diff --git a/Cargo.lock b/Cargo.lock
index 86e13e2..305976a 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -93,6 +93,7 @@ version = "0.1.0"
 dependencies = [
  "cb-kernel",
  "games-ground",
+ "quick-js",
  "serde_json",
 ]
 
@@ -104,6 +105,16 @@ dependencies = [
  "games-ground",
 ]
 
+[[package]]
+name = "cc"
+version = "1.4.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5add81bb678e6cb321aff7fa0dc7689ad82b112dbc032cea19f91d6b8e3582b9"
+dependencies = [
+ "find-msvc-tools",
+ "shlex",
+]
+
 [[package]]
 name = "cfg-if"
 version = "1.0.4"
@@ -162,6 +173,15 @@ version = "1.1.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "c8d4a3bb8b1e0c1050499d1815f5ab16d04f0959b233085fb31653fbfc9d98f9"
 
+[[package]]
+name = "copy_dir"
+version = "0.1.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "543d1dd138ef086e2ff05e3a48cf9da045da2033d16f8538fd76b86cd49b2ca3"
+dependencies = [
+ "walkdir",
+]
+
 [[package]]
 name = "cpufeatures"
 version = "0.2.17"
@@ -243,6 +263,12 @@ version = "1.0.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "877a4ace8713b0bcf2a4e7eec82529c029f1d0619886d18145fea96c3ffe5c0f"
 
+[[package]]
+name = "find-msvc-tools"
+version = "0.1.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5baebc0774151f905a1a2cc41989300b1e6fbb29aff0ceffa1064fdd3088d582"
+
 [[package]]
 name = "games-ground"
 version = "0.1.0"
@@ -331,6 +357,16 @@ version = "0.2.189"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "3eaf3ede3fee6db1a4c2ee091bf8a8b4dccdc6d17f656fb07896ee72867612f2"
 
+[[package]]
+name = "libquickjs-sys"
+version = "0.9.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3f0b24e9bd171b75ae0295bd428fb8fe58410fb23156e5f34a4657a70c3cee96"
+dependencies = [
+ "cc",
+ "copy_dir",
+]
+
 [[package]]
 name = "memchr"
 version = "2.8.3"
@@ -376,6 +412,16 @@ dependencies = [
  "unicode-ident",
 ]
 
+[[package]]
+name = "quick-js"
+version = "0.4.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "19cb4cefcb00f4ab9b332664d06005a74f582ac16aa959c6ad5912957bd83e5f"
+dependencies = [
+ "libquickjs-sys",
+ "once_cell",
+]
+
 [[package]]
 name = "quote"
 version = "1.0.47"
@@ -512,6 +558,12 @@ dependencies = [
  "digest",
 ]
 
+[[package]]
+name = "shlex"
+version = "2.0.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f8fadd59c855ef2080decdef8ff161eb6661b86933c9d82e5ba29dc602a55aba"
+
 [[package]]
 name = "syn"
 version = "2.0.119"
diff --git a/INTENT.md b/INTENT.md
index ebf9939..8fb0fbd 100644
--- a/INTENT.md
+++ b/INTENT.md
@@ -64,6 +64,11 @@ game.
    and scenario tests, simple bots. No rendering, no physics.
 1. **Inspectable 2D table** — card/token/hand/relationship-graph
    visualization, drag-to-propose, debug inspector, hot-seat play.
+   *Open on one human verification (CB-EV-0012 §4): the inspector,
+   drag-to-propose and hot-seat play are evidenced by executing code;
+   the visualization is evidenced only as correctly emitted, because no
+   browser is available to the loop. Run `cb-play --serve 0`, open the
+   printed URL, and confirm the table reads and a drag works.*
 2. **Physical 3D tabletop** — wgpu renderer, Rapier-backed physics, camera
    and pointer controls, snap zones, asset importer.
 3. **Networked sessions** — authoritative host, private projections,
diff --git a/WORK-RECORDS.md b/WORK-RECORDS.md
index ff6ed48..adca1ae 100644
--- a/WORK-RECORDS.md
+++ b/WORK-RECORDS.md
@@ -100,3 +100,6 @@
 | task | CB-WP-0013-T02 | done | — | workplans/CB-WP-0013-instrument-corrections.md |
 | task | CB-WP-0013-T03 | done | — | workplans/CB-WP-0013-instrument-corrections.md |
 | task | CB-WP-0013-T04 | done | — | workplans/CB-WP-0013-instrument-corrections.md |
+| task | CB-WP-0014-T01 | done | — | workplans/CB-WP-0014-execute-the-javascript.md |
+| task | CB-WP-0014-T02 | done | — | workplans/CB-WP-0014-execute-the-javascript.md |
+| task | CB-WP-0014-T03 | done | — | workplans/CB-WP-0014-execute-the-javascript.md |
diff --git a/crates/cb-render-html/Cargo.toml b/crates/cb-render-html/Cargo.toml
index 4435a33..204baaa 100644
--- a/crates/cb-render-html/Cargo.toml
+++ b/crates/cb-render-html/Cargo.toml
@@ -4,16 +4,28 @@ edition.workspace = true
 version.workspace = true
 license-file.workspace = true
 
-# ADR-0007 Decision 1: the browser is the renderer, so this crate has
-# **no third-party dependencies at all** beyond what the game already
+[features]
+# ADR-0009: the JS harness is exposed to cb-play's tests behind a feature,
+# so the engine stays a dev cost there too and never a shipped one.
+js-harness = ["dep:quick-js"]
+
+# ADR-0007 Decision 1: the browser is the renderer, so the rendering path
+# has **no third-party dependencies at all** beyond what the game already
 # carries. Adding one here needs an argument against ADR-0007 §Decision 3.
 [dependencies]
 cb-kernel.workspace = true
 games-ground.workspace = true
-
-[lints]
-workspace = true
+# ADR-0009: an EMBEDDED engine, not `node` — CI runs on rust:1.97, which
+# has no node, so requiring one would make our build acquire a runtime
+# nobody audits while it scored zero on the only instrument that governs
+# dependencies. Optional, and never in the shipped-runtime configuration:
+# AM-4a measures 157,202 against 161,000 and has no room for it.
