5D Library › 分拆经济学、防套利身份锁定与多阶授权阶梯 · English

第三章 转售套利与身份哈希锁

附赠名额为什么会变成生意

个人年卡附赠几个「替亲友测算」的名额,看起来是一项温和的福利。可是对一位手上有客户的算命师来说,每个附赠名额都是一张可以转卖的单次测算:她买一张卡,把名额按略低于官方单价的价格卖给自己的客户,赚取差价。这与经济学中的转售套利完全相同:只要同一件商品对不同买家有不同的价格,而商品可以转手,价差就会被中间人吃掉 (Shapiro and Varian 1998)。

公司另有一条合法的渠道:分销计划。渠道伙伴推荐一位客户全价购买单次深度报告 p2p_2,拿到 γ=50%\gamma = 50\% 的佣金。于是同一位算命师面前有两个选择:推荐,每位客户赚 γp2\gamma p_2;转售,每位客户赚 s−P/(k+1)s – P/(k+1),其中 PP 是卡价,kk 是附赠名额,ss 是她的转售价。

套利何时胜过推荐

程序取 p2=99p_2 = 99、卡价 P=199P = 199,比较附赠 kk 个名额、不加身份锁时的两条路。

推导

# Chapter 3. [p05] Card resale arbitrage against a 50% affiliate channel.
# Single deep report p2 = 99 (USD). Personal annual card P = 199 that
# covers its holder plus k extra readings, with no identity lock, so a
# reseller (a fortune-teller with clients) can use all k + 1 slots.
# She can (i) refer each client as an affiliate and earn g * p2,
# g = 0.5, or (ii) buy cards and resell slots at s, either s = p2 or
# s = 0.8 * p2 (undercutting). Per client served, the firm receives
# p2 (direct), (1 - g) * p2 (affiliate) or P / (k + 1) (resold slot).
P2, CARD, G = 99.0, 199.0, 0.5

def margin(k, s):
    return s - CARD / (k + 1)

if __name__ == "__main__":
    aff = G * P2
    print("   affiliate margin per client g * p2 = {:.2f}".format(aff))
    print("    k  cost/slot  margin s=p2  margin s=0.8p2  firm gets"
          "  vs affiliate")
    for k in range(0, 9):
        cost = CARD / (k + 1)
        print("   {:2d} {:10.2f} {:12.2f} {:15.2f} {:10.2f} {:13.2f}".format(
            k, cost, margin(k, P2), margin(k, 0.8 * P2), cost,
            cost - (1 - G) * P2))
    for s, name in ((P2, "s = p2"), (0.8 * P2, "s = 0.8 p2")):
        k = next(k for k in range(0, 99) if margin(k, s) > aff)
        print("   resale beats referral ({}) from k = {}".format(name, k))
    print("   threshold k + 1 > P / ((1 - g) p2) = {:.3f}".format(
        CARD / ((1 - G) * P2)))
    print("\n   identity lock: the card serves its holder only; each extra")
    print("   chart costs q. Reseller margin p2 - q against g * p2:")
    for q in (29.0, 49.5, 69.0, 99.0):
        m = P2 - q
        print("     q = {:4.1f}: margin {:5.1f}  IC: {:3s}"
              "  loss on a poached direct sale: {:4.1f}".format(
                  q, m, "yes" if m <= aff + 1e-9 else "no", P2 - q))
    print("\nReading: at s = p2 the reseller gains exactly when the firm")
    print("loses against its own affiliate channel. With a lock, any")
    print("q of at least (1 - g) p2 stops resale against referral; only")
    print("q = p2 also leaves a poached direct sale whole.")
#    affiliate margin per client g * p2 = 49.50
#     k  cost/slot  margin s=p2  margin s=0.8p2  firm gets  vs affiliate
#     0     199.00      -100.00         -119.80     199.00        149.50
#     1      99.50        -0.50          -20.30      99.50         50.00
#     2      66.33        32.67           12.87      66.33         16.83
#     3      49.75        49.25           29.45      49.75          0.25
#     4      39.80        59.20           39.40      39.80         -9.70
#     5      33.17        65.83           46.03      33.17        -16.33
#     6      28.43        70.57           50.77      28.43        -21.07
#     7      24.88        74.12           54.33      24.88        -24.62
#     8      22.11        76.89           57.09      22.11        -27.39
#    resale beats referral (s = p2) from k = 4
#    resale beats referral (s = 0.8 p2) from k = 6
#    threshold k + 1 > P / ((1 - g) p2) = 4.020
#
#    identity lock: the card serves its holder only; each extra
#    chart costs q. Reseller margin p2 - q against g * p2:
#      q = 29.0: margin  70.0  IC: no   loss on a poached direct sale: 70.0
#      q = 49.5: margin  49.5  IC: yes  loss on a poached direct sale: 49.5
#      q = 69.0: margin  30.0  IC: yes  loss on a poached direct sale: 30.0
#      q = 99.0: margin   0.0  IC: yes  loss on a poached direct sale:  0.0
#
# Reading: at s = p2 the reseller gains exactly when the firm
# loses against its own affiliate channel. With a lock, any
# q of at least (1 - g) p2 stops resale against referral; only
# q = p2 also leaves a poached direct sale whole.

