diff --git a/.gitea/scripts/tests/test_gate_review_auto_fire.py b/.gitea/scripts/tests/test_gate_review_auto_fire.py
index 72a650cc..d0545d95 100644
--- a/.gitea/scripts/tests/test_gate_review_auto_fire.py
+++ b/.gitea/scripts/tests/test_gate_review_auto_fire.py
@@ -50,14 +50,25 @@ class TestQaReviewDirectTrigger:
             "pull_request_review must include 'submitted' type"
         )
 
-    def test_job_guard_requires_approved_state(self):
+    def test_job_guard_fires_on_pull_request_review(self):
         wf = load_workflow("qa-review.yml")
         guard = _job_guard_string(wf)
-        assert "github.event.review.state == 'APPROVED'" in guard, (
-            "job guard must check review.state for 'APPROVED'"
+        assert "github.event_name == 'pull_request_review'" in guard, (
+            "job guard must fire on pull_request_review events"
         )
-        assert "github.event.review.state == 'approved'" in guard, (
-            "job guard must check review.state for 'approved' (case fallback per #2135)"
+        # #2159: state guard removed because Gitea 1.22.6 does not reliably
+        # expose review.state in the pull_request_review payload. The
+        # evaluator (review-check.sh) validates APPROVED via API anyway.
+        assert "github.event.review.state" not in guard, (
+            "job guard must NOT check review.state (unreliable in Gitea payload)"
+        )
+
+    def test_pull_request_target_includes_labeled(self):
+        wf = load_workflow("qa-review.yml")
+        on = wf[True]
+        types = on["pull_request_target"].get("types", [])
+        assert "labeled" in types, (
+            "qa-review must re-run on labeled events so auto-tier flips the gate (#2396)"
         )
 
     def test_post_step_uses_status_post_token(self):
@@ -91,14 +102,25 @@ class TestSecurityReviewDirectTrigger:
             "pull_request_review must include 'submitted' type"
         )
 
-    def test_job_guard_requires_approved_state(self):
+    def test_job_guard_fires_on_pull_request_review(self):
         wf = load_workflow("security-review.yml")
         guard = _job_guard_string(wf)
-        assert "github.event.review.state == 'APPROVED'" in guard, (
-            "job guard must check review.state for 'APPROVED'"
+        assert "github.event_name == 'pull_request_review'" in guard, (
+            "job guard must fire on pull_request_review events"
         )
-        assert "github.event.review.state == 'approved'" in guard, (
-            "job guard must check review.state for 'approved' (case fallback per #2135)"
+        # #2159: state guard removed because Gitea 1.22.6 does not reliably
+        # expose review.state in the pull_request_review payload. The
+        # evaluator (review-check.sh) validates APPROVED via API anyway.
+        assert "github.event.review.state" not in guard, (
+            "job guard must NOT check review.state (unreliable in Gitea payload)"
+        )
+
+    def test_pull_request_target_includes_labeled(self):
+        wf = load_workflow("security-review.yml")
+        on = wf[True]
+        types = on["pull_request_target"].get("types", [])
+        assert "labeled" in types, (
+            "security-review must re-run on labeled events so auto-tier flips the gate (#2396)"
         )
 
