diff --git a/CHANGELOG.md b/CHANGELOG.md
index a68994a..91ce609 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -133,6 +133,9 @@ repository still gets a decision, never by following the link; no release carrie
 - :memo: docs(changelog): regenerate CHANGELOG.md for the EX-S10 lane
 - :memo: docs(changelog): regenerate CHANGELOG.md for the EX-S10 F2/F5 fixes
 - :memo: docs(changelog): regenerate CHANGELOG.md for the EX-S10 F6 fix
+- :memo: docs(changelog): regenerate CHANGELOG.md for the AUD-S18 CHECK_STAGES fix
+- :memo: docs(changelog): regenerate CHANGELOG.md for the check_coverage_bar narrowing fix
+- :memo: docs(changelog): regenerate CHANGELOG.md for the stale-'15' cosmetic fix
 
 ### Features
 - :sparkles: feat(docs): gate example pack and format claims against dogfood
@@ -152,9 +155,12 @@ repository still gets a decision, never by following the link; no release carrie
 - :bug: fix(docs): pin case+decision pairs, not just case names, in the walkthrough (F1)
 - :bug: fix(examples): scope format extraction to match: lines, not the whole classes: block (F2)
 - :bug: fix(examples): close the same-line trailing-comment gap in format extraction (F6)
+- :bug: fix(audit): sync AUD-S18 CHECK_STAGES pin with Taskfile check task
+- :bug: fix(audit): narrow check_coverage_bar to avoid dogfood-examples' per-pack coverage lines
 
 ### Refactoring
 - :recycle: refactor(examples): discover dogfood packs instead of a hardcoded 3-name loop
+- :art: style(audit): de-stale hardcoded '15 stages' literals in AUD-S18 gate
 
 ### Testing
 - :white_check_mark: test(release): anchor the D-120 note check on its header sentence, not the bare token
diff --git a/hack/audit/exitgate_test.sh b/hack/audit/exitgate_test.sh
index 19a5836..a97519b 100644
--- a/hack/audit/exitgate_test.sh
+++ b/hack/audit/exitgate_test.sh
@@ -6,7 +6,7 @@
 #
 #   (1) REL-07  — the AUD-S01 fail-closed cassettes PASS by name (not "go test
 #                 exited 0", which a `-run` regex matching nothing also does).
-#   (2) RELSE-01/TEST-03 — `task check` is green with all 15 pinned stages
+#   (2) RELSE-01/TEST-03 — `task check` is green with all pinned stages
 #                 actually executed (incl. `changelog-verify`) and the measured
 #                 aggregate coverage at or above the epic's 91.0% bar.
 #   (3) RELSE-05 — the release job still runs the verify-green-on-tag-SHA gate,
@@ -122,9 +122,9 @@ S01_TESTS_CMD=(
   TestFoldSnapshotPathsIncompleteEnumeration
 )
 
-# (2) The 15 stages of `task check`. This list is the authority: the Taskfile's
+# (2) The stages of `task check`. This list is the authority: the Taskfile's
 # `check:` list must equal it, and a real `task check` transcript must show all
-# 15 having run. Grading on the exit code alone is what "aborts at the first
+# of them having run. Grading on the exit code alone is what "aborts at the first
 # failure" makes meaningless.
 CHECK_STAGES=(
   fmt
@@ -133,6 +133,7 @@ CHECK_STAGES=(
   test
   coverage
   build
+  dogfood-examples
   dogfood-comparison
   compare-exitgate-test
   changelog-verify
@@ -141,6 +142,7 @@ CHECK_STAGES=(
   docs-gates
   lint-depguard-test
   lint-workflow-pins-test
+  dogfood-wiring-test
   ci-audit-test
 )
 
@@ -455,7 +457,7 @@ check_s01_cassettes() { # <transcript>
 # (2) RELSE-01 / TEST-03 — `task check` green at the new bar
 # ============================================================================
 
