diff --git a/tools/check.sh b/tools/check.sh
index 3e553f8d..20de302c 100755
--- a/tools/check.sh
+++ b/tools/check.sh
@@ -395,7 +395,10 @@ run_rust_gate() {
 
       if [ "$rc" -eq 0 ] && [ "${MISALIGNED_BEVY_SMOKE:-0}" = "1" ]; then
         step "headless Bevy evidence"
-        bash tools/bevy-headless.sh dark --no-build || rc=1
+        # cargo test and clippy do not produce target/debug/misaligned-bevy.
+        # Let the helper build that real binary so smoke works in a freshly
+        # seeded worktree as well as one with prior local runs.
+        bash tools/bevy-headless.sh dark || rc=1
       elif [ "$rc" -eq 0 ]; then
         echo "headless Bevy evidence: SKIP (set MISALIGNED_BEVY_SMOKE=1)"
       fi
@@ -421,7 +424,7 @@ run_rust_gate() {
 
       if [ "$rc" -eq 0 ] && [ "${MISALIGNED_BEVY_SMOKE:-0}" = "1" ]; then
         step "headless Bevy evidence"
-        bash tools/bevy-headless.sh dark --no-build || rc=1
+        bash tools/bevy-headless.sh dark || rc=1
       fi
       ;;
   esac
diff --git a/wiki/log/2026-07-10-fresh-bevy-smoke.md b/wiki/log/2026-07-10-fresh-bevy-smoke.md
new file mode 100644
index 00000000..6d565a00
--- /dev/null
+++ b/wiki/log/2026-07-10-fresh-bevy-smoke.md
@@ -0,0 +1,37 @@
+# 2026-07-10 — Make Bevy smoke self-sufficient
+
+```
+Type: log
+```
+
+## Intent
+
+Repair the optional frontend smoke gate so it passes in a freshly seeded
+worktree instead of depending on a manually built Bevy binary.
+
+## Changed
+
+- Removed `--no-build` from the frontend and full smoke-gate invocations.
+- Kept `--no-build` available to callers that have deliberately built the
+  binary already; the check gate no longer relies on that hidden precondition.
+- Documented why the smoke harness owns the binary build.
+
+## Design/spec impact
+
+This fixes the existing headless-evidence contract in
+`wiki/process/agent-scale.md`: a land proof must run without a watched window
+or prior manual setup. Cargo tests and clippy build test artifacts, not the
+launchable `misaligned-bevy` binary the harness executes.
+
+## Verification
+
+- Removed the private worktree's `target/debug/misaligned-bevy` first, then
+  ran `MISALIGNED_BEVY_SMOKE=1 ./tools/check.sh --frontend`.
+- The gate rebuilt the missing binary, captured `dark`, and passed its fog
+  audit and PNG validation (SHA-256
+  `a3a935a5cfed53b55fd8b374d0001dd356ee6d93d6aa8e73f32d90ff9cac0ec1`).
+- Repeated the missing-binary proof through
+  `MISALIGNED_BEVY_SMOKE=1 ./tools/check.sh --full`; all workspace tests,
+  clippy, agent smoke, docs gates, and Bevy evidence passed.
+- `./tools/check.sh --land` (docs auto-classification after the explicit
+  frontend smoke evidence)
diff --git a/wiki/log/DEVLOG.md b/wiki/log/DEVLOG.md
index ce2c2e1e..ec9c16a5 100644
--- a/wiki/log/DEVLOG.md
+++ b/wiki/log/DEVLOG.md
@@ -181,6 +181,11 @@ add or amend a session log, then re-run the generator.
 - Intent: Cameron, from an Ears-beat screenshot: the material frame showed concentric noise-field rings and schematic floor language. He wants only **dark shapes for things that are detectable** — not radar rings.
 - Log: [wiki/log/2026-07-10-heard-mass-only.md](2026-07-10-heard-mass-only.md)
 
+## 2026-07-10 - Make Bevy smoke self-sufficient
+
+- Intent: Repair the optional frontend smoke gate so it passes in a freshly seeded worktree instead of depending on a manually built Bevy binary.
+- Log: [wiki/log/2026-07-10-fresh-bevy-smoke.md](2026-07-10-fresh-bevy-smoke.md)
+
 ## 2026-07-10 - The thought fluid goes READY
 
 - Intent: Cameron affirmed all four PROPOSED taste calls on the thought-fluid render spec in one pass ("seems fine to me, proceed").
diff --git a/wiki/process/workflows.md b/wiki/process/workflows.md
index 1193d8f7..47424e11 100644
--- a/wiki/process/workflows.md
+++ b/wiki/process/workflows.md
@@ -306,7 +306,9 @@ MISALIGNED_BEVY_SMOKE=1 ./tools/check.sh --frontend
 
 The helper requires process exit 0, `fog audit OK`, and a valid PNG. The env
 flag keeps this optional in ordinary gates while making it one switch for a
-frontend land.
+frontend land. When enabled through `check.sh`, the harness builds the real
+Bevy binary itself: frontend tests and clippy do not guarantee that a fresh
+worktree already has `target/debug/misaligned-bevy`.
 
 Agent mode is the standard observed-run gate for terminal playability. It is
 command-clocked, deterministic under `--seed`, and produces plain-text frames