diff --git a/Observability/esof_advisor.py b/Observability/esof_advisor.py
index 716e8d6..8be31c7 100755
--- a/Observability/esof_advisor.py
+++ b/Observability/esof_advisor.py
@@ -291,7 +291,10 @@ def compute_esof(now: datetime = None) -> dict:
 
     if   advisory_score >  0.25: advisory_label = "FAVORABLE"
     elif advisory_score >  0.05: advisory_label = "MILD_POSITIVE"
-    elif advisory_score > -0.05: advisory_label = "NEUTRAL"
+    # UNKNOWN (was NEUTRAL): constituent signals in conflict. Empirically the worst
+    # ROI bucket, not a benign mid-state — naming is load-bearing for consumers
+    # making "stand aside vs size-down" decisions.
+    elif advisory_score > -0.05: advisory_label = "UNKNOWN"
     elif advisory_score > -0.25: advisory_label = "MILD_NEGATIVE"
     else:                         advisory_label = "UNFAVORABLE"
 
@@ -394,7 +397,8 @@ CYAN   = "\033[36m"; BOLD   = "\033[1m";  DIM    = "\033[2m";  RST = "\033[0m"
 LABEL_COLOR = {
     "FAVORABLE":    GREEN,
     "MILD_POSITIVE":"\033[92m",
-    "NEUTRAL":      YELLOW,
+    "UNKNOWN":      YELLOW,   # renamed from NEUTRAL — signals-in-conflict
+    "NEUTRAL":      YELLOW,   # backward-compat for historical CH rows / stale HZ snapshots
     "MILD_NEGATIVE":"\033[91m",
     "UNFAVORABLE":  RED,
 }
diff --git a/Observability/esof_gate.py b/Observability/esof_gate.py
index 6941163..5d182e2 100755
--- a/Observability/esof_gate.py
+++ b/Observability/esof_gate.py
@@ -104,13 +104,16 @@ S6_MULT: Dict[str, Dict[int, float]] = {
     #             B0    B1    B2    B3    B4    B5    B6
     "FAVORABLE":     {0: 0.65, 1: 0.50, 2: 0.0, 3: 2.0, 4: 0.20, 5: 0.75, 6: 1.5},
     "MILD_POSITIVE": {0: 0.50, 1: 0.35, 2: 0.0, 3: 2.0, 4: 0.10, 5: 0.60, 6: 1.5},
+    # UNKNOWN replaces NEUTRAL (constituent signals in conflict — empirically the
+    # worst-ROI state). Keep NEUTRAL as alias so historical CH replays still resolve.
+    "UNKNOWN":       {0: 0.40, 1: 0.30, 2: 0.0, 3: 2.0, 4: 0.0,  5: 0.50, 6: 1.5},
     "NEUTRAL":       {0: 0.40, 1: 0.30, 2: 0.0, 3: 2.0, 4: 0.0,  5: 0.50, 6: 1.5},
     "MILD_NEGATIVE": {0: 0.20, 1: 0.20, 2: 0.0, 3: 1.5, 4: 0.0,  5: 0.30, 6: 1.2},
     "UNFAVORABLE":   {0: 0.0,  1: 0.0,  2: 0.0, 3: 1.5, 4: 0.0,  5: 0.0,  6: 1.2},
 }
 
-# Base S6 (NEUTRAL row above) — exposed for quick reference
-S6_BASE: Dict[int, float] = S6_MULT["NEUTRAL"]
+# Base S6 — UNKNOWN/NEUTRAL rows above are identical (alias)
+S6_BASE: Dict[int, float] = S6_MULT["UNKNOWN"]
 
