diff --git a/.github/scripts/cloud_run_env.sh b/.github/scripts/cloud_run_env.sh
index 6d2757fcc..f5388559a 100755
--- a/.github/scripts/cloud_run_env.sh
+++ b/.github/scripts/cloud_run_env.sh
@@ -17,7 +17,7 @@ cloud_run_set_defaults() {
   CLOUD_RUN_MAX_INSTANCES="${CLOUD_RUN_MAX_INSTANCES:-1}"
   # Stage 2-qualified runtime shape, pinned explicitly on every deploy —
   # rationale in docs/migration/cloud-run-operations.md ("Runtime shape").
-  CLOUD_RUN_CONCURRENCY="${CLOUD_RUN_CONCURRENCY:-4}"
+  CLOUD_RUN_CONCURRENCY="${CLOUD_RUN_CONCURRENCY:-6}"
   CLOUD_RUN_WEB_CONCURRENCY="${CLOUD_RUN_WEB_CONCURRENCY:-2}"
   CLOUD_RUN_PORT="${CLOUD_RUN_PORT:-8080}"
   CLOUD_RUN_POLICYENGINE_DB_PASSWORD_SECRET="${CLOUD_RUN_POLICYENGINE_DB_PASSWORD_SECRET:-policyengine-api-prod-db-password:latest}"
diff --git a/changelog.d/cloud-run-concurrency-6.changed.md b/changelog.d/cloud-run-concurrency-6.changed.md
new file mode 100644
index 000000000..3948399d1
--- /dev/null
+++ b/changelog.d/cloud-run-concurrency-6.changed.md
@@ -0,0 +1 @@
+Raise Cloud Run per-instance request concurrency from 4 to 6 to reduce bot-burst scale-out churn.
diff --git a/docs/migration/cloud-run-operations.md b/docs/migration/cloud-run-operations.md
index aed16fa13..7f53c5a53 100644
--- a/docs/migration/cloud-run-operations.md
+++ b/docs/migration/cloud-run-operations.md
@@ -61,9 +61,13 @@ Values measured and justified in
 
 - **`WEB_CONCURRENCY=2`** — the engine is GIL-bound; two worker processes materially
   outperform one on the 4-vCPU instance. Set via the deploy script's env vars.
-- **`--concurrency 4`** — admission control is what keeps the two-worker instance
+- **`--concurrency 6`** — admission control is what keeps the two-worker instance
   memory-safe in 16Gi; excess load queues at the platform and fails as client timeouts
-  rather than dying in-instance. **Both knobs are pinned explicitly on every deploy**:
+  rather than dying in-instance. Raised from the original 4 on 2026-07-21 (re-qualified
+  by a dedicated 30-min soak — memory peaked 61%, zero OOM/5xx; see the 2026-07-21
+  amendment in `history/pr4-stage2-runtime-timing.md`) to stop bot-burst scale-out
+  churn: the autoscaler targets ~60% of concurrency, so c4 spawned instances at ~2.4
+  sustained concurrent requests. **Both knobs are pinned explicitly on every deploy**:
   gcloud inherits unspecified template fields (a mid-campaign CI deploy once shipped an
   inherited test concurrency), and `--concurrency default` resolves to the *platform*
   default (640), not the historical 80.
diff --git a/docs/migration/history/pr4-stage2-runtime-timing.md b/docs/migration/history/pr4-stage2-runtime-timing.md
index 24f8311cd..77cc07f68 100644
--- a/docs/migration/history/pr4-stage2-runtime-timing.md
+++ b/docs/migration/history/pr4-stage2-runtime-timing.md
@@ -131,3 +131,37 @@ memory-safe.
    request mix (likely O(5,000+) with jitter ~50%).
 4. Re-baseline after metadata slimming lands; it dominates warm-parity and
    transfer costs.
+
+## Amendment 2026-07-21 — concurrency re-qualified at 6 (c6)
+
+Context: overnight bot-traffic waves (2026-07-21 00:00–06:00Z) churned Cloud
+Run scale-outs 1→4→1→4 (18 boots); at c4 the autoscaler's ~60% target spawns
+an instance at ~2.4 sustained concurrent requests, and every boot queues
+early-bind-routed requests behind the ~3-min import. Decision (user,
+2026-07-21): raise `--concurrency` to 6, moving the trigger to ~3.6.
+
+Qualification soak per this document's discipline (never change concurrency
+without its own soak — c8 OOM'd a worker at 92% in the original campaign):
+
+- No-traffic revision `policyengine-api-c6-qual` (concurrency 6, 4 vCPU/16Gi,
+  WEB_CONCURRENCY=2), tag URL only, public traffic untouched.
+- 30-min soak, 12 closed-loop clients, cache-busted mix
+  (`scripts/measure_cloud_run_runtime.py --clients 12 --duration-seconds 1800
+  --cache-bust always`), window 14:31:35–15:01:35Z; CPU p99 sustained 35–79%
+  through 15:05Z (fully saturated 6-slot instance with queue pressure).
+- **Gates, evaluated server-side: PASS.** Memory p99-per-2min peaked at 61%
+  (14:55Z) and held 60% steady to soak end — under the 70% bar (vs 92% at c8,
+  69% knife-edge at c4's qualifying soak); zero OOM / worker restarts (only
+  the two initial boots at instance start); zero 5xx — 320/320 requests 200
+  in the revision's request log.
+- Caveat recorded honestly: the load-generator process was killed at soak end
+  before writing its client-side JSON (operator session exit), so client-side
+  latency percentiles were lost. All three gates are server-side observable
+  and were evaluated from Monitoring + request logs over the full window; the
+  loss is informational only.
+- Cleanup: tag removed; revision undeletable while latest-created (same
+  precedent as `s2-restore2`) — zero traffic, scaled to zero, superseded by
+  the next CI deploy. Revision YAML captured at qualification time.
+
+Shipped as `CLOUD_RUN_CONCURRENCY` default 4→6 in
+`.github/scripts/cloud_run_env.sh` with the dry-run invariant updated.
diff --git a/tests/unit/test_cloud_run_deploy_scripts.py b/tests/unit/test_cloud_run_deploy_scripts.py
index eb215817e..32895da97 100644
--- a/tests/unit/test_cloud_run_deploy_scripts.py
+++ b/tests/unit/test_cloud_run_deploy_scripts.py
@@ -371,7 +371,7 @@ def test_deploy_cloud_run_candidate_pins_runtime_shape():
     assert result.returncode == 0, result.stderr
     # Stage 2-qualified values, pinned on every deploy — rationale in
     # docs/migration/cloud-run-operations.md ("Runtime shape and scaling").
-    assert "--concurrency 4 " in result.stdout
+    assert "--concurrency 6 " in result.stdout
     assert "WEB_CONCURRENCY=2 " in result.stdout
     assert "--min-instances 0 " in result.stdout