     def test_post_step_uses_status_post_token(self):
diff --git a/.gitea/scripts/tests/test_sop_auto_tier.sh b/.gitea/scripts/tests/test_sop_auto_tier.sh
new file mode 100755
index 00000000..2cff342b
--- /dev/null
+++ b/.gitea/scripts/tests/test_sop_auto_tier.sh
@@ -0,0 +1,149 @@
+#!/usr/bin/env bash
+# Regression test for #2396 — auto-tier assignment from diff heuristics.
+#
+# Validates the heuristic that computes tier:low|medium|high from a list of
+# changed file paths. The heuristic must be conservative: high-risk paths
+# (security, auth, crypto, secrets, migrations) force tier:high; test-only
+# or doc-only changes default to tier:low; everything else is medium.
+
+set -euo pipefail
+
+PASS=0
+FAIL=0
+
+assert_eq() {
+  local label="$1"
+  local expected="$2"
+  local got="$3"
+  if [ "$expected" = "$got" ]; then
+    echo "  PASS  $label"
+    PASS=$((PASS + 1))
+  else
+    echo "  FAIL  $label"
+    echo "        expected: <$expected>"
+    echo "        got:      <$got>"
+    FAIL=$((FAIL + 1))
+  fi
+}
+
+# ----- Heuristic under test (mirrors sop-tier-check.sh auto-tier block) -----
+# Input: newline-separated filenames (no JSON/jq dependency for testability)
+compute_tier() {
+  local files="$1"
+  local has_high_risk=""
+  local has_non_test=""
+  local file_count
+  file_count=$(echo "$files" | grep -c '^' || true)
+  if [ -z "$files" ] || [ "$file_count" -eq 0 ]; then
+    echo "tier:low"
+    return
+  fi
+  local fname
+  while IFS= read -r fname; do
+    [ -z "$fname" ] && continue
+    case "$fname" in
+      *security*|*auth*|*crypto*|*secret*|*migrations*|*/handlers/secrets*|*/handlers/approval_gate*|*/handlers/workspace_crud*)
+        has_high_risk="yes"
+        ;;
+    esac
+    case "$fname" in
+      # .gitea changes affect CI workflows, SOP gates, and merge policy
+      # for ALL PRs. Reviewer recommendation: treat as high-risk.
+      .gitea/*)
+        has_high_risk="yes"
+        ;;
+    esac
+    case "$fname" in
+      *_test.go|*/docs/*|*.md|*.txt|*.json|*.yaml|*.yml|*.css|*.html)
+        ;;
+      *)
+        has_non_test="yes"
+        ;;
+    esac
+  done <<<"$files"
+
+  if [ -n "$has_high_risk" ]; then
+    echo "tier:high"
+  elif [ -z "$has_non_test" ]; then
+    echo "tier:low"
+  else
+    echo "tier:medium"
+  fi
+}
+
+# Empty PR → low
+echo "test: empty file list → tier:low"
+assert_eq "empty" "tier:low" "$(compute_tier "")"
+
+# Test-only changes → low
+echo "test: test-only changes → tier:low"
+assert_eq "test-only" "tier:low" "$(compute_tier $'foo_test.go\nbar_test.go')"
+
+# Doc-only changes → low
+echo "test: doc-only changes → tier:low"
+assert_eq "docs-only" "tier:low" "$(compute_tier $'docs/README.md\nCHANGELOG.txt')"
+
+# Mixed test + non-test → medium
+echo "test: mixed test + src → tier:medium"
+assert_eq "mixed" "tier:medium" "$(compute_tier $'foo_test.go\nmain.go')"
+
+# Pure src changes → medium
+echo "test: pure src changes → tier:medium"
+assert_eq "src-only" "tier:medium" "$(compute_tier $'internal/handlers/workspace.go\ninternal/db/db.go')"
+
+# Security path → high
+echo "test: security path → tier:high"
+assert_eq "security" "tier:high" "$(compute_tier $'internal/handlers/secrets.go')"
+
+# Auth path → high
+echo "test: auth path → tier:high"
+assert_eq "auth" "tier:high" "$(compute_tier $'internal/auth/middleware.go')"
+
+# Crypto path → high
+echo "test: crypto path → tier:high"
+assert_eq "crypto" "tier:high" "$(compute_tier $'internal/crypto/encrypt.go')"
+
+# Migrations path → high
+echo "test: migrations path → tier:high"
+assert_eq "migrations" "tier:high" "$(compute_tier $'migrations/047_add_column.sql')"
+
+# Mixed with high-risk → high (overrides medium)
+echo "test: mixed with high-risk → tier:high"
+assert_eq "mixed-high" "tier:high" "$(compute_tier $'foo_test.go\ninternal/handlers/secrets.go')"
+
+# .gitea/workflows changes → high (gate-affecting, reviewer recommendation)
+echo "test: .gitea/workflows change → tier:high"
+assert_eq "gitea-workflows" "tier:high" "$(compute_tier $'.gitea/workflows/sop-tier-check.yml')"
+
+# .gitea/scripts changes → high (gate-affecting)
+echo "test: .gitea/scripts change → tier:high"
+assert_eq "gitea-scripts" "tier:high" "$(compute_tier $'.gitea/scripts/sop-tier-check.sh')"
+
+# .gitea change mixed with test → high (high-risk wins)
+echo "test: .gitea/workflows + test → tier:high"
+assert_eq "gitea-mixed" "tier:high" "$(compute_tier $'.gitea/workflows/qa-review.yml\nfoo_test.go')"
+
+# .gitea change with security → high (both high-risk)
+echo "test: .gitea/workflows + security → tier:high"
+assert_eq "gitea-security" "tier:high" "$(compute_tier $'.gitea/workflows/sop-tier-check.yml\ninternal/handlers/secrets.go')"
+
+# Fail-closed: TIER_EXPR must not contain placeholder teams (???)
+# If present, the gate would silently skip clauses instead of failing.
+echo "test: TIER_EXPR has no placeholder teams (fail-closed)"
+# Extract the TIER_EXPR block from sop-tier-check.sh and grep for ???
+SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
+TIER_EXPR_BLOCK=$(sed -n '/declare -A TIER_EXPR=/,/^)/p' "$SCRIPT_DIR/../sop-tier-check.sh")
+TIER_EXPR_PLACEHOLDERS=$(echo "$TIER_EXPR_BLOCK" | grep -c '???' || true)
+if [ "$TIER_EXPR_PLACEHOLDERS" -eq 0 ]; then
+  echo "  PASS  no-placeholder-teams"
+  PASS=$((PASS + 1))
+else
+  echo "  FAIL  no-placeholder-teams"
+  echo "        Found $TIER_EXPR_PLACEHOLDERS ??? placeholder(s) in TIER_EXPR"
+  FAIL=$((FAIL + 1))
+fi
+
+echo
+echo "------"
+echo "PASS=$PASS FAIL=$FAIL"
+[ "$FAIL" -eq 0 ]
diff --git a/.gitea/workflows/qa-review.yml b/.gitea/workflows/qa-review.yml
index 51ff78cf..40f372d1 100644
--- a/.gitea/workflows/qa-review.yml
+++ b/.gitea/workflows/qa-review.yml
@@ -96,7 +96,7 @@ name: qa-review
 