-# The Taskfile's `check:` list must be exactly the 15 pinned stages. Without
+# The Taskfile's `check:` list must be exactly the pinned stages. Without
 # this, deleting `- task: changelog-verify` reopens RELSE-01 with `task check`
 # still exiting 0 — the regression the whole story exists to prevent.
 check_check_wiring() { # <taskfile>
@@ -512,7 +514,7 @@ check_check_wiring() { # <taskfile>
 # Every pinned stage must have actually RUN. go-task prints `task: [<name>] <cmd>`
 # per stage; that transcript is the only evidence a stage executed, because
 # `task check` aborts at the first failure and the exit code cannot distinguish
-# "15 stages green" from "stage list truncated to 2".
+# "all stages green" from "stage list truncated to 2".
 check_check_stages() { # <transcript>
   local tr="$1" rc=0
   [[ -f "$tr" ]] || {
@@ -557,12 +559,18 @@ check_coverage_bar() { # <transcript>
   local plain="$WORK/cov.plain"
   strip_ansi "$tr" >"$plain"
 
-  # Anchored at column 1: `go test -coverprofile` prints a per-PACKAGE
-  # `…  coverage: 92.3% of statements` on every ok line, and grading on the
-  # minimum of those would compare the worst package against the aggregate bar.
-  # Only the gate's own `echo "coverage: ${pct}% (required: …)"` starts the line.
+  # Anchored at column 1 AND on a digit right after the colon-space: `go test
+  # -coverprofile` prints a per-PACKAGE `…  coverage: 92.3% of statements` on
+  # every ok line (not at column 1, already excluded by the anchor), and
+  # EX-S08's dogfood-examples stage echoes ONE "coverage: OK — N rule(s),
+  # every rule tested in both polarities" line per example pack (at column 1,
+  # excluded only by the digit check — "OK" is not `[0-9]`). Grading on either
+  # impostor would either compare the worst package against the aggregate bar
+  # or count N-pack lines as N spurious "aggregate" measurements. Only the
+  # gate's own `echo "coverage: ${pct}% (required: …)"` (Taskfile.yml
+  # coverage:) starts the line with a digit.
   local hits="$WORK/hits.coverage"
-  awk 'index($0, "coverage: ") == 1 { print }' "$plain" >"$hits"
+  awk '/^coverage: [0-9]/ { print }' "$plain" >"$hits"
   local n
   n="$(wc -l <"$hits" | tr -d '[:space:]')"
   if ((n != 1)); then
@@ -1551,7 +1559,7 @@ expect_red check_s01_cassettes "the -run pattern matched NO test (go test exits
 echo
 
 # ---------------------------------------------------------------------------
-echo "== (2) RELSE-01 / TEST-03 — task check green, 15 stages, coverage >= ${COVERAGE_BAR}% =="
+echo "== (2) RELSE-01 / TEST-03 — task check green, ${#CHECK_STAGES[@]} stages, coverage >= ${COVERAGE_BAR}% =="
 
 expect_green check_check_wiring "Taskfile check: runs exactly the ${#CHECK_STAGES[@]} pinned stages" "$TASKFILE"
 expect_green check_coverage_floor "the single-sourced D-010 floor (Taskfile COVERAGE_MIN) is >= ${COVERAGE_BAR}" "$TASKFILE"
@@ -1700,6 +1708,25 @@ mutate_awk "$m" \
   'coverage: 62.5% of statements'
 expect_green check_coverage_bar "a per-package coverage line does not displace the aggregate one" "$m"
 
+# EX-S08's dogfood-examples stage echoes its own "coverage: OK — N rule(s),
+# every rule tested in both polarities" line per example pack — ALSO anchored
+# at column 1, unlike the per-package control above. Without the digit anchor
+# these three impostors plus the real aggregate line make `n == 4`, the exact
+# regression this control pins (the bug this fix closes: reachable in
+# practice as soon as dogfood-examples ran more than zero packs, and reached
+# main unnoticed because check_check_wiring failed first and aborted the
+# script before this check ever ran).
+m="$WORK/check.synth.dogfoodcov"
+cp "$SYNTH_CHECK" "$m"
+mutate_awk "$m" \
+  '{ print } /^task: \[dogfood-examples\]/ {
+     print "coverage: OK — 6 rule(s), every rule tested in both polarities"
+     print "coverage: OK — 8 rule(s), every rule tested in both polarities"
+     print "coverage: OK — 9 rule(s), every rule tested in both polarities"
+   }' \
+  'coverage: OK — 6 rule(s), every rule tested in both polarities'
+expect_green check_coverage_bar "dogfood-examples' per-pack 'coverage: OK — N rule(s)' lines do not displace the aggregate one" "$m"
+
 echo
 
 # ---------------------------------------------------------------------------