+quick-js = { version = "0.4", optional = true }
 
 [dev-dependencies]
 # The coverage gate walks the serialized view; nothing else needs it.
 serde_json.workspace = true
+quick-js = "0.4"
+
+[lints]
+workspace = true
diff --git a/crates/cb-render-html/src/jsrun.rs b/crates/cb-render-html/src/jsrun.rs
new file mode 100644
index 0000000..484462b
--- /dev/null
+++ b/crates/cb-render-html/src/jsrun.rs
@@ -0,0 +1,290 @@
+//! Execute the emitted JavaScript for real (ADR-0009).
+//!
+//! ADR-0007 control 5 says the page **may not construct commands**: it
+//! reports raw pointer facts and Rust decides what they mean. Until now
+//! that contract was held up by a test that greps the emitted script for
+//! game vocabulary — and grepping for the absence of words is a weak
+//! proxy for *"this code cannot construct a command"*.
+//!
+//! This runs it. QuickJS, a DOM stub with exactly the surface `SCRIPT`
+//! touches, and a Rust callback standing in for `fetch` so the test can
+//! see precisely what would have gone on the wire.
+//!
+//! **The scripts are lifted from a real emitted document, not pasted
+//! here.** Both `<script>` blocks are extracted and evaluated in order, so
+//! the endpoint the JavaScript posts to is the one the page actually
+//! carried — token included. A harness that supplied its own endpoint
+//! would pass while the page shipped a broken one.
+//!
+//! What this does **not** prove: that the SVG renders legibly, that a drag
+//! feels like a drag, or that anyone can play a game. QuickJS has no
+//! layout engine (ADR-0009 §What is bought).
+
+use std::sync::{Arc, Mutex};
+
+/// One intercepted `fetch`, exactly as the page would have sent it.
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub struct Posted {
+    pub url: String,
+    pub body: String,
+}
+
+/// Pull every `<script>…</script>` body out of a document, in order.
+pub fn scripts(html: &str) -> Vec<String> {
+    let mut out = Vec::new();
+    let mut rest = html;
+    while let Some(i) = rest.find("<script>") {
+        let after = &rest[i + "<script>".len()..];
+        match after.find("</script>") {
+            Some(j) => {
+                out.push(after[..j].to_string());
+                rest = &after[j..];
+            }
+            None => break,
+        }
+    }
+    out
+}
+
+/// The DOM surface `SCRIPT` touches, and nothing more.
+///
+/// Deliberately minimal: every additional stubbed API is a way for the
+/// script to do something in the test that it could not do in a browser,
+/// or vice versa. If `SCRIPT` ever needs more than this, that is a signal
+/// about the script, not about the stub.
+const DOM: &str = r#"
+var __handlers = {};
+var __status = { textContent: "" };
+var document = {
+  addEventListener: function (type, fn) { __handlers[type] = fn; },
+  getElementById: function (id) { return __status; }
+};
+var window = { location: { reload: function () { __reloaded(); } } };
+function fetch(url, opts) {
+  __post(url, (opts && opts.body) || "");
+  // The script chains .then(...).then(...); give it something to chain on
+  // without ever resolving, so response handling stays out of scope here.
+  var chainable = { then: function () { return chainable; } };
+  return chainable;
+}
+function __down(id) { __handlers['pointerdown']({ target: { id: id } }); }
+function __up(id) { __handlers['pointerup']({ target: { id: id } }); }
+"#;
+
+/// Run the document's scripts, then a pointer gesture, and report what
+/// the page tried to send.
+///
+/// `gesture` is `(down_id, up_id)`. Errors are JS errors, and they are
+/// returned rather than swallowed: a script that throws must not look
+/// like a script that sent nothing.
+pub fn gesture(html: &str, down: &str, up: &str) -> Result<Vec<Posted>, String> {
+    let ctx = quick_js::Context::new().map_err(|e| format!("quickjs init: {e}"))?;
+
+    // `Arc<Mutex<_>>` rather than `Rc<RefCell<_>>`: quick-js requires the
+    // callback to be unwind-safe, since a panic inside it would cross the
+    // C boundary.
+    let posted: Arc<Mutex<Vec<Posted>>> = Arc::new(Mutex::new(Vec::new()));
+    let sink = posted.clone();
+    ctx.add_callback("__post", move |url: String, body: String| {
+        sink.lock().expect("posted").push(Posted { url, body });
+        0i32
+    })
+    .map_err(|e| format!("register __post: {e}"))?;
+    ctx.add_callback("__reloaded", || 0i32)
+        .map_err(|e| format!("register __reloaded: {e}"))?;
+
+    ctx.eval(DOM).map_err(|e| format!("dom stub: {e}"))?;
+
+    let found = scripts(html);
+    if found.is_empty() {
+        return Err("the document carries no <script> block".to_string());
+    }
+    for (i, s) in found.iter().enumerate() {
+        ctx.eval(s)
+            .map_err(|e| format!("script {i} of {}: {e}", found.len()))?;
+    }
+
+    // If the page never registered handlers, the gesture below would be a
+    // no-op and the test would read as "sent nothing" rather than "the
+    // page is broken". Distinguish the two.