当转售价等于官方单价(s=p2s = p_2)时,转售胜过推荐的条件是 P/(k+1)<(1−γ)p2P/(k+1) < (1-\gamma) p_2,即 k+1>P/((1−γ)p2)k + 1 > P / ((1-\gamma) p_2),在这里等于 4.020:从附赠 4 个名额起,转售比推荐划算。这个门槛恰好也是公司相对自己的分销渠道开始亏损的位置:k=3k = 3 时公司每个名额拿到 49.75,比分销渠道的 49.50 只多 0.25;k=4k = 4 时只拿到 39.80,比分销少 9.70。在 s=p2s = p_2 时,算命师赚到的正是公司相对分销渠道亏掉的。 若算命师以八折转售来抢客户,门槛推后到 k=6k = 6。

推导

加上身份锁之后。 卡只服务持卡人本人(k=0k = 0),额外命盘按每份 qq 计费。算命师为客户买额外命盘再转售,每位客户赚 p2−qp_2 – q;推荐则赚 γp2\gamma p_2。转售不划算的条件是 p2−q≤γp2⇔q≥(1−γ)p2.p_2 – q \le \gamma p_2 \iff q \ge (1-\gamma) p_2 . 在 γ=50%\gamma = 50\% 时,额外命盘只要不低于单价的一半(49.5),转售就不再胜过推荐。但如果被转售的客户本来会直接全价购买,公司每单损失 p2−qp_2 – q;只有 q=p2q = p_2 时这一损失为零。

所以提纲中「一卡一身份锁死 + 额外命盘按单次全价计费是唯一激励相容均衡」需要修正:相对分销渠道,激励相容的额外命盘价格是一个区间 q≥(1−γ)p2q \ge (1-\gamma) p_2;全价之所以特别,是因为它同时保护了公司的直销:转售者抢走的若是本来会直接下单的客户,任何低于全价的 qq 都让公司吃亏。

身份锁:一份命盘、一个哈希

身份锁的技术核心是把出生资料规范化后取哈希:日期统一写成「年-月-日」,时间精确到分钟,姓名去空格并统一大小写,再加上本公司的盐值,用 SHA-256 计算摘要 (National Institute of Standards and Technology 2015)。同一个人的两种写法得到同一个键,差一分钟就是另一个键。

推导

# Chapter 3. [p06] An identity lock: one card, one birth-record hash.
# The record is normalised (date as YYYY-MM-DD, time to the minute,
# name case-folded with spaces removed) and hashed with SHA-256 and a
# per-firm salt. Two spellings of one record give one hash. Then the
# birthday problem: how many pairs of DIFFERENT people share a key if
# the key holds only birth data, for 100,000 card holders born over 80
# years, with time known to the double-hour or to the minute.
import hashlib

