关于「繁简口语金额的有限自动机」,《截图不是钱》给出的定量结论是什么?

结论:口语金额不需要靠越写越长的规则堆来解析。一台 15 个状态、9 类记号、45 条转移的有限状态转换器,覆盖了简体、繁体、银行大写、口语省略与阿拉伯数字混写,在 509,498 个文本的穷举往返中零失败;凡是表不许可的说法,一律拒绝。面对听错的近音数字,先让自动机删去不合法的读法,再用待核销池删去不存在的金额,在稠密池中把误读从 0.0092 压到零。可是,读对了金额、对上了末五码,也只说明这张截图「看起来」属于这一单。截图本身只是一个声明。下一章要建立核销的底线:什么证据才能让一单变成「已付」,以及一台核销自动机怎样保证这条底线不被任何事件序列绕过。

数据口径:书中订单、账号末码、金额、截图与哈希均由固定种子合成,参数为示意,全部计算由书中程序生成;本书只讨论核销方如何检出重复或改动过的凭证,不提供任何制作凭证的方法;属非个人化经营方法,不构成法律、财务或理财建议,任何置付与发货都须由主事者人工放行。 第一章用生日问题量化撞单:从 10 单到 100 单,期望撞单配对数增长 110.0 倍,撞单概率只增长 41.3 倍;示意目录的有效金额只有 19.3 种,5.2 单即有对半机会撞单。

English version

对照数据表

下表数字由本章 Python 程序原样打印,可重新运行复现。 程序只用 Python 标准库与结论吻合的打印值: 15, 45

caseexpected yuanparsed yuanverdict
13200.003200.00ok
215000.0015000.00ok
3250.00250.00ok
41002.001002.00ok
515.0015.00ok
6110.00110.00ok
710.5010.50ok
83.523.52ok
91.051.05ok
100.500.50ok
110.080.08ok
121200.001200.00ok
1337.5037.50ok
142340.002340.00ok
1510010.0010010.00ok
1615000.0015000.00ok
173.503.50ok
181280.001280.00ok
19299.00299.00ok
20200800.00200800.00ok
21150000000.00150000000.00ok
22456.78456.78ok
23REJECTREJECTok
24REJECTREJECTok
25REJECTREJECTok
26REJECTREJECTok
27REJECTREJECTok
查看完整验算程序
# Chapter 3. p31: a finite-state transducer for spoken and written
# Chinese amounts (simplified, traditional and banker's capitals)
# Model: each character is mapped to a token class (digit, zero,
# small unit, section unit, yuan, jiao, fen, close mark, Arabic
# number); a transition table over 15 states says which class may
# follow which. Registers hold the running sums; the result is in
# fen (1/100 yuan). Colloquial cut-offs are a rule of the table: a
# digit right after a unit u that ends the integer part counts as
# digit * u/10, so the spoken forms of 3,200, 15,000 and 250 parse as
# people mean them. What the table does not license is rejected,
# never guessed. Chinese characters are written as escapes so the
# code stays ASCII; each line's comment shows them. Test phrases are
# illustrative; only their index and parsed value are printed.
from decimal import Decimal
import re

