diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index bc5584277..dbca8e5c6 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -1,6 +1,17 @@
 name: ci
 
 on:
+  # The only way to run the rate-limit-guard bench lanes in CI. They are gated
+  # on BENCH_LANES inside their suite because spawning a lane is running a
+  # benchmark, and a scheduled lane of their own would be unread advisory
+  # automation outside `ci-status`. A dispatch runs the full contract corpus,
+  # so this input only decides whether the lanes run inside it.
+  workflow_dispatch:
+    inputs:
+      bench_lanes:
+        description: "Run the rate-limit-guard statusline bench lanes"
+        type: boolean
+        default: false
   push:
     branches: [main]
   pull_request:
@@ -1492,6 +1503,9 @@ jobs:
         if: needs.changes.outputs.run_tests == 'true'
         env:
           BASE_REF: ${{ github.base_ref }}
+          # Empty on every event but a dispatch that asked for the lanes, which
+          # is what the gate in the bench suite reads as "stay deferred".
+          BENCH_LANES: ${{ inputs.bench_lanes && '1' || '' }}
         run: |
           if [ "$GITHUB_EVENT_NAME" != pull_request ]; then
             scripts/run-plugin-tests.sh --jobs 3 --shard "$LEG/$LEGS"
diff --git a/plugins/rate-limit-guard/.claude-plugin/plugin.json b/plugins/rate-limit-guard/.claude-plugin/plugin.json
index 96a256eb4..2bf692e1c 100644
--- a/plugins/rate-limit-guard/.claude-plugin/plugin.json
+++ b/plugins/rate-limit-guard/.claude-plugin/plugin.json
@@ -1,7 +1,7 @@
 {
   "$schema": "https://json.schemastore.org/claude-code-plugin-manifest.json",
   "name": "rate-limit-guard",
-  "version": "0.8.18",
+  "version": "0.8.19",
   "description": "Shared rate-limit guard for loop lanes: a statusline wrapper tees the subscription rate-limit windows to a fixed machine-scope file, a StopFailure hook records rate-limit stops reactively, and a reader contract fixes how consuming sessions pause and resume.",
   "author": {
     "name": "Melodic Software",
diff --git a/plugins/rate-limit-guard/CHANGELOG.md b/plugins/rate-limit-guard/CHANGELOG.md
index 30be65ba9..263ca4015 100644
--- a/plugins/rate-limit-guard/CHANGELOG.md
+++ b/plugins/rate-limit-guard/CHANGELOG.md
@@ -3,6 +3,12 @@
 All notable changes to the `rate-limit-guard` plugin are documented here. Format follows
 [Keep a Changelog](https://keepachangelog.com/en/1.1.0/); this plugin uses semantic versioning.
 
+## [0.8.19]
+
+### Changed
+
+- `bench.test.sh` gates its five lane cases on `BENCH_LANES`. Without it the suite runs only the lib-helper assertions and prints a `SKIP:` line naming the coverage that did not run; `BENCH_LANES=1` runs every case, locally or through `ci.yml`'s `bench_lanes` dispatch input.
+
 ## [0.8.18]
 
 ### Changed
diff --git a/plugins/rate-limit-guard/bench/README.md b/plugins/rate-limit-guard/bench/README.md
index e92e1aea0..ed6c7d071 100644
--- a/plugins/rate-limit-guard/bench/README.md
+++ b/plugins/rate-limit-guard/bench/README.md
@@ -83,9 +83,19 @@ HOME="$(mktemp -d)" bash plugins/rate-limit-guard/bench/bench-idle.sh
 