SALT = "fdl-068"

def norm(name, date, time):
    y, m, d = (int(x) for x in date.replace("/", "-").split("-"))
    hh, mm = (int(x) for x in time.split(":"))
    nm = "".join(name.split()).casefold()
    return "{}|{:04d}-{:02d}-{:02d}|{:02d}:{:02d}".format(nm, y, m, d,
                                                          hh, mm)

def key(name, date, time):
    raw = (SALT + "|" + norm(name, date, time)).encode("ascii")
    return hashlib.sha256(raw).hexdigest()[:16]

def pairs(n, m):
    return n * (n - 1) / (2 * m)

if __name__ == "__main__":
    a = key("Tan Mei Ling", "1990-1-5", "7:05")
    b = key("tan meiling", "1990/01/05", "07:05")
    c = key("Tan Mei Ling", "1990-01-05", "07:06")
    print("   {}  {}  {}".format(a, b, c))
    print("   same person, two spellings, one key: {}".format(a == b))
    print("   one minute later, a new key        : {}".format(a != c))
    n = 100000
    days = 80 * 365.25
    for label, m in (("date + double-hour", days * 12),
                     ("date + minute", days * 1440)):
        print("   key = {:18s}: {:12,.0f} keys, {:9,.1f} shared pairs"
              .format(label, m, pairs(n, m)))
    print("\nReading: normalising before hashing makes one person one")
    print("key. Birth data alone are not an identity: at 100,000 users")
    print("a double-hour key is shared by thousands of pairs of strangers,")
    print("so the lock binds the name and a verified contact as well.")
#    a66c2af375f32dcf  a66c2af375f32dcf  63bcf8d2f829f2d7
#    same person, two spellings, one key: True
#    one minute later, a new key        : True
#    key = date + double-hour:      350,640 keys,  14,259.5 shared pairs
#    key = date + minute     :   42,076,800 keys,     118.8 shared pairs
#
# Reading: normalising before hashing makes one person one
# key. Birth data alone are not an identity: at 100,000 users
# a double-hour key is shared by thousands of pairs of strangers,
# so the lock binds the name and a verified contact as well.

但出生资料本身不是身份。假设 100,000 名持卡人出生在 80 年之内,若键只含日期与时辰,共有 350,640 个不同的键,按生日问题的公式,预期有 14,259.5 对陌生人共享同一个键;精确到分钟时仍有 118.8 对。双胞胎更不必说。所以锁不能只锁出生资料,还要绑定姓名与一个经过验证的联系方式;而「同一张卡出现第二个出生哈希」应当触发的是人工核实,而不是自动封号。

假说 · 尚未验证

作者的诠释(尚未验证): 在作者的产品中,「个人年卡只绑定持卡人本人,额外命盘按全价计费,分销佣金只付给非本人购买」可以把转售套利压到可以忽略的水平,同时不损害正常用户。什么会证明这一判断错:若实施身份锁后,正常用户因身份核实而放弃购买的比例,超过锁所阻止的转售损失(第八章的实验可以同时测量两者),这一设计就得不偿失。

小结

附赠名额在 k+1k + 1 超过 P/((1−γ)p2)P/((1-\gamma)p_2) 时变成套利,本书参数下从 4 个名额开始;在 s=p2s = p_2 时,算命师的套利收益正好等于公司相对分销渠道的损失。身份锁之下,额外命盘不低于单价的一半就足以防止转售胜过推荐,全价则同时保护直销。出生哈希必须连同姓名与联系方式一起锁定。

参考文献

National Institute of Standards and Technology. 2015. Secure Hash Standard (SHS). Federal Information Processing Standards Publication Nos. 180-4. NIST.
Shapiro, Carl, and Hal R. Varian. 1998. Information Rules: A Strategic Guide to the Network Economy. Harvard Business School Press.

Fan, Y. (2026). Unbundling Economics, Anti-Arbitrage Identity Locking and Multi-Tier Licensing Ladders: Orthogonal Demand Partitioning, Resale Cannibalization Bounds, Affiliate Margin Preservation and Five-Figure B2B Upsell Equilibria [分拆经济学、防套利身份锁定与多阶授权阶梯] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090184 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).