DIG = {}
for chars, v in (("\u4e00\u58f9\u5e7a", 1),          # yi, capital, yao
                 ("\u4e8c\u8d30\u8cb3\u4e24\u5169", 2),  # er, liang
                 ("\u4e09\u53c1\u53c3", 3), ("\u56db\u8086", 4),
                 ("\u4e94\u4f0d", 5), ("\u516d\u9646\u9678", 6),
                 ("\u4e03\u67d2", 7), ("\u516b\u634c", 8),
                 ("\u4e5d\u7396", 9)):
    for c in chars:
        DIG[c] = v
CLS = {c: ("D", v) for c, v in DIG.items()}
for c in "\u96f6\u3007":                              # ling, circle zero
    CLS[c] = ("Z", 0)
for chars, v in (("\u5341\u62fe", 10), ("\u767e\u4f70", 100),
                 ("\u5343\u4edf", 1000)):            # shi, bai, qian
    for c in chars:
        CLS[c] = ("U", v)
for chars, v in (("\u4e07\u842c", 10 ** 4), ("\u4ebf\u5104", 10 ** 8)):
    for c in chars:                                  # wan, yi (1e8)
        CLS[c] = ("W", v)
for c in "\u5143\u5706\u5713\u5757\u584a":           # yuan, kuai
    CLS[c] = ("Y", 0)
for c in "\u89d2\u6bdb":                              # jiao, mao
    CLS[c] = ("J", 0)
CLS["\u5206"] = ("F", 0)                             # fen
for c in "\u6574\u6b63":                              # zheng (close)
    CLS[c] = ("E", 0)
# spaces, commas, currency signs, "renminbi" and "qian" are dropped
FILLER = re.compile("[\\s,\u00a5\uffe5]|\u4eba\u6c11\u5e01|\u94b1")

# state -> token classes it licenses ("$" is the end of input)
TABLE = {
    "S": "DZUA", "D": "UWYJF$", "U": "DZWY$", "Z": "DU$", "W": "DZY$",
    "A": "UWY$", "AU": "Y$", "Y": "DZE$", "YD": "J$", "YZ": "D",
    "YZD": "F$", "J": "DE$", "JD": "F$", "F": "E$", "E": "$",
}


def tokens(text):
    text = FILLER.sub("", text)
    out, i = [], 0
    while i < len(text):
        m = re.match(r"[0-9]+(\.[0-9]+)?", text[i:])
        if m:
            out.append(("A", Decimal(m.group(0))))
            i += len(m.group(0))
        elif text[i] in CLS:
            out.append(CLS[text[i]])
            i += 1
        else:
            return None
    return out


def parse(text):
    """Return the amount in fen, or None if the table rejects it."""
    toks = tokens(text)
    if not toks:
        return None
    st, big, sec, pend, cut, last = "S", 0, 0, None, 0, 10 ** 9
    yuan, jiao, fen = None, 0, 0
    for cls, v in toks + [("$", 0)]:
        if cls == "A" and st in ("Y", "YZ", "J") and v in range(10):
            cls, v = "D", int(v)          # an Arabic jiao or fen digit
        if cls not in TABLE[st]:
            return None
        if cls in "YJF$" and yuan is None and st in "DUWZ":
            if st == "D" and cls in "JF":
                if big or sec:
                    return None           # "11 mao" is ambiguous
                yuan = 0                  # a bare "5 mao" or "8 fen"
            else:
                if st == "D":
                    sec += pend * cut if cut else pend
                yuan, pend = big + sec, None
        if cls == "D":
            if st in ("S", "Z"):
                cut = 0
            pend = v
            st = {"Y": "YD", "YZ": "YZD", "J": "JD"}.get(st, "D")
        elif cls == "Z":
            st, cut = ("YZ" if st == "Y" else "Z"), 0
        elif cls == "U" and st == "A":
            val = pend * v
            if val != int(val):
                return None
            yuan, st = int(val), "AU"
        elif cls == "U":
            if st in ("S", "Z"):
                if v != 10:
                    return None
                pend = 1                  # a leading "shi" means 10
            if v >= last:
                return None
            sec += pend * v
            pend, last, cut, st = None, v, v // 10, "U"
        elif cls == "W" and st == "A":
            val = pend * v
            if val != int(val):
                return None
            yuan, st = int(val), "AU"
        elif cls == "W":
            if st == "D":
                sec += pend
            if sec == 0 and big == 0:
                return None
            big = (big + sec) * v if v > 10 ** 4 else big + sec * v
            sec, pend, last, cut, st = 0, None, 10 ** 9, v // 10, "W"
        elif cls == "A":
            pend, st = v, "A"
        elif cls == "Y":
            if st == "A":
                yuan = pend
            st = "Y"
        elif cls == "J":
            jiao, st = pend, "J"
        elif cls == "F":
            fen, st = pend, "F"
        elif cls == "E":
            st = "E"
        elif st == "YD":
            jiao = pend
        elif st in ("JD", "YZD"):
            fen = pend
        elif st == "A":
            yuan = pend
    total = Decimal(yuan or 0) * 100 + jiao * 10 + fen
    return int(total) if total == int(total) else None


