关于「试投淘汰:先试哪几把刀,什么时候换」,《存量资产七日点火与火力收敛》给出的定量结论是什么?

结论:第五章说明七天窗口里第一天后全押只有 0.550 的机会押对。

数据口径:书中参数为示意,价格用无币种示意单位;全部计算由书中程序生成;本书是非个人化经营建议,不承诺任何收入。 第二章定义火力收敛系数,示意书目里只推前三件仅得最优的 0.805。

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: 0.550

A. buyers in 7 days (mean of 200 runs); oracle = 147.0
policymeanp10p90best_last_daybest_weekP(top=best)
uniform83.272940.1200.1200.000
commit118.7861470.5490.4540.550
commit2115.9931340.6310.4430.640
thompson107.7901240.4920.3450.715
查看完整验算程序
# Chapter 5. p51: trial-firing eight candidate knives for seven days
# Eight candidates with true conversion rates from 0.010 to 0.035
# (illustrative), 600 visitors a day for 7 days, traffic assigned in
# batches of 50. Policies: uniform split; explore day 1 (or days 1-2) then
# commit to the leader; batch Thompson sampling (Beta(1,1) priors, traffic in
# proportion to how often each arm wins 20 posterior draws). 200 runs.
import math
import random
import statistics

rng = random.Random(41551)
P = (0.010, 0.012, 0.015, 0.018, 0.020, 0.024, 0.028, 0.035)
DAY, DAYS, BATCH, RUNS = 600, 7, 50, 200
NB = DAY * DAYS // BATCH


def serve(arm, n):
    k = 0
    for _ in range(n):
        if rng.random() < P[arm]:
            k += 1
    return k


def split(weights, n):
    # integer weights -> integer traffic, largest remainders first
    tot = int(math.fsum(weights))
    raw = [n * v / tot for v in weights]
    out = [int(r) for r in raw]
    left = n - int(math.fsum(out))
    order = sorted(range(len(weights)),
                   key=lambda i: (-(raw[i] - out[i]), i))
    for i in order[:left]:
        out[i] += 1
    return out


def run(policy):
    succ = [0] * 8
    tri = [0] * 8
    last_day = [0] * 8
    buyers = 0
    for b in range(NB):
        explore = NB // DAYS * (2 if policy == "commit2" else 1)
        if policy == "uniform" or (policy.startswith("commit")
                                   and b < explore):
            alloc = split([1] * 8, BATCH)
        elif policy.startswith("commit"):
            lead = max(range(8), key=lambda i: (succ[i] / max(1, tri[i]), -i))
            alloc = [BATCH if i == lead else 0 for i in range(8)]
        else:
            wins = [0] * 8
            for _ in range(20):
                draws = [rng.betavariate(1 + succ[i], 1 + tri[i] - succ[i])
                         for i in range(8)]
                wins[max(range(8), key=lambda i: draws[i])] += 1
            alloc = split(wins, BATCH)
        for i in range(8):
            if alloc[i]:
                k = serve(i, alloc[i])
                succ[i] += k
                tri[i] += alloc[i]
                buyers += k
                if b >= NB - NB // DAYS:
                    last_day[i] += alloc[i]
    top = max(range(8), key=lambda i: (last_day[i], i))
    return (buyers, last_day[7] / float(DAY), tri[7] / float(DAY * DAYS),
            1.0 if top == 7 else 0.0)


oracle = DAY * DAYS * P[7]
print("A. buyers in 7 days (mean of {} runs); oracle = {:.1f}".format(
    RUNS, oracle))
print("policy    mean   p10  p90  best_last_day  best_week  P(top=best)")
for pol in ("uniform", "commit", "commit2", "thompson"):
    res = [run(pol) for _ in range(RUNS)]
    b = sorted(r[0] for r in res)
    print("{:<8} {:6.1f}  {:4d} {:4d}  {:13.3f}  {:9.3f}  {:11.3f}".format(
        pol, statistics.fmean(b), b[RUNS // 10], b[RUNS * 9 // 10],
        statistics.fmean(r[1] for r in res),
        statistics.fmean(r[2] for r in res),
        statistics.fmean(r[3] for r in res)))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《重装数字资产库首周现金流点火、火力收敛系数与饱和变现突击母典》第 5 章「试投淘汰:先试哪几把刀,什么时候换」(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-415 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026415,
  author    = {Fan, Yafu},
  title     = {Seven-Day Cashflow Ignition from Large Digital Asset Catalogs: Firepower Concentration Ratios and Saturation Monetisation Sprints},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论