diff --git a/.github/scripts/changelog-armed.sh b/.github/scripts/changelog-armed.sh
new file mode 100755
index 0000000..c2c37df
--- /dev/null
+++ b/.github/scripts/changelog-armed.sh
@@ -0,0 +1,108 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+# changelog-armed.sh [<changelog>] [<version-file>] — assert that
+# CHANGELOG.md is ARMED: that there is a heading for the next PR's entry to
+# land under, and that it is the right one for the state this tree is in.
+#
+# The failure it exists to catch (#108, heavy-duty/rig#66) leaves no trace:
+# the ceremony PR stamps '## Unreleased' into '## X.Y.Z — DATE' by hand, and
+# nothing puts the heading back. A PR authored BEFORE the release wrote its
+# entry under '## Unreleased'; that heading is gone by the time it merges, so
+# git lands the entry under whatever heading now occupies that position — the
+# just-shipped section — CLEANLY, with no conflict. The one signal an author
+# would trust ("git told me to look") is absent exactly when the result is
+# wrong, and the drift is only ever discovered by reading the file.
+#
+# The rule, keyed on VERSION, because the two states are genuinely different:
+#
+#   VERSION ends in -dev  ->  the top section MUST be '## Unreleased'
+#   VERSION is bare       ->  the top section may be '## Unreleased' (armed,
+#                             the ceremony's own re-arm) or the stamped
+#                             section for exactly that VERSION
+#
+# Keying on VERSION is the whole design, and the reason this is not simply
+# "require '## Unreleased'". That unconditional form is what rig#44 and
+# heavy-duty/cast#108 had to REVERT: it is false by construction on the
+# ceremony PR's own tree, which makes the release unshippable through a green
+# CI. Anyone tempted to simplify this back should read those two first.
+#
+# The consequence worth stating plainly: a ceremony PR that stamps and forgets
+# to re-arm still passes here — its VERSION is bare, and a bare tree is
+# allowed to be stamped. It goes red the moment the '-dev' bump lands on main,
+# which release.yml does automatically in the same job as the publish. So the
+# guard does not block the release; it refuses to let main SIT disarmed, which
+# is the window a late PR can fall into.
+#
+# A file of its own (not inlined in ci.yml) so test/release.sh can drive it
+# against constructed trees for both states — the same discipline as
+# release-notes.sh.
+
+changelog="${1:-CHANGELOG.md}"
+version_file="${2:-VERSION}"
+
+[ -f "$changelog" ]    || { echo "changelog-armed: no such file: $changelog" >&2; exit 1; }
+[ -f "$version_file" ] || { echo "changelog-armed: no such file: $version_file" >&2; exit 1; }
+
+ver="$(tr -d '[:space:]' < "$version_file")"
+[ -n "$ver" ] || { echo "changelog-armed: $version_file is empty" >&2; exit 1; }
+
+# The TOP section: the first '## ' heading in the file. Everything above it is
+# the changelog's own preamble and belongs to no section.
+top="$(grep -m1 '^## ' "$changelog" || true)"
+[ -n "$top" ] || {
+  echo "changelog-armed: $changelog has no '## ' section at all — nothing for a PR entry to land under" >&2
+  exit 1
+}
+
+# '## 0.7.0 — 2026-07-19' -> '0.7.0'. Split on whitespace, same shape
+# release-notes.sh matches on, so the two cannot disagree about what a
+# section header is.
+top_ver="$(printf '%s\n' "$top" | awk '{ print $2 }')"
+
+case "$ver" in
+  *-dev)
+    if [ "$top_ver" != "Unreleased" ]; then
+      cat >&2 <<EOF
+changelog-armed: VERSION is '$ver' (a development tree) but the top section of
+  $changelog is:
+
+    $top
+
+  A -dev tree MUST carry '## Unreleased' at the top. Without it, a PR that
+  wrote its entry under '## Unreleased' before the release merges CLEANLY into
+  the section above — the one that already shipped — and the changelog quietly
+  misattributes it (#108, heavy-duty/rig#66).
+
+  The fix is to re-arm: add an empty '## Unreleased' immediately above
+  '$top'. The release ceremony is supposed to do this in the same edit that
+  stamps the version — see CONTRIBUTING.md, "Releases".
