"""PTM 事業拡大の順番シミュレーション（モンテカルロ）
数字はすべて仮置き。予測ではなく「順番の違いがどう効くか」を比べるための模型。
期間: 2026-12 (m0) 〜 2028-11 (m23)
"""
import numpy as np, json, sys

N = 4000
M = 24

# チャネル定義: lam0=月あたり見込み客, p=成約率, rev=(下限,上限)円, hrs=1件の時間,
# case=公開できる事例になる確率, cw=事例の重み(ロボットの信頼にどれだけ効くか), robot=ロボット仕事か
CH = {
    "spot":    dict(lam0=0.30, p=0.6, rev=(30e3, 60e3),   hrs=3,   case=0.0, cw=0.0, robot=True, lag=1),
    "package": dict(lam0=0.12, p=0.25, rev=(300e3, 800e3), hrs=40,  case=0.5, cw=1.0, robot=True, lag=2),
    "partner": dict(lam0=0.08, p=0.20, rev=(1e6, 3e6),     hrs=120, case=0.3, cw=1.5, robot=True, lag=3),
    "gov":     dict(lam0=0.30, p=0.7, rev=(25e3, 40e3),    hrs=8,   case=0.3, cw=0.7, robot=True, lag=4),
    "ai":      dict(lam0=0.40, p=0.30, rev=(200e3, 500e3), hrs=30,  case=0.4, cw=0.3, robot=False, lag=1),
}
# 信頼(C)が見込み客と成約率を押し上げる強さ
K_LEAD, K_CONV = 0.25, 0.10

SCEN = {
    "A 現状延長(発信だけで待つ)": {},
    "B スポット相談→見積もり→型づくり": {"spot": 0, "package": 3, "partner": 9},
    "C ロボット会社向けの型づくり先行": {"partner": 0, "package": 6},
    "D 行政・支援機関ルート先行": {"gov": 0, "package": 4, "partner": 12},
    "E AI導入支援で稼いでから": {"ai": 0, "package": 6, "gov": 9},
    "F <地域>の地域実証先行": {"minamisoma": 4, "gov": 6, "package": 10},
    "G 行政＋スポット並走→見積もり→型づくり": {"gov": 0, "spot": 0, "package": 3, "partner": 12},
}

def run(scen, rng, P):
    C = P["C0"]  # 公開事例・再現記録の信頼の蓄え
    rev_robot = rev_all = 0.0
    first_robot = None
    research_h = deal_h = 0.0
    cases = 0
    accepted = rng.random() < P["minamisoma_p"]
    backlog = []
    for m in range(M):
        # 使える時間（PTM分）
        if m < 4:
            cap = 60
        else:
            cap = 120 if (accepted and "minamisoma" in scen) else (110 if accepted else 70)
        deal_cap = cap * 0.5  # MASTER: 研究の記録が半分を下回らない
        used = 0.0
        # <地域>の実証: 合格時、m10ごろに地元の事例ができる
        if "minamisoma" in scen and accepted and m == scen["minamisoma"] + 6 and rng.random() < 0.6:
            C += 1.5; cases += 1
        # 発信からの問い合わせ（全シナリオ共通）→ 見積もり相当
        lam_in = P["inbound"] * (1 + K_LEAD * C)
        for _ in range(rng.poisson(lam_in)):
            if rng.random() < min(0.5, 0.15 * (1 + K_CONV * C)):
                backlog.append(("package", m))
        # 各チャネル
        for ch, start in scen.items():
            if ch == "minamisoma" or m < start + CH[ch]["lag"]:
                continue
            c = CH[ch]
            lam0 = c["lam0"] * (P["gov_mult"] if ch == "gov" else 1) * (P["partner_mult"] if ch == "partner" else 1)
            lam = lam0 * (1 + K_LEAD * C)
            for _ in range(rng.poisson(lam)):
                if rng.random() < min(0.9, c["p"] * (1 + K_CONV * C)):
                    backlog.append((ch, m))
            # 行政の訪問・AI支援からロボットの見積もりへの橋渡し
            if ch in ("gov", "ai"):
                bridge = 0.10 if ch == "gov" else 0.04
                for _ in range(rng.poisson(lam * c["p"])):
                    if rng.random() < bridge:
                        backlog.append(("package", m))
        # 時間の範囲でこなす（入りきらない分は翌月、3か月で流れる）
        nb = []
        for ch, mm in backlog:
            c = CH[ch]
            if m - mm > 3:
                continue
            if used + c["hrs"] <= deal_cap or (c["hrs"] > deal_cap and used == 0):
                used += min(c["hrs"], deal_cap)
                r = rng.uniform(*c["rev"])
                rev_all += r
                if c["robot"]:
                    rev_robot += r
                    if first_robot is None and ch != "spot":
                        first_robot = m
                if rng.random() < c["case"]:
                    C += c["cw"]; cases += 1
            else:
                nb.append((ch, mm))
        backlog = nb
        deal_h += used
        research_h += cap - used
        C += P["content_growth"]  # 研究の記録そのものが少しずつ信頼になる
    return dict(rev_robot=rev_robot, rev_all=rev_all, first=first_robot,
                research=research_h / (research_h + deal_h), cases=cases)

