diff --git a/prod/docs/uv_cert/UV_EXIT_DOCTRINE_PARITY_PASS_20260712.md b/prod/docs/uv_cert/UV_EXIT_DOCTRINE_PARITY_PASS_20260712.md
index 582004f..6f73213 100644
--- a/prod/docs/uv_cert/UV_EXIT_DOCTRINE_PARITY_PASS_20260712.md
+++ b/prod/docs/uv_cert/UV_EXIT_DOCTRINE_PARITY_PASS_20260712.md
@@ -134,3 +134,16 @@ is therefore PART OF PARITY.** Transcribe it regardless of verdict; analyze
 its net contribution separately (candidate: flight recorder + trade_events
 per-leg attribution over historical windows). Parity is fidelity to what
 runs, not to what should run.
+
+**⚠ OPERATOR CAUTION (×3): the OB TP coefficient may be DORMANT qua code
+path.** Wiring exists (`esf_alpha_orchestrator.py:842` attaches ob_engine to
+exit_manager) but MUST NOT be transcribed as active on faith. Before porting
+TP modulation, VERIFY LIVE: (1) the attach call executes in the live trader
+with a REAL OBFeatureEngine (not None/MockOBProvider); (2) empirically:
+BLUE logs `tp_mod_factor` on EVERY decision (aem diag) — sweep the live
+decision record/logs for tp_mod_factor ≠ 1.0 occurrences. If none ever fire,
+the coefficient is dormant-in-effect and UV must transcribe it in the SAME
+dormant state (present, wired identically, firing identically — i.e. never)
+or parity breaks in UV's favor of a mechanism BLUE doesn't actually express.
+Same verification applies to each modulation branch separately (cascade vs
+withdrawal vs per-asset — one may fire while others are dead).