+    let armed: bool = ctx
+        .eval_as("!!(__handlers['pointerdown'] && __handlers['pointerup'])")
+        .map_err(|e| format!("handler check: {e}"))?;
+    if !armed {
+        return Err("the page registered no pointer handlers".to_string());
+    }
+
+    ctx.eval(&format!(
+        "__down({}); __up({});",
+        json_lit(down),
+        json_lit(up)
+    ))
+    .map_err(|e| format!("gesture: {e}"))?;
+
+    let out = posted.lock().expect("posted").clone();
+    Ok(out)
+}
+
+fn json_lit(s: &str) -> String {
+    let mut out = String::with_capacity(s.len() + 2);
+    out.push('"');
+    for c in s.chars() {
+        match c {
+            '"' => out.push_str("\\\""),
+            '\\' => out.push_str("\\\\"),
+            c => out.push(c),
+        }
+    }
+    out.push('"');
+    out
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use cb_kernel::PlayerId;
+    use games_ground::{Action, GroundCommand};
+
+    fn page() -> String {
+        let legal = vec![
+            GroundCommand::SelectAction {
+                action: Action::Ground,
+                target: None,
+                problem: None,
+            },
+            GroundCommand::SelectAction {
+                action: Action::Attack,
+                target: Some(PlayerId(1)),
+                problem: None,
+            },
+        ];
+        crate::doc::document(
+            &crate::testfix::view(Some(PlayerId(0))),
+            &legal,
+            "/command?t=deadbeef",
+            Some(PlayerId(0)),
+            false,
+        )
+    }
+
+    /// **The one this pass exists for.** The emitted JavaScript runs, and
+    /// what it puts on the wire is two element ids and nothing else.
+    #[test]
+    fn the_emitted_javascript_reports_element_ids_and_nothing_else() {
+        let posts = gesture(&page(), "action-attack", "seat-1").expect("the script runs");
+        assert_eq!(posts.len(), 1, "expected exactly one fetch, got {posts:?}");
+        let p = &posts[0];
+
+        // The endpoint comes from the page, token and all — not from the
+        // harness. A page that shipped a broken endpoint fails here.
+        assert_eq!(p.url, "/command?t=deadbeef", "the page's own endpoint");
+        assert_eq!(p.body, "down=action-attack&up=seat-1");
+
+        // Control 5, asserted rather than grepped.
+        //
+        // The naive check — "the body must not contain 'attack'" — is
+        // WRONG, and writing it was instructive: the body legitimately
+        // contains `action-attack`, because that is the id of an element
+        // a finger landed on. An element may name an action; that is not
+        // the page deciding anything.
+        //
+        // What actually distinguishes reporting from deciding is the
+        // *shape*: exactly two fields, named `down` and `up`, whose values
+        // are the two ids and nothing derived from them.
+        let fields: Vec<&str> = p
+            .body
+            .split('&')
+            .map(|kv| kv.split('=').next().unwrap())
+            .collect();
+        assert_eq!(
+            fields,
+            vec!["down", "up"],
+            "the page sent fields beyond the gesture"
+        );
+        for marker in ["SelectAction", "{", "}", "\"", "kind", "target", "problem"] {
+            assert!(
+                !p.body.contains(marker),
+                "the body carries {marker:?} — the page is constructing a command: {}",
+                p.body
+            );
+        }
+    }
+
+    /// And the ids it reports resolve, in Rust, to the command the
+    /// aggregate offered — closing the loop the grep test could not.
+    #[test]
+    fn what_the_page_sends_resolves_to_a_legal_command() {
+        let legal = vec![
+            GroundCommand::SelectAction {
+                action: Action::Ground,
+                target: None,
+                problem: None,
+            },
+            GroundCommand::SelectAction {
+                action: Action::Attack,
+                target: Some(PlayerId(1)),
+                problem: None,
+            },
+        ];
+        let posts = gesture(&page(), "action-attack", "seat-1").expect("runs");
+        let fact = crate::PointerFact::parse(&posts[0].body).expect("a pointer fact");
+        assert_eq!(crate::resolve(&fact, &legal, PlayerId(0)), Ok(1));
+    }
+
+    /// A gesture that means nothing is still *reported* faithfully — the
+    /// page does not filter, because filtering would be deciding.
+    #[test]
+    fn the_page_reports_a_meaningless_gesture_rather_than_suppressing_it() {
+        let posts = gesture(&page(), "seat-1", "seat-2").expect("runs");
+        assert_eq!(posts.len(), 1);
+        assert_eq!(posts[0].body, "down=seat-1&up=seat-2");
+    }
+
+    /// EXPECT-VACUOUS: a harness that ran the script but delivered no
+    /// events would report "sent nothing" and read as a pass. Prove the
+    /// gesture is what causes the send.
+    #[test]
+    fn without_a_gesture_the_page_sends_nothing() {
+        let html = page();
+        let ctx = quick_js::Context::new().expect("ctx");
+        let seen = std::sync::Arc::new(std::sync::Mutex::new(0usize));
+        let sink = seen.clone();
+        ctx.add_callback("__post", move |_u: String, _b: String| {
+            *sink.lock().expect("seen") += 1;
+            0i32
+        })
+        .expect("cb");
+        ctx.add_callback("__reloaded", || 0i32).expect("cb");
+        ctx.eval(DOM).expect("dom");
+        for s in scripts(&html) {
+            ctx.eval(&s).expect("script");
+        }
+        assert_eq!(
+            *seen.lock().expect("seen"),
+            0,
+            "the page sent something unprompted"
+        );
+    }
+
+    /// The harness must not report success when the page is broken.
+    #[test]
+    fn a_page_with_no_script_or_no_handlers_is_an_error_not_a_silence() {
+        assert!(gesture("<p>no script here</p>", "a", "b").is_err());
+        // A script that registers nothing must not look like one that
+        // registered handlers and chose to send nothing.
+        let e = gesture("<script>var x = 1;</script>", "a", "b").unwrap_err();
+        assert!(e.contains("no pointer handlers"), "{e}");
+    }
+
+    #[test]
+    fn both_script_blocks_are_extracted_in_order() {
+        let found = scripts(&page());
+        assert_eq!(
+            found.len(),
+            2,
+            "the page carries the endpoint and the script"
+        );
+        assert!(found[0].contains("CB_ENDPOINT"), "{}", found[0]);
+        assert!(found[1].contains("pointerdown"), "{}", found[1]);
+    }
+}
diff --git a/crates/cb-render-html/src/lib.rs b/crates/cb-render-html/src/lib.rs
index d910f7e..c39b300 100644
--- a/crates/cb-render-html/src/lib.rs
+++ b/crates/cb-render-html/src/lib.rs
@@ -43,6 +43,8 @@
 
 pub mod doc;
 pub mod input;
+#[cfg(any(test, feature = "js-harness"))]
+pub mod jsrun;
 pub mod serve;
 
 pub use doc::{document, text_of};
diff --git a/decisions/ADR-0009-embed-the-js-engine.md b/decisions/ADR-0009-embed-the-js-engine.md
new file mode 100644
index 0000000..f41f32b
--- /dev/null
+++ b/decisions/ADR-0009-embed-the-js-engine.md
@@ -0,0 +1,121 @@
+# ADR-0009: embed the JS engine; do not put `node` in the toolchain
+
+status: accepted
+date: 2026-08-02
+decided by: agent, under the standing loop authorization
+tier: M (structural M — adds an external dependency to the toolchain,
+InnerLoop v1.6; chaos d4=2 → no override). Tier M merges survey and
+decision into one document, which this is.