def evaluate(P, seed=1):
    out = {}
    for name, sc in SCEN.items():
        rng = np.random.default_rng(seed)
        rs = [run(sc, rng, P) for _ in range(N)]
        rr = np.array([r["rev_robot"] for r in rs]) / 1e4
        ra = np.array([r["rev_all"] for r in rs]) / 1e4
        f = np.array([r["first"] if r["first"] is not None else 99 for r in rs])
        out[name] = dict(
            p_first12=float((f < 12).mean()),
            first_med=float(np.median(f)),
            robot_p10=float(np.percentile(rr, 10)), robot_med=float(np.median(rr)), robot_p90=float(np.percentile(rr, 90)),
            all_med=float(np.median(ra)),
            cases=float(np.mean([r["cases"] for r in rs])),
            research=float(np.mean([r["research"] for r in rs])),
        )
    return out

BASE = dict(C0=0.5, inbound=0.04, minamisoma_p=0.5, gov_mult=1.0, partner_mult=1.0, content_growth=0.03)

if __name__ == "__main__":
    res = {"base": evaluate(BASE)}
    sens = {
        "行政ルートが通りにくい(×0.4)": dict(BASE, gov_mult=0.4),
        "ロボット会社が乗ってくる(×2)": dict(BASE, partner_mult=2.0),
        "<地域>に受かる(100%)": dict(BASE, minamisoma_p=1.0),
        "<地域>に落ちる(0%)": dict(BASE, minamisoma_p=0.0),
        "発信の問い合わせが3倍": dict(BASE, inbound=0.12),
    }
    for k, P in sens.items():
        res[k] = evaluate(P)
    json.dump(res, open("/home/claude/ptm_sim_results.json", "w"), ensure_ascii=False, indent=1)
    b = res["base"]
    print(f"{'シナリオ':<28}{'1年内初案件':>8}{'初案件月(中央)':>10}{'ロボ売上24か月 p10/中央/p90(万円)':>30}{'全売上中央':>8}{'事例数':>6}{'研究比':>6}")
    for n, v in b.items():
        print(f"{n:<28}{v['p_first12']:>8.0%}{v['first_med']:>10.0f}   {v['robot_p10']:>6.0f}/{v['robot_med']:>5.0f}/{v['robot_p90']:>5.0f}{v['all_med']:>10.0f}{v['cases']:>6.1f}{v['research']:>6.0%}")
    print()
    for k in sens:
        rk = sorted(res[k].items(), key=lambda x: -x[1]["robot_med"])
        print(k, " → ロボ売上中央の順位:", " > ".join(f"{n[0]}({v['robot_med']:.0f})" for n, v in rk))
