关于「通关题库:难度与区分度、围绕通关线的测验信息与测试效应的反直觉」,《专著改编的信息保留代数》给出的定量结论是什么?

结论:第四章给题目做体检、围绕通关线组卷,并提出核心反常识命题:测试效应使不给反馈的测验组一周后仍以 0.3004 领先反复学习组的 0.0253。

数据口径:书中参数为示意;全部计算由书中程序生成;本书不是法律意见,也不是个人化的经营或投资建议。 第六章在 40 个制作工时下求出播客、投影片与题库的最优组合,价值 2.0685,不做题库只有 1.4164;事实核对应放在上游;每周 2 次提醒使期望学习次数最高,为 6.2792,过量提醒有害,对脆弱学习者阈值更低;只改编有权改编的书,合成内容必须标识。

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: 0.0253, 0.3004

planafter 5 minutesafter 2 daysafter 7 days
SSSS0.95420.31700.0253
SSST0.90580.49300.1220
STTT0.56680.43280.3004
STTT overtakes SSSS after 1.36 days
units ever encoded: SSSS 0.9375, STTT 0.5000
查看完整验算程序
# Chapter 4. p43: restudy against retrieval practice, a two-trace model.
# Four short sessions on one passage: SSSS (study four times), SSST and
# STTT (S = study, T = free-recall test without feedback). Each idea unit
# is either encoded or not. A study session encodes a missing unit with
# probability 0.5 (stability 1 day) or multiplies an encoded unit's
# stability by 1.3; every study also lifts a short-lived familiarity
# m = 1 - 0.55^(study sessions) that decays with time constant 0.4 day.
# A test retrieves an encoded unit with probability exp(-gap / S) and a
# success multiplies S by 2.4; a missed unit stays missed (no feedback).
# Recall at delay t: encoded 1 - (1 - exp(-t/S))(1 - m exp(-t/0.4)),
# missed 0.3 m exp(-t/0.4). Exact expectation over all branches.
# Parameters illustrative; the design echoes the conditions of Roediger
# and Karpicke (2006), the numbers do not reproduce theirs.
import math

P_ENC, STUDY_X, TEST_X, TAU_M, GAP = 0.5, 1.3, 2.4, 0.4, 0.004


def states(plan):
    """List of (probability, encoded, stability) after the sessions."""
    cur = [(1.0, False, 0.0)]
    for step in plan:
        nxt = []
        for p, enc, s in cur:
            if step == "S":
                if enc:
                    nxt.append((p, True, s * STUDY_X))
                else:
                    nxt.append((p * P_ENC, True, 1.0))
                    nxt.append((p * (1.0 - P_ENC), False, 0.0))
            elif enc:
                r = math.exp(-GAP / s)
                nxt.append((p * r, True, s * TEST_X))
                nxt.append((p * (1.0 - r), True, s))
            else:
                nxt.append((p, False, 0.0))
        cur = nxt
    return cur


def recall(plan, t):
    m = (1.0 - 0.55 ** plan.count("S")) * math.exp(-t / TAU_M)
    terms = []
    for p, enc, s in states(plan):
        if enc:
            terms.append(p * (1.0 - (1.0 - math.exp(-t / s)) * (1.0 - m)))
        else:
            terms.append(p * 0.3 * m)
    return math.fsum(terms)


print("plan  after 5 minutes  after 2 days  after 7 days")
for plan in ("SSSS", "SSST", "STTT"):
    print("{:4s}  {:15.4f}  {:12.4f}  {:12.4f}".format(
        plan, recall(plan, 5.0 / 1440.0), recall(plan, 2.0),
        recall(plan, 7.0)))
lo, hi = 5.0 / 1440.0, 7.0
for _ in range(60):
    mid = (lo + hi) / 2.0
    if recall("STTT", mid) > recall("SSSS", mid):
        hi = mid
    else:
        lo = mid
print("STTT overtakes SSSS after {:.2f} days".format(hi))
print("units ever encoded: SSSS {:.4f}, STTT {:.4f}".format(
    math.fsum(p for p, e, _ in states("SSSS") if e),
    math.fsum(p for p, e, _ in states("STTT") if e)))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《一书多吃内容裂变代数:从硬核专著自动编译双人播客、导学投影片与通关题库》第 4 章「通关题库:难度与区分度、围绕通关线的测验信息与测试效应的反直觉」(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-151 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026151,
  author    = {Fan, Yafu},
  title     = {Content Fission Algebra: Compiling Hard-Core Monographs into Two-Host Podcasts, Study Slides and Quiz Banks},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论