diff --git a/pyproject.toml b/pyproject.toml
index 70681c4ed6c..20cf8b9c115 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -287,6 +287,11 @@ paths_to_mutate = [
 ]
 tests_dir = [
     "tests/test_litellm/proxy/management_endpoints/",
+    # PR1 (key Tier-1) behavior-pinning suite — see
+    # tests/proxy_behavior/management/regression_replay/README.md for the
+    # G4 evidence and tests/proxy_behavior/management/mutmut_triage/pr1.md
+    # for the G5 survivor-classification protocol.
+    "tests/proxy_behavior/management/",
 ]
 also_copy = [
     "litellm/",
diff --git a/tests/proxy_behavior/management/mutmut_triage/pr1.md b/tests/proxy_behavior/management/mutmut_triage/pr1.md
new file mode 100644
index 00000000000..fe31eea0498
--- /dev/null
+++ b/tests/proxy_behavior/management/mutmut_triage/pr1.md
@@ -0,0 +1,66 @@
+# G5 — Mutmut survivor triage for PR1 (Key Tier-1)
+
+PR1 wires the new behavior suite into `[tool.mutmut].tests_dir` in
+`pyproject.toml`. This file records the **first manually-triggered mutmut
+run** that exercises `litellm/proxy/management_endpoints/key_management_endpoints.py`
+against the new suite, and classifies every surviving mutant inside the 6
+Tier-1 handler functions:
+
+  * `generate_key_fn` (and helpers `_common_key_generation_helper`,
+    `key_generation_check`, `_team_key_generation_check`,
+    `_team_key_operation_team_member_check`, `_get_user_in_team`)
+  * `update_key_fn` (and `_check_key_admin_access` helper)
+  * `regenerate_key_fn` (and `_get_and_validate_existing_key` helper)
+  * `info_key_fn`
+  * `list_keys`
+  * `delete_key_fn`
+
+## How to run
+
+```bash
+# Manual trigger of the workflow:
+gh workflow run mutation-test.yml --ref litellm_/silly-wright-1b8559
+gh run watch
+```
+
+Or locally (~hours, plan around it):
+
+```bash
+uv run --with mutmut==3.5.0 mutmut run
+```
+
+Once the run completes, download `mutation-report.md` from the workflow
+artifact and:
+
+  1. Filter the survivors to the 6 handler functions listed above.
+  2. For each surviving mutant in those functions, add a row to the table
+     below classifying it **killed** (write a new test → mutant dies →
+     rerun confirms) or **accepted** with a one-line reason (equivalent
+     mutation, defensive-only branch, unreachable under any realistic
+     world-seed).
+
+G5 is the binding gate for the mutation signal: **zero unreviewed
+survivors** in the Tier-1 handler set. The aggregate kill rate is recorded
+as telemetry under "Baseline metrics" below — never as a goalpost.
+
+## Baseline metrics (filled after the first mutmut run lands)
+
+| Metric | Value | Notes |
+|--------|-------|-------|
+| Surviving mutants (all paths_to_mutate) | _TBD_ | |
+| Surviving mutants in Tier-1 handler funcs | _TBD_ | G5 gate target: 0 unreviewed |
+| Killed mutants | _TBD_ | |
+| Kill rate (killed / (killed + survived)) | _TBD_ | Telemetry only |
+| Wall-clock minutes | _TBD_ | |
+
+## Survivor triage
+
+| # | Mutant location (file:line, op) | Classification | Reason / new test |
+|---|---------------------------------|----------------|-------------------|
+| _none yet — first mutmut run pending_ | | | |
+
+## PR2 / PR3 delta
+
+When PR2 (Team Tier-1) ships, append a new "Baseline metrics" block here
+keyed on the PR — never overwrite. The headline number to report is the
+**delta** vs. PR1's baseline, not the absolute kill rate.