diff --git a/scripts/build_blend_bsdf_panel.py b/scripts/build_blend_bsdf_panel.py
index f0a4c4a..db92d5d 100644
--- a/scripts/build_blend_bsdf_panel.py
+++ b/scripts/build_blend_bsdf_panel.py
@@ -32,6 +32,7 @@
 import numpy as np
 
 from sm_tias_lightsim.evaluation.normalize import apply_mean_anchor
+from sm_tias_lightsim.evaluation.roi_presets import SIM_ROI, extract_roi_xywh
 
 # ---------------------------------------------------------------------------
 # 定数
@@ -44,9 +45,6 @@
 BASELINE_NPY = ROOT / "output" / "baseline" / "luminance.npy"
 BLEND_OUTPUT_BASE = ROOT / "output" / "blend_sweep"
 
-# ROI 設定（configs/eval_baseline.yaml / eval_baseline_dng_target.yaml と整合）
-SIM_ROI = {"x": 24, "y": 37, "w": 464, "h": 450}
-
 # mean 正規化の target_mean（eval_baseline_dng_target.yaml の target_mean と整合）
 TARGET_MEAN = 207.82
 
@@ -82,12 +80,6 @@
     return f"w{_format_param(weight)}_a{_format_param(alpha)}"
 
 
-def _extract_roi(arr: np.ndarray, roi: dict) -> np.ndarray:
-    """ROI 矩形を切り出して float64 に変換する."""
-    x, y, w, h = roi["x"], roi["y"], roi["w"], roi["h"]
-    return arr[y: y + h, x: x + w].astype(np.float64)
-
-
 def _rec709_luma(rgb: np.ndarray) -> np.ndarray:
     """RGB 配列を Rec.709 luma に変換する (H, W, 3) → (H, W)."""
     return 0.2126 * rgb[..., 0] + 0.7152 * rgb[..., 1] + 0.0722 * rgb[..., 2]
@@ -112,7 +104,7 @@
     arr = np.load(npy_path)
     if arr.ndim == 3:
         arr = _rec709_luma(arr)
-    roi = _extract_roi(arr, SIM_ROI)
+    roi = extract_roi_xywh(arr, SIM_ROI)
     normalized, _ = apply_mean_anchor(roi, TARGET_MEAN)
     return normalized
 
diff --git a/scripts/build_led_distribution_panel.py b/scripts/build_led_distribution_panel.py
index 2b579af..455d88d 100644
--- a/scripts/build_led_distribution_panel.py
+++ b/scripts/build_led_distribution_panel.py
@@ -22,6 +22,8 @@
 import matplotlib.font_manager as fm
 import matplotlib.pyplot as plt
 
+from sm_tias_lightsim.evaluation.roi_presets import SIM_ROI, extract_roi_xywh
+
 ROOT = Path(__file__).resolve().parents[1]
 SWEEP_BASE = ROOT / "output" / "led_distribution_sweep"
 SUMMARY_CSV = SWEEP_BASE / "summary.csv"
@@ -66,7 +68,7 @@
         if not npy.exists():
             lums.append(None)
             continue
-        arr = np.load(npy)
+        arr = extract_roi_xywh(np.load(npy), SIM_ROI)
         lums.append(arr.mean(axis=-1) if arr.ndim == 3 else arr)
 
     valid = [x for x in lums if x is not None]
diff --git a/scripts/build_led_group_sanity_panel.py b/scripts/build_led_group_sanity_panel.py
index 0d40937..dcca696 100644
--- a/scripts/build_led_group_sanity_panel.py
+++ b/scripts/build_led_group_sanity_panel.py
@@ -32,6 +32,8 @@
 import matplotlib.font_manager as fm
 import matplotlib.pyplot as plt
 
