diff --git a/docs/re/mission-phase-membership.md b/docs/re/mission-phase-membership.md
index bf49e47..9961003 100644
--- a/docs/re/mission-phase-membership.md
+++ b/docs/re/mission-phase-membership.md
@@ -237,6 +237,37 @@ frozen guest: the banner fires, and the unpacked value is `(True, 0)`.
 yaw/pitch/target across thousands of samples means a dead world, not a patient
 pilot.
 
+## 🟡 2026-08-25 — Turret-biased run, VERIFIED LIVE: still no phase advance
+
+Re-ran it with the freeze witness armed, and checked liveness **before**
+investing in the run this time (`frozen.py` → `animating, max_delta=254`).
+
+This one was real: **823 distinct speeds** logged, the player's hull moved
+1500 → 1485, the escort fell 100 % → 30.7 %, and the ADAN population dropped
+**127 → 89 — 38 enemies destroyed** over ~1064 s. The probe printed no
+`GUEST FROZEN` banner.
+
+**The phase mirror stayed 0 throughout.** So: 38 enemy losses, escort still
+alive, no advance.
+
+### 🔴 But this still does not test the prediction
+
+The prediction is that phase 1 clears when **ADN110/ADN111/ADN112** — 27
+specific Turrets — are destroyed. What was measured is that *38 enemies of some
+kind* died. `SYLPH_PREFER=Turret` biases target choice, but the pilot switched to
+`DEFEND` and spent most of the run on `e010_ADAN_Attacker_S`, protecting the
+escort.
+
+**The missing instrument is per-squadron liveness.** Nothing currently maps a
+live craft back to its roster squadron — that is the same gap
+[roster-to-craft-link](roster-to-craft-link.md) records as unsolved. Until a
+craft can be attributed to `ADN110`, a kill count cannot say whether the
+condition was met, and "no advance after 38 kills" is compatible with both
+"the condition is wrong" and "the right 27 were never killed".
+
+**Honest state: three live-run attempts, no phase advance observed, and the
+decisive experiment is still blocked on attribution rather than on flying time.**
+
 ### 🟡 The prediction to test
 
 If phase 1 clears on those three squadrons dying, then destroying **27 Turrets**