+references: [CB-WP-0014](../workplans/CB-WP-0014-execute-the-javascript.md),
+[ADR-0007](ADR-0007-render-html-not-a-port.md) D3 (the acquisition rule)
+and D5 (control 5),
+[ADR-0008](ADR-0008-instrument-corrections.md) D2 (AM-4b left uncorrected),
+[CB-EV-0010](../evidence/CB-EV-0010-render-port.md) §4
+
+## Context
+
+ADR-0007 control 5 says the emitted page **may not construct commands** —
+it reports raw pointer facts and Rust decides what they mean. That
+contract is currently held up by a test that greps the emitted script for
+game vocabulary. **Grepping for the absence of words is a weak proxy for
+"this code cannot construct a command."**
+
+And the script has never been executed at all, which is the sole remaining
+reason INTENT stage 1 is open.
+
+## The survey
+
+`node` v24.11.1 is on this machine, so the obvious move is to shell out to
+it. Measured alternatives, same method as `dep-weight.py`, marginal against
+the 29-crate dev-toolchain graph, under the positive control:
+
+| option | marginal Rust lines | notes |
+|---|---:|---|
+| `boa_engine` | 896,410 | 27× AM-4b's headroom |
+| `rquickjs` | 69,985 | 2.1× headroom |
+| **`quick-js`** | **11,434** | bindings + vendored QuickJS C |
+| `node`, shelled out | **0** | and that zero is the problem |
+
+AM-4b headroom is **32,979** (317,021 of 350,000), so `quick-js` fits at
+35% of it and `rquickjs` does not.
+
+## Decision — embed `quick-js`; `node` is refused
+
+**This is ADR-0007 Decision 3's acquisition rule biting its author, which
+is the only real test of whether it was written honestly.** The rule:
+
+> AM-4 counts third-party code the project causes to be **acquired**. It
+> does not count runtimes the user already has independently of us. It
+> **does** count a library our build or install instructions cause to be
+> fetched, pinned, or linked, whether or not its source is Rust.
+
+A browser is not counted because a developer has one regardless of us. But
+**CI runs on `rust:1.97`, which has no `node`** — so adding this test
+would make our CI fetch a JavaScript runtime. That is our build causing an
+acquisition, of tens of millions of lines nobody here will audit, scoring
+**zero** on the only instrument that governs dependencies.
+
+Taking `node` would mean using the rule to exempt a browser we do not
+install while also exempting a runtime we do. `quick-js` costs 11,434
+lines that are vendored, pinned, auditable, and counted.
+
+The secondary reasons matter less but all point the same way: the test
+runs anywhere `cargo test` runs, needs no CI change, cannot skip because a
+binary is missing, and pins one engine version rather than whatever the
+image happens to ship.
+
+**`python3` is not a precedent for `node`.** It is already a toolchain
+dependency, and the honest reading is that it was never argued — it
+predates the acquisition rule. Leaning on it would be using an unexamined
+decision to license a second one. It is left alone here and noted as
+owed.
+
+### What is bought, and what is not
+
+Executing the script proves the **input contract**: what the page puts on
+the wire in response to a pointer gesture. It does **not** prove the SVG
+renders legibly, that a drag feels like a drag, or that anyone can play a
+game. QuickJS has no layout engine and this ADR claims no rendering
+evidence.
+
+### Cost accepted — and a correction, measured after the fact
+
+The paragraph originally here read *"35% of AM-4b's remaining headroom,
+for a test."* **That was wrong, and finding out how wrong is the more
+important result of this pass.**
+
+After landing, `make dep-weight` reported AM-4b **unchanged at 317,021**.
+`quick-js` is a dev-dependency of `cb-render-html`, and AM-4b measures
+`cargo tree -p games-ground --edges normal` — one package, no dev edges.
+It cannot see it. Measured:
+
+| | crates | lines |
+|---|---:|---:|
+| AM-4b as instrumented (`games-ground`, normal) | 29 | 317,021 |
+| the whole workspace, including dev edges | 57 | 725,258 |
+| **uncounted by AM-4b** | **28** | **408,237** |
+
+**The dev-toolchain budget is blind to more source than its entire
+target** — `criterion`, `clap`, `ciborium`, and now `quick-js`.
+
+This is the same defect this ADR refuses `node` for: a real acquisition
+scoring zero because the instrument does not look there. The difference is
+that `quick-js` is *auditable and pinned* and `node` is neither, so the
+decision stands. But it stands on the acquisition rule, **not** on an
+affordability argument, because there is no affordability argument to be
+had until AM-4b can see what it is buying.
+
+Recorded as owed, and it is now the third defect in the AM-4 family: the
+shipped-runtime proc-macro count (fixed, ADR-0008 D2), AM-4b's own
+proc-macro share (owed), and AM-4b's scope (this).
+
+## Consequences
+
+- `quick-js` is a **dev-dependency of `cb-render-html` only**. It must
+  never reach the shipped-runtime configuration; AM-4a would catch that,
+  and now measures 157,202 against 161,000 with no room for it.
+- The grep test stays. It is cheap, and it fails faster and more legibly
+  than an execution test when someone adds a word to the script.