+from sm_tias_lightsim.evaluation.roi_presets import SIM_ROI, extract_roi_xywh
+
 # ---------------------------------------------------------------------------
 # 定数
 # ---------------------------------------------------------------------------
@@ -63,13 +65,13 @@
         if not npy_path.exists():
             print(f"エラー: {npy_path} が見つかりません", file=sys.stderr)
             return 1
-        lums[g] = np.load(npy_path)
+        lums[g] = extract_roi_xywh(np.load(npy_path), SIM_ROI)
 
     # ベースライン (全 ON) 読み込み
     if not BASELINE_NPY.exists():
         print(f"エラー: ベースライン {BASELINE_NPY} が見つかりません", file=sys.stderr)
         return 1
-    baseline = np.load(BASELINE_NPY)
+    baseline = extract_roi_xywh(np.load(BASELINE_NPY), SIM_ROI)
 
     # スカラ輝度に変換 (RGB なら平均，2D ならそのまま)
     def to_2d(arr: np.ndarray) -> np.ndarray:
diff --git a/scripts/build_luminance_evolution_panel.py b/scripts/build_luminance_evolution_panel.py
index 5d4a809..689a0c4 100644
--- a/scripts/build_luminance_evolution_panel.py
+++ b/scripts/build_luminance_evolution_panel.py
@@ -32,14 +32,11 @@
 
 from sm_tias_lightsim.evaluation.normalize import apply_mean_anchor
 from sm_tias_lightsim.evaluation.oetf import apply_srgb_oetf, solve_exposure
+from sm_tias_lightsim.evaluation.roi_presets import REAL_ROI, SIM_ROI, extract_roi_xywh
 
 ROOT = Path(__file__).resolve().parents[1]
 TARGET_MEAN = 146.7146
 
-# ROI 設定（configs と整合）
-SIM_ROI = {"x": 24, "y": 37, "w": 464, "h": 450}  # configs/eval_baseline.yaml
-REAL_ROI = {"x": 539, "y": 1015, "w": 1400, "h": 1409}  # configs/eval_real.yaml
-
 # 共有 colorbar の値域決定方針: 実機 ROI のパーセンタイル
 # 5〜95 パーセンタイルで外れ値（センサノイズ・最暗最明画素）を除く．
 REAL_VMIN_PERCENTILE = 5.0
@@ -51,12 +48,6 @@
     return 0.2126 * rgb[..., 0] + 0.7152 * rgb[..., 1] + 0.0722 * rgb[..., 2]
 
 
-def _extract_roi(arr: np.ndarray, roi: dict) -> np.ndarray:
-    """ROI 矩形を切り出して float64 に変換する."""
-    x, y, w, h = roi["x"], roi["y"], roi["w"], roi["h"]
-    return arr[y : y + h, x : x + w].astype(np.float64)
-
-
 def _load_real() -> np.ndarray:
     """実機 PNG → Rec.709 luma → ROI 抽出 → mean アンカー（恒等近い）."""
     png = (
@@ -68,7 +59,7 @@
         luma = _rec709_luma(img.astype(np.float64))
     else:
         luma = img.astype(np.float64)
-    roi = _extract_roi(luma, REAL_ROI)
+    roi = extract_roi_xywh(luma, REAL_ROI)
     # 実機は既に mean ≈ 146.71 なので alpha ≈ 1.0（揺れ吸収）
     normalized, _ = apply_mean_anchor(roi, TARGET_MEAN)
     return normalized
@@ -79,7 +70,7 @@
     arr = np.load(npy_path)
     if arr.ndim == 3:
         arr = _rec709_luma(arr)
-    roi = _extract_roi(arr, SIM_ROI)
+    roi = extract_roi_xywh(arr, SIM_ROI)
     normalized, _ = apply_mean_anchor(roi, TARGET_MEAN)
     return normalized
 