+EOF
+      exit 1
+    fi
+    ;;
+  *)
+    # A bare VERSION is the ceremony tree and the merge commit that publishes
+    # it. Both arrangements are legal there: re-armed ('## Unreleased' back on
+    # top, above the section just stamped) or not yet re-armed (the stamped
+    # section still on top). What is NOT legal is a stamped top section naming
+    # some OTHER version — that is a ceremony that stamped the wrong number,
+    # and release.yml would publish a body that is not this release's.
+    if [ "$top_ver" != "Unreleased" ] && [ "$top_ver" != "$ver" ]; then
+      cat >&2 <<EOF
+changelog-armed: VERSION is '$ver' but the top section of $changelog is:
+
+    $top
+
+  A bare VERSION means this tree is a release. Its top section must be either
+  '## Unreleased' (re-armed after stamping) or the stamped section for '$ver'
+  itself. A stamped section naming a different version means the ceremony
+  stamped the wrong number, and the published release body would come from
+  the wrong section.
+EOF
+      exit 1
+    fi
+    ;;
+esac
+
+echo "changelog-armed: VERSION '$ver' agrees with the top section ($top_ver)"
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 1ba0a22..8d5bce0 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -25,6 +25,12 @@ jobs:
         run: bash test/labels-reconcile.sh
       - name: release-flow tests
         run: bash test/release.sh
+      # The changelog is ARMED for the next entry (#108). Its own step rather
+      # than a line inside test/release.sh: this one asserts a fact about THIS
+      # tree, not about the release machinery, so when it goes red the log
+      # says which check found the drift without anyone reading a suite.
+      - name: changelog is armed for the next entry
+        run: bash .github/scripts/changelog-armed.sh
 
   # The multi-user rehearsal, on a REAL incus — a GitHub runner is root on a
   # disposable VM, which is exactly the substrate the rehearsal needs. It runs
diff --git a/CHANGELOG.md b/CHANGELOG.md
index f5ae264..3b01b16 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -34,6 +34,36 @@ which records not just what changed but what each drill run proved.
 
 ### Fixed
 
+- **The release ceremony re-arms `CHANGELOG.md`, and CI refuses to let
+  `main` sit disarmed** (#108) — the ceremony stamps `## Unreleased` into
+  `## X.Y.Z — DATE` by hand, and nothing put the heading back, so `main`
+  sat with no `## Unreleased` from the release until the next PR that
+  happened to re-create one. A PR authored *before* the release wrote its
+  entry under `## Unreleased`; with that heading gone, git lands the entry
+  under whatever now occupies the position — **the section that just
+  shipped** — and it merges **cleanly**. No conflict, no error, no red X:
+  the one signal an author would trust is absent exactly when the outcome
+  is wrong, and the changelog credits a released version with a change it
+  does not contain until a human reads the file. Confirmed in the sibling
+  repo (heavy-duty/rig#66); box has not drifted yet, and the reason is
+  luck rather than design — 0.6.0's ceremony (`77599ab`) added its heading
+  *without* removing `## Unreleased`, so main was never disarmed, while
+  0.7.0 did disarm it and left a window that nothing happened to cross.
+  Two halves land together. The ceremony step in `CONTRIBUTING.md` is now
+  explicitly **two edits**: stamp, then put an empty `## Unreleased` back
+  above the section just stamped — it belongs there and not in
+  `release.yml`, which only ever touches `VERSION`. And
+  `.github/scripts/changelog-armed.sh` enforces it in CI, keyed on
+  `VERSION` because the two states are genuinely different: a `-dev` tree
+  must carry `## Unreleased` on top, a bare-`VERSION` tree (the ceremony
+  PR, and the merge that publishes it) may carry either that or its own
+  stamped section. The keying is the whole design and not an
+  over-complication — box previously had **no** top-section guard at all,
+  and the obvious one, an unconditional `## Unreleased` requirement, is
+  false by construction on the ceremony PR's own tree, which is why rig#44
+  and heavy-duty/cast#108 both had to revert it. So a forgotten re-arm
+  does not block the release; it turns `main` red on the very next push,
+  the automatic `-dev` bump the release itself makes.