CASES = [
    ("\u4e09\u5343\u4e8c", 320000),                  # san qian er
    ("\u4e00\u4e07\u4e94", 1500000),                 # yi wan wu
    ("\u4e24\u767e\u4e94", 25000),                   # liang bai wu
    ("\u4e00\u5343\u96f6\u4e8c", 100200),            # yi qian ling er
    ("\u5341\u4e94\u5757", 1500),                    # shi wu kuai
    ("\u4e00\u767e\u4e00", 11000),                   # yi bai yi
    ("\u5341\u5757\u4e94", 1050),                    # shi kuai wu
    ("\u4e09\u5757\u4e94\u6bdb\u4e8c", 352),         # 3 kuai 5 mao 2
    ("\u4e00\u5757\u96f6\u4e94", 105),               # 1 kuai ling 5
    ("\u4e94\u6bdb", 50),                            # wu mao
    ("\u516b\u5206", 8),                             # ba fen
    ("\u58f9\u4edf\u8d30\u4f70\u5706\u6574", 120000),  # capitals
    ("\u53c1\u62fe\u67d2\u5713\u4f0d\u89d2", 3750),  # traditional
    ("\u5169\u5343\u4e09\u767e\u56db\u5341\u5143", 234000),
    ("\u4e00\u4e07\u96f6\u4e00\u5341\u5757", 1001000),
    ("1.5\u4e07", 1500000), ("3\u57575", 350),        # mixed digits
    ("\uffe51,280", 128000), ("299\u5143", 29900),
    ("\u4e8c\u5341\u4e07\u96f6\u516b\u767e", 20080000),
    ("\u4e00\u4ebf\u4e94", 15000000000),             # yi yi wu
    ("\u4eba\u6c11\u5e01\u8086\u4f70\u4f0d\u62fe\u9646"
     "\u5143\u67d2\u89d2\u634c\u5206", 45678),       # full capitals
    ("\u4e09\u5757\u4e94\u5757", None),              # two yuan marks
    ("\u4e94\u6bdb\u4e09\u6bdb", None),              # two mao marks
    ("\u767e\u4e94", None),                          # bai without digit
    ("\u4e00\u5341\u767e", None),                    # units rising
    ("\u4e09\u4e94\u767e", None),                    # "300 to 500"
    ("12.345\u5143", None),                          # beyond the fen
]
ok = 0
print("case  expected yuan    parsed yuan  verdict")
for i, (text, want) in enumerate(CASES, 1):
    got = parse(text)
    ok += got == want
    w = "REJECT" if want is None else "%.2f" % (want / 100)
    g = "REJECT" if got is None else "%.2f" % (got / 100)
    print("%4d  %13s  %13s  %s" % (i, w, g,
                                   "ok" if got == want else "FAIL"))
print("states %d, token classes 9, licensed transitions %d"
      % (len(TABLE), len("".join(TABLE.values()))))
print("cases passed: %d of %d" % (ok, len(CASES)))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《多模态汇款凭证与口语订单容错解析、转账末五码核销自动机与重复截图欺诈阻断母典》第 3 章「繁简口语金额的有限自动机」(版本 1.0.0). 5D Library. DOI 待发布

DOI 待发布 · 专著页正式发布后上线(本站暂无专著页) · 5DL-B-267 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026267,
  author    = {Fan, Yafu},
  title     = {Fault-Tolerant Parsing of Payment Receipts and Spoken Orders: Last-Five-Digit Write-Off Automata and Blocking Duplicate-Screenshot Fraud},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论