@@ -89,7 +80,7 @@
     arr = np.load(npy_path)
     if arr.ndim == 3:
         arr = _rec709_luma(arr)
-    roi = _extract_roi(arr, SIM_ROI)
+    roi = extract_roi_xywh(arr, SIM_ROI)
     scale, _ = solve_exposure(roi, TARGET_MEAN)
     return apply_srgb_oetf(roi, scale, quantize_uint8=True)
 
diff --git a/scripts/build_max_depth_panel.py b/scripts/build_max_depth_panel.py
index e4cced1..99c785e 100644
--- a/scripts/build_max_depth_panel.py
+++ b/scripts/build_max_depth_panel.py
@@ -29,6 +29,8 @@
 import matplotlib.font_manager as fm
 import matplotlib.pyplot as plt
 
+from sm_tias_lightsim.evaluation.roi_presets import SIM_ROI, extract_roi_xywh
+
 ROOT = Path(__file__).resolve().parents[1]
 SWEEP_BASE = ROOT / "output" / "max_depth_sweep"
 SUMMARY_CSV = SWEEP_BASE / "summary.csv"
@@ -75,7 +77,7 @@
             print(f"警告: {npy} 見つからずスキップ", file=sys.stderr)
             lums.append(None)
             continue
-        arr = np.load(npy)
+        arr = extract_roi_xywh(np.load(npy), SIM_ROI)
         lums.append(arr.mean(axis=-1) if arr.ndim == 3 else arr)
 
     # 共有 vmin/vmax
diff --git a/scripts/build_reflectance_panel.py b/scripts/build_reflectance_panel.py
index bd7da9b..fed3750 100644
--- a/scripts/build_reflectance_panel.py
+++ b/scripts/build_reflectance_panel.py
@@ -28,6 +28,8 @@
 import matplotlib.font_manager as fm
 import matplotlib.pyplot as plt
 
+from sm_tias_lightsim.evaluation.roi_presets import SIM_ROI, extract_roi_xywh
+
 ROOT = Path(__file__).resolve().parents[1]
 SWEEP_BASE = ROOT / "output" / "reflectance_sweep"
 SUMMARY_CSV = SWEEP_BASE / "summary.csv"
@@ -74,7 +76,7 @@
         if not npy.exists():
             lums.append(None)
             continue
-        arr = np.load(npy)
+        arr = extract_roi_xywh(np.load(npy), SIM_ROI)
         lums.append(arr.mean(axis=-1) if arr.ndim == 3 else arr)
 
     valid = [x for x in lums if x is not None]
diff --git a/scripts/build_sim_vs_darkroom_panel.py b/scripts/build_sim_vs_darkroom_panel.py
index 2b34419..ed1f090 100644
--- a/scripts/build_sim_vs_darkroom_panel.py
+++ b/scripts/build_sim_vs_darkroom_panel.py
@@ -46,6 +46,7 @@
     calc_radial_min_max_ratio,
     calc_radial_profile,
 )
+from sm_tias_lightsim.evaluation.roi_presets import REAL_ROI, SIM_ROI, extract_roi_xywh
 from sm_tias_lightsim.evaluation.zones import (
     calc_zone_stats,
     classify_zones,
@@ -62,12 +63,10 @@
 REAL_META = (
     ROOT / "reference_measurement" / "quantitative" / "MiniTIAS_QM_20260601_124500.meta.json"
 )
-# 白板 ROI (portrait, 旧 PNG と同一座標. DNG portrait は 2448x3264 で同一フレーミング)
-REAL_ROI = {"x": 539, "y": 1015, "w": 1400, "h": 1409}
+# REAL_ROI（白板 portrait）/ SIM_ROI は roi_presets に集約（configs と整合）.
 