+- If `quick-js` becomes unmaintained, the fallback is not `node` — it is
+  `rquickjs` plus an AM-4b decision, or dropping the execution test and
+  saying so.
diff --git a/evidence/CB-EV-0012-execute-the-javascript.md b/evidence/CB-EV-0012-execute-the-javascript.md
new file mode 100644
index 0000000..c180a64
--- /dev/null
+++ b/evidence/CB-EV-0012-execute-the-javascript.md
@@ -0,0 +1,150 @@
+# CB-EV-0012 — the loop is closed, and the budget is blind
+
+CB-WP-0014 T03. Measured 2026-08-02 at `55212d7`+. Pass kind `product`,
+tier **M** (chaos d4=2, no override). Declaration 9 of 12.
+
+**Applying CB-EV-0011 §4's proposed rule, and hereby adopting it:** the
+cost table below quotes the **previous** pass's final figure and marks
+this pass's own as provisional. See §5.
+
+---
+
+## 1. What was actually blocking stage 1, and what it cost to find out
+
+CB-EV-0010 §4 said the emitted JavaScript had never run. That sentence was
+carried for a pass on an assumption nobody checked. **`node` v24.11.1 was
+on the machine the whole time.**
+
+That is the fourth consecutive pass whose central finding was a claim
+carried because nobody ran the command that settles it — after the 3,750
+headroom figure, the "empty" sub-100k dependency region, and SH-3's 0.0%.
+The pattern is stable enough now to name: **this project's characteristic
+error is not faulty reasoning, it is unexecuted verification.** Every one
+was cheap to check and none had been checked.
+
+## 2. The loop, closed
+
+The real server serves the real page → QuickJS runs *that page's own two
+`<script>` blocks* → the gesture goes over a real socket → the seat's
+`Choice` comes back.
+
+Before this, every link was tested and the chain was not. The page was
+asserted against as a parsed document; the socket was driven by synthetic
+HTTP this test suite wrote itself. **A page whose JavaScript sent
+something else entirely would have passed everything.**
+
+Control 5 — *the page may not construct commands* — was held up by a test
+that grepped the emitted script for game vocabulary. It is now asserted on
+what the script actually puts on the wire, and the mutation that makes the
+JS post `command=SelectAction&target=…` turns it red.
+
+**A wrong assertion worth keeping.** The first draft required the body not
+to contain `"attack"`. It legitimately does — the body is
+`down=action-attack&up=seat-1`, and `action-attack` is *the id of an
+element a finger landed on*. An element may name an action; that is not
+the page deciding anything. What distinguishes reporting from deciding is
+the **shape**: exactly two fields, `down` and `up`, carrying two ids and
+nothing derived from them.
+
+## 3. AM-4b is blind to more source than its own target
+
+ADR-0009's first draft priced `quick-js` at *"35% of AM-4b's remaining
+headroom."* After it landed, `make dep-weight` reported AM-4b
+**unchanged**.
+
+| | crates | lines |
+|---|---:|---:|
+| AM-4b as instrumented (`games-ground`, `--edges normal`) | 29 | 317,021 |
+| the whole workspace, including dev edges | 57 | 725,258 |
+| **uncounted** | **28** | **408,237** |
+
+AM-4b measures one package and no dev edges. `criterion`, `clap`,
+`ciborium` and now `quick-js` are invisible to it.
+
+**This is the same defect the ADR refuses `node` for** — a real
+acquisition scoring zero because the instrument does not look there. The
+decision still stands, because `quick-js` is pinned and auditable and
+`node` is neither, but it stands on the **acquisition rule alone**. The
+affordability argument is withdrawn: there is none to be had until AM-4b
+can see what it is buying.
+
+Third defect in the AM-4 family, and the second found by a pass that was
+*using* the instrument rather than auditing it:
+
+| defect | status |
+|---|---|
+| shipped-runtime counted proc-macro crates (89,048 lines) | fixed, ADR-0008 D2 |
+| AM-4b's own proc-macro share | unmeasured, owed |
+| **AM-4b's scope: one package, no dev edges (408,237 lines)** | **found here, owed** |
+
+## 4. INTENT stage 1, deliverable by deliverable
+
+| deliverable | status |
+|---|---|
+| card/token/hand/**relationship-graph** visualization | **emitted and gated** — 41 of 42 leaf paths asserted present in the *parsed* document. **Never seen.** |
+| drag-to-propose | **evidenced end to end** — JS gesture → socket → `Choice` |
+| debug inspector | **evidenced** (CB-WP-0011) — walks bundles and scenarios, hash-verified |
+| hot-seat play | **evidenced** — two seats, one listener, the projection follows the seat |
+
+The last one was the closest to being claimed on the strength of the code
+path existing. `SeatPolicy` hands every human seat a handle on one shared
+`Server`, so turn-taking "obviously" worked — and nothing drove more than
+one seat until this pass. The property that matters is not that two turns
+happen but that **the same tab, asked twice, shows two different hands**;
+mutating the projection to serve P1's view to every seat turns it red.
+
+### The stage stays open, on one named blocker
+
+Three of four deliverables are evidenced by executing code. The first is
+evidenced only as *correctly emitted*: QuickJS has no layout engine, no
+browser is available here, and no test in this repo can establish that an
+SVG table is legible or that a drag feels like a drag.
+
+**Everything testable from here has been tested.** What remains is one
+human action: run `cb-play --serve 0`, open the printed URL, and confirm
+the table reads and a drag works. That is the whole of the remaining
+blocker, it is the maintainer's to do, and it is named here so the stage
+is no longer carried on a vague reservation.
+
+Closing it unilaterally would be the failure mode stage 0 avoided.
+Carrying it *vaguely* was the failure mode this pass found.
+
+## 5. Cost, shape, and the rule now adopted
+
+| pass | kind | responses | cost | $/response |
+|---|---|---|---|---|
+| CB-WP-0012 | product | 97 | $14.52 | 0.150 |
+| **CB-WP-0013** (final) | meta | 47 | **$8.26** | **0.176** |
+| CB-WP-0014 | product | *provisional — see below* | | |
+
+**The rule is adopted.** CB-EV-0011 proposed that an evidence file quote
+the previous pass's final cost and never its own. CB-WP-0013 reported
+itself at $5.78/34 responses mid-flight; final is **$8.26/47** — under by
+43%. Four for four, always low. This file quotes CB-WP-0013 and declines
+to quote itself, and future evidence files should do the same.
