from __future__ import annotations import json import math from pathlib import Path import matplotlib.pyplot as plt import numpy as np import pandas as pd from matplotlib.colors import ListedColormap from matplotlib.font_manager import FontProperties ROOT = Path(__file__).resolve().parent OUT = ROOT / "outputs" FIG = OUT / "figures" OUT.mkdir(exist_ok=True) FIG.mkdir(exist_ok=True) FONT_PATH = ROOT.parents[1] / "public" / "fonts" / "NotoSansTC-Regular.ttf" FONT = FontProperties(fname=str(FONT_PATH)) if FONT_PATH.exists() else None plt.rcParams["axes.unicode_minus"] = False audit = pd.read_csv(ROOT / "case-audit.csv") qual = pd.read_csv(ROOT / "qualitative-cases.csv") binary_columns = [ "operator_controlled_site", "operator_controlled_social", "verified_ecommerce", "delivery_observed", "heritage_narrative", "relationship_narrative", "professional_advice", "local_story", "b2b_signal", "public_price", ] def wilson_interval(successes: int, total: int, z: float = 1.959963984540054) -> tuple[float, float]: p = successes / total denominator = 1 + z * z / total centre = (p + z * z / (2 * total)) / denominator radius = z * math.sqrt((p * (1 - p) + z * z / (4 * total)) / total) / denominator return centre - radius, centre + radius def savefig(name: str) -> None: plt.tight_layout() plt.savefig(FIG / name, dpi=200, bbox_inches="tight", facecolor="white") plt.close() # Figure 1: region-balanced sampling frame. region_order = ["北", "中", "南", "東"] region_counts = audit["region"].value_counts().reindex(region_order) plt.figure(figsize=(8.2, 4.4)) bars = plt.bar(region_order, region_counts, color=["#315c4c", "#6f3519", "#b26a32", "#d9a86c"]) plt.title("圖1　60家樣本的區域配置", fontproperties=FONT, fontsize=15) plt.ylabel("案例數", fontproperties=FONT) plt.xticks(fontproperties=FONT) plt.yticks(range(0, 17, 3), fontproperties=FONT) for bar, value in zip(bars, region_counts): plt.text(bar.get_x() + bar.get_width() / 2, value + 0.25, str(value), ha="center", fontproperties=FONT) plt.ylim(0, 17) savefig("figure-1-regional-sample.png") # Figure 2: visibility tiers, deliberately neutral colors and no quality ranking. tier_counts = audit["visibility_tier"].value_counts().reindex(["A", "B", "C"], fill_value=0) tier_labels = ["A 行政／目錄訊號", "B 第三方市場頁", "C 業者控制通路"] plt.figure(figsize=(8.2, 4.7)) bars = plt.barh(tier_labels, tier_counts.values, color=["#c8c2ba", "#d9a86c", "#315c4c"]) plt.title("圖2　標準搜尋可觀察的最高公開層級", fontproperties=FONT, fontsize=15) plt.xlabel("案例數（不是能力或品質評分）", fontproperties=FONT) plt.xticks(fontproperties=FONT) plt.yticks(fontproperties=FONT) for bar, value in zip(bars, tier_counts.values): plt.text(value + 0.5, bar.get_y() + bar.get_height() / 2, f"{value}（{value / len(audit):.1%}）", va="center", fontproperties=FONT) plt.xlim(0, max(tier_counts.values) * 1.18) savefig("figure-2-visibility-tiers.png") # Figure 3: observable signals. Zero means not observed on the reviewed public pages. signal_labels = { "operator_controlled_site": "自有網站", "operator_controlled_social": "自有社群／LINE", "verified_ecommerce": "可驗證線上交易", "delivery_observed": "明示配送", "heritage_narrative": "世代／歷史敘事", "relationship_narrative": "關係敘事", "professional_advice": "選米／烹煮知識", "local_story": "地方／生產者故事", "b2b_signal": "商用服務訊號", "public_price": "公開規格價格", } signal_counts = audit[binary_columns].sum().sort_values() plt.figure(figsize=(9, 6.2)) bars = plt.barh([signal_labels[x] for x in signal_counts.index], signal_counts.values, color="#6f3519") plt.title("圖3　60家樣本中的公開可觀察行銷訊號", fontproperties=FONT, fontsize=15) plt.xlabel("觀察到的案例數；未觀察到不等於實際沒有", fontproperties=FONT) plt.xticks(fontproperties=FONT) plt.yticks(fontproperties=FONT) for bar, value in zip(bars, signal_counts.values): plt.text(value + 0.12, bar.get_y() + bar.get_height() / 2, str(int(value)), va="center", fontproperties=FONT) plt.xlim(0, max(10, signal_counts.max() + 2)) savefig("figure-3-observable-signals.png") # Figure 4: regional distribution of controlled channels; descriptive only. controlled = audit.assign(controlled=(audit["visibility_tier"] == "C").astype(int)) region_controlled = controlled.groupby("region")["controlled"].agg(["sum", "count"]).reindex(region_order) plt.figure(figsize=(8.2, 4.7)) bars = plt.bar(region_order, region_controlled["sum"], color="#315c4c") plt.title("圖4　各區域觀察到的業者控制通路", fontproperties=FONT, fontsize=15) plt.ylabel("案例數（各區樣本皆為15家）", fontproperties=FONT) plt.xticks(fontproperties=FONT) plt.yticks(range(0, 5), fontproperties=FONT) for bar, value in zip(bars, region_controlled["sum"]): plt.text(bar.get_x() + bar.get_width() / 2, value + 0.08, str(int(value)), ha="center", fontproperties=FONT) plt.ylim(0, 4) savefig("figure-4-regional-controlled-channels.png") # Figure 5: a sensitivity funnel, not a conversion funnel or causal sequence. funnel_labels = ["納入抽樣框", "具市場內容頁（B或C）", "業者控制通路（C）", "可驗證線上交易", "明示配送", "公開規格價格"] funnel_values = [len(audit), int((audit.visibility_tier != "A").sum()), int((audit.visibility_tier == "C").sum()), int(audit.verified_ecommerce.sum()), int(audit.delivery_observed.sum()), int(audit.public_price.sum())] plt.figure(figsize=(9, 5.2)) bars = plt.barh(funnel_labels[::-1], funnel_values[::-1], color=["#c8c2ba", "#d9a86c", "#b26a32", "#6f3519", "#315c4c", "#315c4c"]) plt.title("圖5　不同嚴格度下的數位轉譯可觀察範圍", fontproperties=FONT, fontsize=15) plt.xlabel("案例數；各列是定義門檻，不代表顧客轉換路徑", fontproperties=FONT) plt.xticks(fontproperties=FONT) plt.yticks(fontproperties=FONT) for bar, value in zip(bars, funnel_values[::-1]): plt.text(value + 0.6, bar.get_y() + bar.get_height() / 2, str(value), va="center", fontproperties=FONT) plt.xlim(0, 66) savefig("figure-5-definition-sensitivity.png") # Figure 6: qualitative matrix uses two discrete colors and cell labels. qual_columns = ["relationship_narrative", "heritage_narrative", "professional_advice", "operator_controlled_channel", "transaction_function", "delivery_signal", "product_story", "service_contact"] qual_labels = ["關係敘事", "歷史傳承", "專業建議", "自有通路", "交易功能", "配送訊號", "產品故事", "服務聯絡"] matrix = qual[qual_columns].to_numpy(dtype=int) plt.figure(figsize=(10.5, 6.2)) plt.imshow(matrix, aspect="auto", cmap=ListedColormap(["#eeeae4", "#315c4c"]), vmin=0, vmax=1) plt.xticks(range(len(qual_labels)), qual_labels, rotation=35, ha="right", fontproperties=FONT) plt.yticks(range(len(qual)), qual["case_name"], fontproperties=FONT) plt.title("圖6　資訊豐富案例的質性文本訊號矩陣", fontproperties=FONT, fontsize=15) for i in range(matrix.shape[0]): for j in range(matrix.shape[1]): plt.text(j, i, "有" if matrix[i, j] else "未見", ha="center", va="center", color="white" if matrix[i, j] else "#6b625a", fontsize=8, fontproperties=FONT) plt.figtext(0.5, 0.005, "灰色＝未觀察到；綠色＝觀察到。這不是成效評分。", ha="center", fontproperties=FONT, fontsize=9) savefig("figure-6-qualitative-matrix.png") controlled_n = int((audit.visibility_tier == "C").sum()) market_facing_n = int((audit.visibility_tier != "A").sum()) ci_low, ci_high = wilson_interval(controlled_n, len(audit)) tier_summary = pd.DataFrame({ "visibility_tier": ["A", "B", "C"], "count": tier_counts.values, "share": tier_counts.values / len(audit), }) tier_summary.to_csv(OUT / "visibility-tier-summary.csv", index=False, encoding="utf-8-sig") signal_summary = pd.DataFrame({ "signal": signal_counts.index, "count": signal_counts.values.astype(int), "share": signal_counts.values / len(audit), }) signal_summary.to_csv(OUT / "signal-summary.csv", index=False, encoding="utf-8-sig") region_controlled.reset_index().to_csv(OUT / "regional-controlled-summary.csv", index=False, encoding="utf-8-sig") results = { "sample_n": int(len(audit)), "regions": {k: int(v) for k, v in region_counts.items()}, "visibility_tiers": {k: int(v) for k, v in tier_counts.items()}, "market_facing_pages_n": market_facing_n, "market_facing_pages_share": market_facing_n / len(audit), "operator_controlled_channel_n": controlled_n, "operator_controlled_channel_share": controlled_n / len(audit), "operator_controlled_channel_wilson_95": [ci_low, ci_high], "signals": {key: int(audit[key].sum()) for key in binary_columns}, "qualitative_cases_n": int(len(qual)), "interpretation_boundary": "All values are publicly observable content signals as of 2026-07-17, not business capability, quality, revenue, loyalty, or causal effects.", } (ROOT / "analysis-results.json").write_text(json.dumps(results, ensure_ascii=False, indent=2) + "\n", encoding="utf-8") print(json.dumps(results, ensure_ascii=False, indent=2))