 # シミュ baseline (max_depth=16, R=0.85, spp=1024)
 SIM_NPY = ROOT / "output" / "baseline" / "luminance.npy"
-SIM_ROI = {"x": 24, "y": 37, "w": 464, "h": 450}  # configs/eval_baseline_dng_target.yaml
 
 OUT_PATH = ROOT / "output" / "_slide_panels" / "sim_vs_darkroom_normalized.png"
 
@@ -77,12 +76,6 @@
     return _REC709[0] * rgb[..., 0] + _REC709[1] * rgb[..., 1] + _REC709[2] * rgb[..., 2]
 
 
-def _extract_roi(arr: np.ndarray, roi: dict) -> np.ndarray:
-    """ROI 矩形を float64 で切り出す."""
-    x, y, w, h = roi["x"], roi["y"], roi["w"], roi["h"]
-    return arr[y : y + h, x : x + w].astype(np.float64)
-
-
 def _upsample(grid: np.ndarray, shape: tuple[int, int]) -> np.ndarray:
     """(rows, cols) グリッドを (H, W) へバイリニア補間する (run_uniformity_dng.py と同一)."""
     rows, cols = grid.shape
@@ -117,7 +110,7 @@
     rgb = rgb * _lsc_gain_rgb(REAL_META, rgb.shape[:2])  # LSC 順方向適用 (ビネット除去)
     luma = _rec709_luma(rgb) * 255.0  # 0〜255 linear スケール
     luma = np.rot90(luma, -1)  # portrait (PNG / ROI と同じ向き)
-    return _extract_roi(luma, REAL_ROI)
+    return extract_roi_xywh(luma, REAL_ROI)
 
 
 def _load_sim() -> np.ndarray:
@@ -125,7 +118,7 @@
     arr = np.load(SIM_NPY)
     if arr.ndim == 3:
         arr = _rec709_luma(arr)
-    return _extract_roi(arr, SIM_ROI)
+    return extract_roi_xywh(arr, SIM_ROI)
 
 
 def _minmax_normalize(roi: np.ndarray) -> np.ndarray:
diff --git a/src/sm_tias_lightsim/evaluation/roi_presets.py b/src/sm_tias_lightsim/evaluation/roi_presets.py
new file mode 100644
index 0000000..e0c3332
--- /dev/null
+++ b/src/sm_tias_lightsim/evaluation/roi_presets.py
@@ -0,0 +1,36 @@
+"""キャリブレーション基準 ROI のプリセットと dict ベース抽出ヘルパ.
+
+スライド／比較パネル生成スクリプト（scripts/build_*_panel.py 等）が共通で使う
+ROI 座標を一元管理する．座標は評価パイプライン（sm-evaluate）の
+configs/eval_baseline*.yaml / eval_real.yaml と同一．
+
+評価本体（シーンスキーマ ``RoiConfig`` を使う側）は
+``sm_tias_lightsim.evaluation.roi.extract_roi`` を使うこと．
+本モジュールは ``{"x", "y", "w", "h"}`` dict を扱うスクリプト向けの軽量ヘルパ．
+"""
+
+from __future__ import annotations
+
+import numpy as np
+
+# シミュ luminance.npy 用 ROI（configs/eval_baseline*.yaml と整合）.
+SIM_ROI = {"x": 24, "y": 37, "w": 464, "h": 450}
+
+# 実機測定画像用 ROI（configs/eval_real.yaml と整合）.
+REAL_ROI = {"x": 539, "y": 1015, "w": 1400, "h": 1409}
+
+
+def extract_roi_xywh(arr: np.ndarray, roi: dict) -> np.ndarray:
+    """ROI 矩形（``{"x", "y", "w", "h"}``）を切り出して float64 で返す.
+
+    先頭 2 軸を (H, W) とみなすため，2D（輝度）・3D（RGB）どちらの配列も扱える．
+
+    Args:
+        arr: 先頭 2 軸が (H, W) の配列．
+        roi: ``{"x", "y", "w", "h"}`` 形式の ROI 辞書（``SIM_ROI`` / ``REAL_ROI``）．
+
+    Returns:
+        ROI でクロップした float64 配列（入力と同じ次元数）．
+    """
+    x, y, w, h = roi["x"], roi["y"], roi["w"], roi["h"]
+    return arr[y : y + h, x : x + w].astype(np.float64)