+
+**Meta budget: 29% `[OVER]`** the soft 25%, driven by CB-WP-0013's $8.26
+sitting in a trailing three with two cheap product passes. Reported under
+the rule that requires reporting it; no product work was displaced, and
+CB-WP-0013 was an instrument repair, which ADR-0006 D2 exempts anyway.
+
+**SH-1 is at 347,720 `[HARD]`** against a 300,000 ceiling. The remedy is a
+compaction and it is free, and it is not something this session can do for
+itself. **CB-EV-0009's standing prediction is now live and testable**: if
+the next pass opens above the SH-1 hard line, it should cost more per
+response than 0.123. It has been unfalsified for three passes because
+every one of them opened below the line.
+
+## 6. Open
+
+- **AM-4b's scope defect (408,237 uncounted lines)** and its unmeasured
+  proc-macro share. §3.
+- **`python3` as a toolchain dependency was never argued.** ADR-0009
+  declined to lean on it as precedent and left it owed.
+- **INTENT stage 1: one human verification.** §4.
+- **AM-4a still cannot survive stage 2** — 1,741,979 against 161,000,
+  10.8×. Reserved for the maintainer since ADR-0007.
+- **ADR-0007 D3's acquisition rule** remains unratified — and it has now
+  decided two dependency questions, which raises rather than lowers the
+  cost of leaving it unratified.
+- **Chaos: 9 of 12 declarations, 1 override.**
diff --git a/tools/cb-play/Cargo.toml b/tools/cb-play/Cargo.toml
index 3bafd06..63afc67 100644
--- a/tools/cb-play/Cargo.toml
+++ b/tools/cb-play/Cargo.toml
@@ -14,5 +14,10 @@ serde_json.workspace = true
 # Scenario args are YAML values; the prompt renders them (T02).
 serde_yaml.workspace = true
 
+[dev-dependencies]
+# ADR-0009: the real page, driven by a real JS engine, into the real
+# socket. Dev-only — cb-play ships without it.
+cb-render-html = { workspace = true, features = ["js-harness"] }
+
 [lints]
 workspace = true
diff --git a/tools/cb-play/src/hotseat.rs b/tools/cb-play/src/hotseat.rs
index f53fdd4..d75d2af 100644
--- a/tools/cb-play/src/hotseat.rs
+++ b/tools/cb-play/src/hotseat.rs
@@ -370,4 +370,126 @@ mod tests {
         );
         assert!(replies[1].contains("200 OK"));
     }
+
+    /// **The loop, closed.** The real server serves the real page; a real
+    /// JavaScript engine runs the page's own script and produces a pointer
+    /// gesture; what it produces goes over a real socket; and the seat's
+    /// choice comes back.
+    ///
+    /// Before ADR-0009 every link in that chain was tested and the chain
+    /// was not. The page was asserted against as a parsed document and the
+    /// socket was driven by synthetic HTTP that this test suite wrote
+    /// itself — so a page whose JavaScript sent something else entirely
+    /// would have passed everything.
+    #[test]
+    fn a_gesture_in_javascript_becomes_a_move_in_the_game() {
+        let server = Server::bind(0).expect("bind");
+        let port = server.listener.local_addr().unwrap().port();
+        let token = server.url().rsplit("t=").next().unwrap().to_string();
+
+        let tok = token.clone();
+        let client = std::thread::spawn(move || {
+            let token = tok;
+            // 1. fetch the page the server actually serves
+            let mut s = TcpStream::connect(("127.0.0.1", port)).expect("connect");
+            s.write_all(get(&token).as_bytes()).expect("write");
+            let mut page = String::new();
+            let _ = s.read_to_string(&mut page);
+            assert!(page.contains("200 OK"), "{page}");
+
+            // 2. run ITS script, in a real engine, with a real gesture
+            let posts = cb_render_html::jsrun::gesture(&page, "action-ground", "table")
+                .expect("the served page's script runs");
+            assert_eq!(posts.len(), 1, "{posts:?}");
+
+            // 3. send exactly what the JavaScript produced — not what this
+            //    test thinks it should have produced
+            let mut s = TcpStream::connect(("127.0.0.1", port)).expect("connect");
+            s.write_all(post(&token, &posts[0].body).as_bytes())
+                .expect("write");
+            let mut reply = String::new();
+            let _ = s.read_to_string(&mut reply);
+            (posts[0].clone(), reply)
+        });
+
+        let choice = server
+            .next_choice(&state(), PlayerId(0), &ground_only(), false)
+            .expect("a choice");
+        assert_eq!(choice, Choice::Command(0));
+
+        let (posted, reply) = client.join().expect("client thread");
+        assert_eq!(posted.body, "down=action-ground&up=table");
+        // The endpoint the JS used carries the server's own token, which
+        // the page cannot mint — so this also proves the token round-trips
+        // through the emitted document.
+        assert!(posted.url.contains(&token), "{}", posted.url);
+        assert!(reply.contains("200 OK"), "{reply}");
+        assert!(server.refusals().is_empty(), "{:?}", server.refusals());
+    }
+
+    /// **Hot-seat**: two seats, one browser, one listener — and each seat
+    /// is shown its own hand and nobody else's.
+    ///
+    /// This is the deliverable that was closest to being claimed on the
+    /// strength of the code path existing. `SeatPolicy` gives every human
+    /// seat a handle on one shared `Server`, so turn-taking "obviously"
+    /// works; nothing drove more than one seat until this test.
+    ///
+    /// The property that matters is not that two turns happen. It is that
+    /// the *projection follows the seat* — the same tab, asked twice,
+    /// must show two different hands.