 - **`box restore` asks before it destroys — and the confirmation prompt is
   now the row's, not rm's** (#105) — `restore` and `rm` both irreversibly
   discard user state, and only one of them asked. The table gave `restore`
diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
index 394985f..227250f 100644
--- a/CONTRIBUTING.md
+++ b/CONTRIBUTING.md
@@ -50,7 +50,48 @@ A release is a PR, and merging it ships it
 1. **The release PR** — `release: X.Y.Z`, labeled `release` — bumps `VERSION`
    from `X.Y.Z-dev` and stamps the `## Unreleased` section with version +
    date (feature PRs land their changelog entry as part of the PR, so the
-   section is already written). This PR is where the release ritual hangs:
+   section is already written).
+
+   **Stamping is two edits, not one — the second is re-arming.** After
+   rewriting `## Unreleased` into `## X.Y.Z — DATE`, put an **empty
+   `## Unreleased` back at the top**, immediately above the section you just
+   stamped:
+
+   ```markdown
+   ## Unreleased
+
+   ## 0.7.1 — 2026-07-19
+
+   ### Fixed
+   ...
+   ```
+
+   Not cosmetic, and not deferrable to the next PR that happens to need it.
+   Between the stamp and the next re-creation of that heading, `main` has no
+   `## Unreleased`. A PR authored *before* the release wrote its entry under
+   that heading; with the heading gone, git lands the entry under whatever
+   now occupies the position — **the section that just shipped** — and it
+   merges **cleanly**, no conflict, no signal. The changelog then credits a
+   released version with a change it does not contain, and nothing but a
+   human reading the file will ever say so
+   ([#108](https://github.com/heavy-duty/box/issues/108); confirmed in the
+   sibling repo as
+   [heavy-duty/rig#66](https://github.com/heavy-duty/rig/issues/66)).
+
+   CI enforces the arming rule with
+   [.github/scripts/changelog-armed.sh](.github/scripts/changelog-armed.sh),
+   keyed on `VERSION`: a `-dev` tree must carry `## Unreleased` on top; a
+   bare-`VERSION` tree (the ceremony PR, and the merge that publishes it) may
+   carry either `## Unreleased` or its own stamped section. That is why the
+   guard cannot simply demand `## Unreleased` unconditionally — the
+   unconditional form is false on the ceremony PR's own tree and makes the
+   release unshippable, which is why rig and cast both reverted it. The
+   practical consequence: forgetting to re-arm does **not** block the release
+   PR, it turns `main` red on the very next push — the automatic `-dev` bump
+   the release itself makes. Do it in the ceremony PR and main is never
+   disarmed at all.
+
+   This PR is where the release ritual hangs:
    the full drill on real hardware, recorded in
    [drill/RUNS.md](drill/RUNS.md) — CI proves the tier's semantics on every
    PR, a release still proves the boundary.
diff --git a/test/release.sh b/test/release.sh
index 8e8a5e9..d02583d 100644
--- a/test/release.sh
+++ b/test/release.sh
@@ -179,6 +179,81 @@ check "release.yml: the release bumps main to the next -dev itself" 0 "" \
 check "release.yml: ...with a PR fallback when the direct push is refused" 0 "" \
   grep -qF "opening the bump PR instead" "$RY"
 
+# ---------------------------------------------------------------------------
+# changelog-armed.sh (#108) — the changelog has a heading for the NEXT entry.
+#
+# The drift this catches produces no conflict and no error: the ceremony
+# stamps '## Unreleased' away, and a PR authored before the release merges its
+# entry cleanly into the section that just shipped. box has no top-section
+# guard at all today — the two checks above pin only that the 0.6.0 and 0.5.0
+# sections still extract, which a disarmed main passes happily.
+#
+# BOTH states are constructed as real trees and the real script is run against
+# them, because the failure mode of the naive fix is precisely a state
+# mismatch: an unconditional '## Unreleased' requirement is green on main and
+# false on the ceremony PR's own tree, which is why rig#44 and
+# heavy-duty/cast#108 both had to revert one. A test that only drives the
+# -dev state would have shipped that bug again.