 The **benchmarks gate nothing**. Wall-clock numbers on shared CI runners are noise, so no lane's
 timing ever runs in CI. What does run is `bench.test.sh`, a contract smoke suite discovered by
-`scripts/run-plugin-tests.sh` like every other `*.test.sh`: it unit-tests the lib helpers and
-runs each lane once with tiny parameters against the repo tee under an isolated `HOME`, asserting
-behaviour and output shape, never timing. That keeps the harness runnable from a clean checkout
-(an unrunnable harness is exactly the defect that made #2521's measurements unreproducible) and
-maps these files into `scripts/affected-tests.sh` coverage. The tee's behavioural coverage lives
-in `../scripts/statusline-tee.test.sh`.
+`scripts/run-plugin-tests.sh` like every other `*.test.sh`, which maps these files into
+`scripts/affected-tests.sh` coverage. It splits in two:
+
+- **Every run** unit-tests the lib helpers: `median`, `pace_sleep_arg`, `now_ms` and the refusal
+  on a bash without `EPOCHREALTIME`. No lane is spawned, so it costs milliseconds.
+- **`BENCH_LANES=1`** adds the lane cases: one tiny-parameter run of each lane against the repo
+  tee under an isolated `HOME`, plus the two failing-render aborts, asserting behaviour and
+  output shape, never timing. Spawning a lane is running a benchmark whatever the parameters, so
+  those five cases are gated; without the variable the suite prints a `SKIP:` line and the
+  runner's summary names the coverage that did not run.
+
+Run them locally with `BENCH_LANES=1 bash plugins/rate-limit-guard/bench/bench.test.sh`, or in CI
+by dispatching `ci.yml` with its `bench_lanes` input set. That is the deliberate run that keeps
+the harness runnable from a clean checkout; an unrunnable harness is exactly the defect behind the
+unreproducible measurements in #2521. The tee's behavioural coverage lives in
+`../scripts/statusline-tee.test.sh`.
diff --git a/plugins/rate-limit-guard/bench/bench.test.sh b/plugins/rate-limit-guard/bench/bench.test.sh
index f866e57fd..1b4c6745e 100755
--- a/plugins/rate-limit-guard/bench/bench.test.sh
+++ b/plugins/rate-limit-guard/bench/bench.test.sh
@@ -8,6 +8,12 @@
 # unrunnable harness is exactly the defect that let #2521's measurements go
 # unreproducible (#2582).
 #
+# The lane runs are GATED on BENCH_LANES: running a lane is running a benchmark
+# however small its parameters, so an ordinary CI run stops after the lib
+# assertions and reports the lanes as skipped coverage. BENCH_LANES=1 runs them
+# — locally, or in CI through ci.yml's `bench_lanes` dispatch input, which is
+# the deliberate run the #2582 guard rests on.
+#
 # Self-contained: defines its own assertion helpers — installed plugins are
 # cache-isolated with no shared test lib.
 
@@ -26,6 +32,13 @@ ok() {
   echo "ok: $*"
   PASS=$((PASS + 1))
 }
+# One summary shape for both exits: the gated stop after the lib assertions and
+# the full run.
+summary() {
+  echo
+  echo "PASS=$PASS FAIL=$FAIL"
+  [[ $FAIL -eq 0 ]]
+}
 
 WORK="$(mktemp -d)"
 trap 'rm -rf "$WORK"' EXIT
@@ -77,6 +90,25 @@ else
   fail "lib: EPOCHREALTIME guard rc=$RC out=$OUT"
 fi
 
+# --- the gate ----------------------------------------------------------------
+# Everything above asserts pure functions and costs milliseconds. Everything
+# below SPAWNS the lanes, which is a benchmark run whatever the parameters, and
+# is the only part of this suite that spends real wall-clock seconds on a
+# shared runner.
+#
+# The deferral prints a SKIP line rather than counting an ok. scripts/
+# run-plugin-tests.sh reads `^SKIP:` and names the suite under "Suites with
+# skipped coverage (exit 0 here is NOT evidence those cases ran)", which is
+# what this is: five cases that did not run. An ok would make the aggregate
+# read as full coverage. --strict-skips therefore fails here, correctly — a
+# caller declaring a fully provisioned environment is asking for every case to
+# run, and BENCH_LANES=1 is how it gets them.
+if [[ -z "${BENCH_LANES:-}" ]]; then
+  echo "SKIP: bench lanes deferred; set BENCH_LANES=1 to run them"
+  summary
+  exit
+fi
+
 # --- bench-idle: smoke run against the repo tee, isolated HOME ---------------
 HOME1="$WORK/home1"
 mkdir -p "$HOME1"
@@ -132,6 +164,4 @@ else
   fail "trace-probe: rc=$RC out=${OUT:0:400}"
 fi
 
-echo
-echo "PASS=$PASS FAIL=$FAIL"
-[[ $FAIL -eq 0 ]]
+summary
diff --git a/scripts/run-plugin-tests-serial.txt b/scripts/run-plugin-tests-serial.txt
index 8e62de661..b386522cc 100644
--- a/scripts/run-plugin-tests-serial.txt
+++ b/scripts/run-plugin-tests-serial.txt
@@ -39,7 +39,10 @@ plugins/context-guard/hooks/zone-gate.test.sh
 # A/B timing harness; its verdicts are elapsed-time comparisons.
 plugins/performance/scripts/ab.test.sh
 
-# The statusline bench measures render latency.
+# The statusline bench measures render latency. Its lane runs are gated on
+# BENCH_LANES, so an ordinary run stops before them and needs no seat here; the
+# entry is for the runs that DO set it (a local full corpus with the lanes
+# enabled), where the lanes measure exactly what contention distorts.
 plugins/rate-limit-guard/bench/bench.test.sh
 
 # Cancellation-window cases kill a parked rename shim on a short fuse.