What does Distilling Books into Skill Packs conclude, in numbers, about “Governance: Least Privilege, Machine-Checked Red Lines, Disclaimers, Provenance and Rule Retirement”?

Answer: R04, the reply template for price objections, fired 96 times and was overruled 31 times; its interval runs from 0.2378 to 0.4217 and the lower bound is above 0.20, so the template is probably wrong and is retired for a rewrite. R05 fired 18 times with 3 overrides; the rate of 0.1667 looks modest, but the upper bound reaches 0.3922, so it cannot be cleared yet and goes to review. R07 fired only 12 times with 1 override but drew a red-line complaint, and is suspended at once. R03 and R11 do not have traced anchors and are suspended. R08, the guard on stock lines, did not fire once in ninety days and is kept. S13, a birthday-offer rule from an earlier version, did not fire in ninety days and is retired to the archive.

Scope of the numbers: All samples are invented; pricing and closing rules are written in general terms and are not legal or compliance advice.

中文版

Data table

The numbers below are printed verbatim by the chapter's Python program and can be reproduced by running it. standard-library Python onlyprinted values that match the answer: 0.1667, 0.20, 0.2378, 0.3922, 0.4217, 03

rule | fires | overrides | rate| Wilson 95%| decision
R01|140 |2 | 0.0143 | 0.0039 to 0.0506 | KEEP
R02|410 |9 | 0.0220 | 0.0116 to 0.0412 | KEEP
R03|1260 |41 | 0.0325 | 0.0241 to 0.0438 | SUSPEND (anchor not traced)
R04|96 |31 | 0.3229 | 0.2378 to 0.4217 | RETIRE (usually overruled)
R05|18 |3 | 0.1667 | 0.0584 to 0.3922 | REVIEW (bound crosses 0.20)
R06|55 |4 | 0.0727 | 0.0286 to 0.1726 | KEEP
R07|12 |1 | 0.0833 | 0.0149 to 0.3539 | SUSPEND (red-line complaint)
R08|0 |0 | 0.0000 | 0.0000 to 1.0000 | KEEP (silent guard)
R10|230 |12 | 0.0522 | 0.0301 to 0.0890 | KEEP
R11|64 |1 | 0.0156 | 0.0028 to 0.0833 | SUSPEND (anchor not traced)
S13|0 |0 | 0.0000 | 0.0000 to 1.0000 | RETIRE (dead in 90 days)
zero overrides in 30 fires still leaves an upper bound of 0.1135
fires needed with zero overrides before the upper bound drops below 0.20: 16
Show the full program
# Chapter 7. p73: When to retire a rule: Wilson intervals on the human-override rate over 90 days.
# For each rule: times it fired, times a reviewer overrode it, red-line complaints, whether its source
# anchor is still TRACED (p72), and whether it is a guard (a rule that blocks or refuses rather than sells).
# Wilson 95% interval (z = 1.96) on overrides / fires.
# Decision, first match wins:
#   SUSPEND  any red-line complaint (threat, invented scarcity or deadline, promised outcome)
#   SUSPEND  source anchor not TRACED
#   RETIRE   lower bound > 0.20 (the rule is usually overruled)
#   KEEP     a guard that never fired (silence is what a guard is for)
#   RETIRE   a selling rule that fired 0 times in 90 days (dead; archive it with its golden cases)
#   REVIEW   upper bound > 0.20 (cannot yet rule out a bad rule; collect more or inspect cases)
#   KEEP     otherwise
import math

Z, BAR = 1.96, 0.20


def wilson(k, n):
    if n == 0:
        return 0.0, 1.0
    p = k / n
    den = 1 + Z * Z / n
    mid = (p + Z * Z / (2 * n)) / den
    half = Z * math.sqrt(p * (1 - p) / n + Z * Z / (4 * n * n)) / den
    return max(0.0, mid - half), min(1.0, mid + half)


STATS = [   # rule, fires, overrides, red-line complaints, anchor traced, guard
    ("R01", 140, 2, 0, True, True),
    ("R02", 410, 9, 0, True, True),
    ("R03", 1260, 41, 0, False, False),
    ("R04", 96, 31, 0, True, False),
    ("R05", 18, 3, 0, True, False),
    ("R06", 55, 4, 0, True, False),
    ("R07", 12, 1, 1, True, False),
    ("R08", 0, 0, 0, True, True),
    ("R10", 230, 12, 0, True, False),
    ("R11", 64, 1, 0, False, True),
    ("S13", 0, 0, 0, True, False),
]


def decide(fires, over, complaints, traced, guard):
    lo, hi = wilson(over, fires)
    if complaints:
        return "SUSPEND (red-line complaint)", lo, hi
    if not traced:
        return "SUSPEND (anchor not traced)", lo, hi
    if fires and lo > BAR:
        return "RETIRE (usually overruled)", lo, hi
    if fires == 0:
        return ("KEEP (silent guard)", lo, hi) if guard else ("RETIRE (dead in 90 days)", lo, hi)
    if hi > BAR:
        return "REVIEW (bound crosses 0.20)", lo, hi
    return "KEEP", lo, hi


if __name__ == "__main__":
    print("rule | fires | overrides | rate   | Wilson 95%        | decision")
    for rid, n, k, c, t, g in STATS:
        d, lo, hi = decide(n, k, c, t, g)
        rate = k / n if n else 0.0
        print("{:4} | {:5d} | {:9d} | {:.4f} | {:.4f} to {:.4f} | {}".format(rid, n, k, rate, lo, hi, d))
    lo, hi = wilson(0, 30)
    print("zero overrides in 30 fires still leaves an upper bound of {:.4f}".format(hi))
    n = 1
    while wilson(0, n)[1] > BAR:
        n += 1
    print("fires needed with zero overrides before the upper bound drops below 0.20: {}".format(n))

Single source and citation

Fan, Y. (2026). Lossless Distillation of Monographs into Agent Skill Packs, Executable Rule Compilation and Self-Evolving Systems (Version 0.1.0), Chapter 7: Governance: Least Privilege, Machine-Checked Red Lines, Disclaimers, Provenance and Rule Retirement. 5D Library. DOI pending

DOI pending · monograph page goes live after publication · 5DL-B-150 · v0.1.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026150,
  author    = {Fan, Yafu},
  title     = {Lossless Distillation of Monographs into Agent Skill Packs, Executable Rule Compilation and Self-Evolving Systems},
  year      = {2026},
  publisher = {5D Library},
  edition   = {0.1.0},
  note      = {DOI pending}
}

Other chapter answers in this book