+    #[test]
+    fn two_seats_take_turns_through_one_listener_and_see_different_hands() {
+        let server = Server::bind(0).expect("bind");
+        let port = server.listener.local_addr().unwrap().port();
+        let token = server.url().rsplit("t=").next().unwrap().to_string();
+
+        let tok = token.clone();
+        let client = std::thread::spawn(move || {
+            let mut pages = Vec::new();
+            for _ in 0..2 {
+                let mut s = TcpStream::connect(("127.0.0.1", port)).expect("connect");
+                s.write_all(get(&tok).as_bytes()).expect("write");
+                let mut page = String::new();
+                let _ = s.read_to_string(&mut page);
+                pages.push(page);
+
+                let mut s = TcpStream::connect(("127.0.0.1", port)).expect("connect");
+                s.write_all(post(&tok, "down=action-ground&up=table").as_bytes())
+                    .expect("write");
+                let mut reply = String::new();
+                let _ = s.read_to_string(&mut reply);
+            }
+            pages
+        });
+
+        let state = state();
+        for seat in [0u8, 1] {
+            let choice = server
+                .next_choice(&state, PlayerId(seat), &ground_only(), false)
+                .expect("a choice");
+            assert_eq!(choice, Choice::Command(0), "seat P{}", seat + 1);
+        }
+
+        let pages = client.join().expect("client thread");
+        assert_eq!(pages.len(), 2);
+        for (i, page) in pages.iter().enumerate() {
+            let text = cb_render_html::text_of(page);
+            assert!(
+                text.contains(&format!("viewing as P{}", i + 1)),
+                "turn {i} was not served to P{}",
+                i + 1
+            );
+            // K13: exactly one open hand per page, and it is this seat's.
+            assert_eq!(
+                text.matches("cards)").count(),
+                1,
+                "turn {i} showed {} open hands",
+                text.matches("cards)").count()
+            );
+        }
+        // And the two turns were genuinely different views, not the same
+        // page served twice — which is how this test would pass vacuously.
+        assert_ne!(pages[0], pages[1], "both turns served an identical page");
+    }
 }
diff --git a/tools/status.py b/tools/status.py
index 5dd7c05..5b5b411 100644
--- a/tools/status.py
+++ b/tools/status.py
@@ -214,14 +214,21 @@ def report():
     print(f"  branch     {branch}"
           + (f"  ({len(dirty)} uncommitted)" if dirty else "  (clean)"))
 
+    # The list grows by one line per pass forever, and this report has a
+    # length limit it is meant to keep. Collapse the fully-closed ones to
+    # a single line rather than raising the limit — the precedent is
+    # LOOP-LINT's four loadability breaches, each fixed structurally.
     print("\n  workplans")
+    closed = []
     for wid, title, status, tasks, _kind in plans:
         done = sum(1 for _, s, _ in tasks if s == "done")
         if status == "done" and done == len(tasks):
-            print(f"    {wid}  {status:<12} {done}/{len(tasks)}")
+            closed.append(wid.replace("CB-WP-", ""))
             continue
         short = (title or "")[:44]
         print(f"    {wid}  {status:<12} {done}/{len(tasks)} done   {short}")
+    if closed:
+        print(f"    {len(closed)} closed and complete: " + " ".join(closed))
 
     nxt = next_task(plans)
     if nxt:
diff --git a/workplans/CB-WP-0014-execute-the-javascript.md b/workplans/CB-WP-0014-execute-the-javascript.md
new file mode 100644
index 0000000..71a694f
--- /dev/null
+++ b/workplans/CB-WP-0014-execute-the-javascript.md
@@ -0,0 +1,203 @@
+---
+id: CB-WP-0014
+kind: product
+title: "Execute the JavaScript, and close or keep stage 1"
+status: done
+---
+
+# Purpose
+
+```
+structural tier  M   (adds an external dependency to the toolchain —
+                      InnerLoop v1.6 tier table)
+chaos            d4 = 2  → no override
+declared tier    M
+```
+
+Declaration 9 of 12. Tier M merges the survey and the ADR into one
+document; the adversarial review is optional.
+
+## The one thing blocking stage 1
+
+CB-EV-0010 §4, stated plainly at the time:
+
+> **The emitted JavaScript has never run.** Every test is on the Rust
+> side: the socket loop is driven by synthetic HTTP, and the page is
+> asserted against as a parsed document. No browser engine has executed
+> `SCRIPT`.
+
+All four of INTENT stage 1's named deliverables now exist. That sentence
+is the only reason the stage is still open, and it has been carried
+unexamined for one pass on the assumption that executing it was not
+possible here.
+
+**It is possible.** `node` v24.11.1 is on this machine. The assumption was
+never checked, which makes it the same shape as every other finding in the
+last three passes: a claim carried because nobody ran the one command that
+would settle it.
+
+## What executing it actually proves
+
+Not that the table *looks* right — that needs a rendering engine and a
+human eye, and neither is available here. What it proves is the contract
+ADR-0007 control 5 rests on:
+
+> the page reports **raw pointer facts and nothing else**, and Rust
+> decides what command they mean.
+
+That contract is currently asserted by a test that greps the emitted
+script for game vocabulary. Grepping for the absence of words is a weak
+proxy for "this code cannot construct a command". Running it, feeding it
+synthetic pointer events, and observing exactly what it puts on the wire
+is the real check.
+
+## Task: decide whether a JS engine joins the toolchain
+
+```task
+id: CB-WP-0014-T01
+status: done
+priority: high
+```
+
+Write `decisions/ADR-0009-*.md` (tier M: survey and decision in one).
+
+**This is a real dependency decision, not a formality.** CI runs on the
+`rust:1.97` image, which has no `node`. Adding a test that needs one means
+either installing it in CI or letting the test skip — and **a test that
+skips silently is the harness-does-nothing class this project has found
+five times.** If it can skip, it must fail loudly instead.
+
+Decide, and price it under ADR-0007 Decision 3's acquisition rule — the
+rule this project adopted precisely so that "it's only a dev tool" is not
+an automatic pass:
+
+- `python3` is already a toolchain dependency, so this is the second, not
+  the first. Say whether that makes it cheaper or is merely a precedent
+  being leaned on.