 
 # ── IRP filter threshold tables keyed by advisory_label (Strategy S6_IRP) ─────
@@ -121,13 +124,14 @@ S6_BASE: Dict[int, float] = S6_MULT["NEUTRAL"]
 IRP_PARAMS: Dict[str, Dict[str, float]] = {
     "FAVORABLE":     {"alignment_min": 0.15, "noise_max": 640.0, "latency_max": 24},
     "MILD_POSITIVE": {"alignment_min": 0.17, "noise_max": 560.0, "latency_max": 22},
-    "NEUTRAL":       {"alignment_min": 0.20, "noise_max": 500.0, "latency_max": 20},
+    "UNKNOWN":       {"alignment_min": 0.20, "noise_max": 500.0, "latency_max": 20},
+    "NEUTRAL":       {"alignment_min": 0.20, "noise_max": 500.0, "latency_max": 20},  # alias
     "MILD_NEGATIVE": {"alignment_min": 0.22, "noise_max": 440.0, "latency_max": 18},
     "UNFAVORABLE":   {"alignment_min": 0.25, "noise_max": 380.0, "latency_max": 15},
 }
 
-# Gold-spec thresholds (NEUTRAL row)
-IRP_GOLD: Dict[str, float] = IRP_PARAMS["NEUTRAL"]
+# Gold-spec thresholds (UNKNOWN/NEUTRAL row)
+IRP_GOLD: Dict[str, float] = IRP_PARAMS["UNKNOWN"]
 
 
 # ── GateResult ─────────────────────────────────────────────────────────────────
@@ -157,7 +161,8 @@ def strategy_A_lev_scale(adv: dict) -> GateResult:
     mult_map = {
         "UNFAVORABLE":    0.50,
         "MILD_NEGATIVE":  0.75,
-        "NEUTRAL":        1.00,
+        "UNKNOWN":        1.00,
+        "NEUTRAL":        1.00,  # alias — historical CH replays
         "MILD_POSITIVE":  1.00,
         "FAVORABLE":      1.00,
     }
diff --git a/prod/docs/ESOF_LABEL_MIGRATION.md b/prod/docs/ESOF_LABEL_MIGRATION.md
new file mode 100644
index 0000000..fddce34
--- /dev/null
+++ b/prod/docs/ESOF_LABEL_MIGRATION.md
@@ -0,0 +1,26 @@
+# EsoF Label Migration — `NEUTRAL` → `UNKNOWN`
+
+Landed in `exp/green-s6-esof-aem-shadow-2026-04-21` on 2026-04-21.
+
+## What changed
+
+`Observability/esof_advisor.py` emits `UNKNOWN` where it previously emitted `NEUTRAL` (mid-band, `-0.05 < advisory_score <= 0.05`). All downstream consumers keep a `NEUTRAL` alias so historical CH rows and replays continue to resolve.
+
+## Why
+
+Findings from the 637-trade EsoF retrospective (`prod/docs/EsoF_BLUE_IMPLEMENTATION_CURR_AND_RESEARCH.md`) show the mid-band isn't a benign middle — it's the **worst-ROI regime**, corresponding to states where the constituent liq/session/DoW/slot/cell signals are **in conflict**. "NEUTRAL" connoted "no strong read, probably fine"; the data says the opposite. The orchestrator-top EsoF gate uses `UNKNOWN` → `0.25x` sizer mult to encode "stand mostly aside" instead of "trade normally".
+
+## Touched files
+
+- `Observability/esof_advisor.py` — emitter renamed; `LABEL_COLOR` carries both keys.
+- `Observability/esof_gate.py` — `S6_MULT`, `IRP_PARAMS`, Strategy A mult map all keyed on `UNKNOWN` with a `NEUTRAL` alias entry.
+- `nautilus_dolphin/nautilus_dolphin/nautilus/esf_alpha_orchestrator.py` — new GREEN-only `esof_sizing_table` looks up by label; missing/None label defaults to `UNKNOWN`.
+
+## CH / HZ
+
+- CH `dolphin.trade_events`, `dolphin.esof_advisory_log`: no DDL change (labels are strings). Historical rows keep `NEUTRAL`; new rows get `UNKNOWN`. Any dashboard/query filtering on the label needs to `IN ('NEUTRAL','UNKNOWN')` or be migrated.
+- HZ `DOLPHIN_FEATURES.esof_advisor_latest`: next advisor tick overwrites with `UNKNOWN`. Consumers reading stale snapshots across the cutover should treat both as equivalent.
+
+## Rollback
+
+Revert the three files above. The `NEUTRAL` alias means a partial rollback (advisor only) is safe without cascading breakage.