts/check-plugin-count.sh
index ac50d423c9..fbb4510437 100755
--- a/scripts/check-plugin-count.sh
+++ b/scripts/check-plugin-count.sh
@@ -109,7 +109,6 @@ echo ":: Canonical plugin count derived from plugins/: ${N} (${typed_dirs[*]})"
 SURFACES=(
   "apps/www/src/components/landing/comparison.tsx"
   "apps/www/src/app/blog/announcing-atlas/page.tsx"
-  "apps/www/public/llms.txt"
   "apps/www/content/social/twitter-launch.md"
   "apps/www/content/social/linkedin-launch.md"
   "apps/docs/content/shared/comparisons/metabase.mdx"
diff --git a/scripts/ci-local.sh b/scripts/ci-local.sh
index bcb1254acd..74550d58ee 100755
--- a/scripts/ci-local.sh
+++ b/scripts/ci-local.sh
@@ -332,6 +332,7 @@ launch gate-fixtures-wired       bash scripts/check-gate-fixtures-wired.sh
 launch runtime-stage-upgrades    bash scripts/check-runtime-stage-upgrades.sh
 launch pricing-parity            bash scripts/check-pricing-parity.sh
 launch plugin-count              bash scripts/check-plugin-count.sh
+launch launch-sentence           bash scripts/check-launch-sentence.sh
 launch plugin-lockstep           bun scripts/check-plugin-lockstep.ts
 launch enforcement-parity        bash scripts/check-enforcement-parity.sh
 launch schema-drift              bash scripts/check-schema-drift.sh