+# ---------------------------------------------------------------------------
+ARMED="$ROOT/.github/scripts/changelog-armed.sh"
+check "changelog-armed: runnable bash" 0 "" bash -n "$ARMED"
+
+# tree <dir> <version> <changelog-body...> — a two-file tree to run against
+tree() {
+  local d="$WORK/$1" v="$2"; shift 2
+  mkdir -p "$d"
+  printf '%s\n' "$v" > "$d/VERSION"
+  { echo "# Changelog"; echo; printf '%s\n' "$@"; } > "$d/CHANGELOG.md"
+  echo "$d"
+}
+armed() { bash "$ARMED" "$1/CHANGELOG.md" "$1/VERSION"; }
+
+# --- the -dev steady state: armed is the only legal shape ------------------
+T="$(tree dev-armed 0.7.1-dev '## Unreleased' '' '- **A pending entry**' '' '## 0.7.0 — 2026-07-19' '' '- **Shipped**')"
+check "armed: a -dev tree with '## Unreleased' on top passes" 0 "agrees" armed "$T"
+T="$(tree dev-disarmed 0.7.1-dev '## 0.7.0 — 2026-07-19' '' '- **Shipped**')"
+check "armed: a -dev tree WITHOUT it fails — the #108 drift, caught" 1 "MUST carry" armed "$T"
+check "armed: ...and the failure says how to fix it (re-arm)" 1 "re-arm" armed "$T"
+check "armed: ...naming the issue and its origin" 1 "heavy-duty/rig#66" armed "$T"
+
+# --- the ceremony PR: bare VERSION, BOTH arrangements legal ----------------
+# This is the pair that the reverted guards got wrong. Neither may fail, or
+# the release PR cannot go green and the ceremony is unshippable.
+T="$(tree rel-stamped 0.7.1 '## 0.7.1 — 2026-07-19' '' '- **This release**')"
+check "armed: a bare VERSION with its OWN stamped section on top passes" 0 "agrees" armed "$T"
+T="$(tree rel-rearmed 0.7.1 '## Unreleased' '' '## 0.7.1 — 2026-07-19' '' '- **This release**')"
+check "armed: ...and so does the RE-ARMED ceremony tree (the shape #108 asks for)" \
+  0 "agrees" armed "$T"
+# The one bare-VERSION arrangement that is wrong: a stamp naming another
+# version. release.yml would publish a body that is not this release's.
+T="$(tree rel-wrong 0.7.1 '## 0.7.0 — 2026-07-19' '' '- **Some other release**')"
+check "armed: a bare VERSION under someone ELSE's stamped section fails" 1 "wrong number" armed "$T"
+
+# --- degenerate trees refuse rather than pass by accident ------------------
+T="$(tree no-sections 0.7.1-dev 'Prose and no headings at all.')"
+check "armed: a changelog with no '## ' section at all fails" 1 "no '## ' section at all" armed "$T"
+check "armed: a missing changelog refuses by path" 1 "no such file" \
+  bash "$ARMED" "$WORK/nope.md" "$ROOT/VERSION"
+check "armed: a missing VERSION refuses by path" 1 "no such file" \
+  bash "$ARMED" "$ROOT/CHANGELOG.md" "$WORK/nope-version"
+mkdir -p "$WORK/empty-ver"; : > "$WORK/empty-ver/VERSION"
+check "armed: an empty VERSION refuses" 1 "is empty" \
+  bash "$ARMED" "$ROOT/CHANGELOG.md" "$WORK/empty-ver/VERSION"
+
+# --- and the tree under test, which is the assertion that actually fires ---
+check "armed: THIS tree's VERSION and CHANGELOG.md agree" 0 "agrees" \
+  bash "$ARMED" "$ROOT/CHANGELOG.md" "$ROOT/VERSION"
+
+# The guard is only a guard if CI runs it, and the ceremony is only re-armed
+# if the ceremony step says so. Fail-closed pins on both, since a guard nobody
+# invokes and a step nobody wrote are the two ways this reverts silently.
+check "ci.yml: runs the changelog-armed guard" 0 "" \
+  grep -qF 'changelog-armed.sh' "$ROOT/.github/workflows/ci.yml"
+check "CONTRIBUTING: the ceremony re-arms '## Unreleased' after stamping" 0 "" \
+  grep -qF 'Stamping is two edits, not one' "$ROOT/CONTRIBUTING.md"
+check "CONTRIBUTING: ...and names the guard that enforces it" 0 "" \
+  grep -qF 'changelog-armed.sh' "$ROOT/CONTRIBUTING.md"
+
 # ---------------------------------------------------------------------------
 # latest_release_tag — extracted from install.sh (the source-the-pure-function
 # trick) and driven against a shim curl. The shim serves the ONE seam the