diff --git a/.github/workflows/governance-reusable.yml b/.github/workflows/governance-reusable.yml
index f467a11f..5a8a1929 100644
--- a/.github/workflows/governance-reusable.yml
+++ b/.github/workflows/governance-reusable.yml
@@ -1131,27 +1131,22 @@ jobs:
           done
           [ $failed -eq 1 ] && { echo "Add SPDX header + permissions:"; exit 1; }
           echo "All workflows have SPDX headers + permissions"
-      - name: Check SHA-pinned actions
+      - name: Check locked or SHA-pinned actions
         env:
           GH_TOKEN: ${{ github.token }}
         run: |
           if [ -f .github/workflows/actions.lock ]; then
-            # The lockfile records transitive dependency evidence, while direct
-            # workflow references remain visibly SHA-pinned. Keep both layers:
-            # external analysers and GitHub's sha_pinning_required setting do
-            # not infer direct pins from actions.lock.
+            # actions.lock is the authoritative immutable resolution for both
+            # direct actions and their transitive dependencies. Do not also
+            # rewrite direct refs to raw SHAs: gh actions-lock omits refs that
+            # no tag or branch contains, and GitHub then rejects the workflow
+            # at startup. Measured in oikosbot PR #78 on 2026-08-29: five
+            # previously executable workflows became startup_failure after the
+            # redundant direct-SHA conversion; restoring their locked version
+            # refs made GitHub's native resolver accept them again.
             gh extension install github/gh-actions-lock
             bash "$RUNNER_TEMP/update-actions-lock.sh" --verify-local
-            unpinned=$(grep -rnE --include='*.yml' --include='*.yaml' \
-              "^[[:space:]]+uses:" .github/workflows/ | \
-              grep -v "@[a-f0-9]\{40\}" | \
-              grep -v "uses: \./\|uses: docker://\|uses: hyperpolymath/standards/" || true)
-            if [ -n "$unpinned" ]; then
-              echo "ERROR: direct workflow references not SHA-pinned:"
-              echo "$unpinned"
-              exit 1
-            fi
-            echo "Lockfile coverage verified; direct references SHA-pinned"
+            echo "Immutable direct and transitive lockfile coverage verified"
           else
             unpinned=$(grep -rnE --include='*.yml' --include='*.yaml' \
               "^[[:space:]]+uses:" .github/workflows/ | \
diff --git a/scripts/check-action-pins-resolve.sh b/scripts/check-action-pins-resolve.sh
index 9aec9509..f5eaaa51 100755
--- a/scripts/check-action-pins-resolve.sh
+++ b/scripts/check-action-pins-resolve.sh
@@ -5,9 +5,10 @@ set -uo pipefail
 # check-action-pins-resolve.sh — verify every SHA-pinned action actually EXISTS.
 #
 # ── Why this gate exists ────────────────────────────────────────────────────
-# The governance linter's "Check SHA-pinned actions" step verifies the *shape*
-# of a pin (`@` + 40 hex chars). It cannot tell a real commit from an invented
-# one, because a fabricated SHA is a perfectly well-formed 40-hex string.
+# The governance linter's "Check locked or SHA-pinned actions" step delegates
+# lockfile integrity to gh actions-lock. For repositories without a lockfile it
+# verifies the *shape* of each direct pin (`@` + 40 hex chars), but cannot tell
+# a real commit from an invented one: a fabricated SHA is still well formed.
 #
 # That gap is not theoretical. Measured across the estate on 2026-07-28:
 #