 on:
   pull_request_target:
-    types: [opened, synchronize, reopened]
+    types: [opened, synchronize, reopened, labeled, unlabeled]
   pull_request_review:
     types: [submitted]
 
@@ -110,13 +110,21 @@ jobs:
   approved:
     # Gate the job:
     #   - On pull_request_target events: always run.
-    #   - On pull_request_review_approved events: run so the gate flips
-    #     immediately when a team member submits an APPROVE review.
+    #   - On pull_request_review events: always run.  We do NOT guard on
+    #     review.state here because Gitea 1.22.6's payload shape for this
+    #     event does not reliably expose the state field that the GitHub-
+    #     style guard expects (issue #2159).  The evaluator
+    #     (review-check.sh) reads the actual reviews from the API and
+    #     checks for a real APPROVE, so running on COMMENT or
+    #     REQUEST_CHANGES reviews is harmless (read-only, idempotent).
+    #     sop-tier-check.yml uses the same pattern (no state guard) and
+    #     provably fires on every review event.
+    #   - On labeled/unlabeled events: re-evaluate when tier label changes.
+    #     This ensures qa-review flips when auto-tier assigns a label (#2396).
     # Comment-triggered refires live in sop-checklist.yml review-refire job.
     if: |
       github.event_name == 'pull_request_target' ||
-      (github.event_name == 'pull_request_review' &&
-       (github.event.review.state == 'APPROVED' || github.event.review.state == 'approved'))
+      github.event_name == 'pull_request_review'
     runs-on: ubuntu-latest
     steps:
       - name: Privilege check (A1.1 — INFORMATIONAL log only, NOT a gate)
diff --git a/.gitea/workflows/security-review.yml b/.gitea/workflows/security-review.yml
index c214cb1c..dc261802 100644
--- a/.gitea/workflows/security-review.yml
+++ b/.gitea/workflows/security-review.yml
@@ -23,7 +23,7 @@ name: security-review
 
 on:
   pull_request_target:
-    types: [opened, synchronize, reopened]
+    types: [opened, synchronize, reopened, labeled, unlabeled]
   pull_request_review:
     types: [submitted]
 
@@ -37,13 +37,21 @@ jobs:
   approved:
     # Gate the job:
     #   - On pull_request_target events: always run.
-    #   - On pull_request_review_approved events: run so the gate flips
-    #     immediately when a team member submits an APPROVE review.
+    #   - On pull_request_review events: always run.  We do NOT guard on
+    #     review.state here because Gitea 1.22.6's payload shape for this
+    #     event does not reliably expose the state field that the GitHub-
+    #     style guard expects (issue #2159).  The evaluator
+    #     (review-check.sh) reads the actual reviews from the API and
+    #     checks for a real APPROVE, so running on COMMENT or
+    #     REQUEST_CHANGES reviews is harmless (read-only, idempotent).
+    #     sop-tier-check.yml uses the same pattern (no state guard) and
+    #     provably fires on every review event.
+    #   - On labeled/unlabeled events: re-evaluate when tier label changes.
+    #     This ensures security-review flips when auto-tier assigns a label (#2396).
     # Comment-triggered refires live in sop-checklist.yml review-refire job.
     if: |
       github.event_name == 'pull_request_target' ||
-      (github.event_name == 'pull_request_review' &&
-       (github.event.review.state == 'APPROVED' || github.event.review.state == 'approved'))
+      github.event_name == 'pull_request_review'
     runs-on: ubuntu-latest
     steps:
       - name: Privilege check (A1.1 — INFORMATIONAL log only, NOT a gate)