+- What the alternatives cost: no execution at all (status quo, and the
+  reason stage 1 is open), or a JS interpreter as a Rust crate (measure it
+  — do not estimate, and note that AM-4b now has an unmeasured proc-macro
+  share of its own).
+- Whether the engine is required by `make all` or by a separate target.
+
+**AM-4b is the budget this lands in**, and it is the instrument
+CB-WP-0013 explicitly declined to correct. Do not use its uncorrected
+headroom as an argument.
+
+**Done 2026-08-02.**
+[ADR-0009](../decisions/ADR-0009-embed-the-js-engine.md) — **embed
+`quick-js`; `node` is refused.** Measured, marginal, under the control:
+`boa_engine` 896,410 · `rquickjs` 69,985 · **`quick-js` 11,434** · `node`
+**0**, and that zero is the problem.
+
+This is ADR-0007 D3's acquisition rule biting its author. CI runs on
+`rust:1.97`, which has no `node` — so the test would make *our build*
+fetch a JS runtime of tens of millions of unaudited lines, scoring zero on
+the only instrument that governs dependencies. A browser is exempt because
+a developer has one regardless of us; a CI-installed runtime is not.
+
+**And the cost argument in the first draft was wrong.** It claimed 35% of
+AM-4b's headroom. After landing, AM-4b did not move at all — it measures
+`games-ground --edges normal`, one package, no dev edges. Measured: the
+workspace including dev edges is **725,258** lines against AM-4b's
+**317,021**, so **408,237 lines are uncounted — more than the target
+itself**. The decision stands on the acquisition rule; the affordability
+argument is withdrawn, because there is none to be had until the
+instrument can see what it is buying. Filed as the third AM-4 defect.
+
+## Task: run it, end to end, through a real socket
+
+```task
+id: CB-WP-0014-T02
+status: done
+priority: high
+```
+
+Execute `cb_render_html::doc::SCRIPT` in a real JS engine against a
+minimal DOM, synthesize a pointer-down and pointer-up on two real element
+ids from a real emitted document, and let its `fetch` reach the **actual
+`Server`** — the same `next_choice` loop `cb-play --serve` runs.
+
+Acceptance: the seat's `Choice` comes back from a pointer gesture that
+originated in JavaScript. That is the first time anything in this project
+has closed that loop.
+
+**Controls, and they are the point.** The test must go red when:
+
+- the JS is mutated to POST a command name instead of element ids — this
+  is **control 5 asserted rather than grepped**, and if it does not fire,
+  the grep test was the only thing holding the contract up;
+- the pointer events are not delivered at all — the EXPECT-VACUOUS case, a
+  harness that runs a script and asserts nothing about what it did;
+- the token is stripped from the endpoint the page was given.
+
+**Done 2026-08-02.** `crates/cb-render-html/src/jsrun.rs` and
+`hotseat::tests::a_gesture_in_javascript_becomes_a_move_in_the_game`.
+
+**The loop is closed.** The real server serves the real page; QuickJS runs
+*that page's own scripts*; the gesture it produces goes over a real socket;
+the seat's `Choice` comes back. Both `<script>` blocks are lifted from the
+served document, so the endpoint under test is the one the page actually
+carried — token included.
+
+Three controls, each red for its stated reason:
+
+| mutation | result |
+|---|---|
+| the JS posts `command=SelectAction&target=…` | body assertion red — **control 5 asserted, not grepped** |
+| the gesture is not delivered (EXPECT-VACUOUS) | `expected exactly one fetch, got []` |
+| the token dropped from the endpoint | the end-to-end token assertion red |
+
+**A wrong assertion, worth keeping.** The first draft asserted the body
+must not contain `"attack"`. It legitimately does: the body is
+`down=action-attack&up=seat-1`, and `action-attack` is *the id of an
+element a finger landed on*. An element may name an action; that is not
+the page deciding anything. The real test of reporting-versus-deciding is
+the **shape** — exactly two fields, `down` and `up`, carrying two ids and
+nothing derived from them. That is what it asserts now.
+
+## Task: close stage 1, or say what still blocks it
+
+```task
+id: CB-WP-0014-T03
+status: done
+priority: high
+```
+
+`evidence/CB-EV-0012-*.md`, and then a decision on INTENT.
+
+**Be precise about what is and is not established.** A passing JS test
+proves the input contract and the loop. It does **not** prove that an SVG
+table renders legibly, that a drag feels like a drag, or that a human can
+play a game — none of which any test here can reach.
+
+State which of stage 1's four deliverables are evidenced and which rest on
+inspection, then close the stage or name the remaining blocker. Carrying
+it open by default is as unexamined as closing it by default was going to
+be.
+
+**Adopt or reject the self-quoting rule** proposed in CB-EV-0011 §4: an
+evidence file quotes the previous pass's final cost and never its own.
+This pass is the first that can apply it — CB-WP-0013's figure is now
+final. Decide it rather than inheriting it.
+
+**Done 2026-08-02.**
+[CB-EV-0012](../evidence/CB-EV-0012-execute-the-javascript.md).
+
+- **Stage 1 stays open on one named blocker**, not a vague reservation.
+  Three of four deliverables are evidenced by executing code; the
+  visualization is evidenced only as *correctly emitted*, because QuickJS
+  has no layout engine and no browser is available here. What remains is
+  one human action: `cb-play --serve 0`, open the URL, confirm the table
+  reads and a drag works.
+- **Hot-seat was the deliverable closest to being claimed on the strength
+  of the code path existing.** Nothing drove more than one seat until this
+  pass. Now: two seats, one listener, and the projection follows the seat —
+  mutating it to serve P1's view to everyone turns it red.
+- **The self-quoting rule is ADOPTED.** CB-WP-0013 reported itself at
+  $5.78/34 mid-flight; final $8.26/47, under by 43%. Four for four.
+- **AM-4b is blind to 408,237 lines** — more than its own target. Found by
+  a pass that was using it, not auditing it.
+- **SH-1 at 347,720 `[HARD]`.** CB-EV-0009's standing prediction is now
+  live: the next pass opened above that line should cost more than 0.123
+  per response.