What does Beyond Distillation conclude, in numbers, about “The Ceiling of Distillation: Compression Keeps, Synthesis Generates”?
Answer: Now read the table. With all twelve rules kept, the description is 522 bytes long and nothing in the target market goes unhandled. Cut to ten rules, the description shrinks to 415 bytes, a ratio of 0.795, and home-market coverage is still 0.9439, which looks almost lossless. Yet in the target market 0.2266 of cases now have no rule to follow, and 0.6045 of the hard-limit cases are missed. Cut to eight rules, home-market coverage still reads 0.8758, while the share of hard-limit cases missed reaches 1.0000: every one of them is lost. Cut to two rules, the description is down to 75 bytes, a ratio of 0.144, and 0.6673 of target-market cases are handled by nobody.
Scope of the numbers: Parameters are illustrative and all calculations are generated by the book's programs. This is general business analysis, not investment, legal or medical advice, and the code it generates is teaching material deployed to no site.
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: 0.144, 0.2266, 0.6045, 0.6673, 0.795, 0.8758
| source share of hard rules 0.0920 | |||||
| target share of hard rules 0.3780 | |||||
| k | bytes | ratio | src-cover | tgt-miss | hard-miss |
| 12 | 522 | 1.000 | 1.0000 | 0.0000 | 0.0000 |
| 10 | 415 | 0.795 | 0.9439 | 0.2266 | 0.6045 |
| 8 | 318 | 0.609 | 0.8758 | 0.3958 | 1.0000 |
| 6 | 241 | 0.462 | 0.7895 | 0.4546 | 1.0000 |
| 4 | 156 | 0.299 | 0.6713 | 0.5366 | 1.0000 |
| 2 | 75 | 0.144 | 0.4834 | 0.6673 | 1.0000 |
Show the full program
# Chapter 1. p12: distilling harder drops the rare rules that carry limits.
# Model: a mother book yields 12 operating rules. In the source market the
# rule frequencies follow a Zipf law (rank r has weight 1 / r). Three rare
# rules (ranks 9, 11, 12) are hard limits: a compliance red line, a margin
# floor in a second unit, a tap-target gate. "Distilling harder" keeps
# only the top-k rules by frequency. In the target market the hard limits
# fire far more often (weight x6). We draw 20,000 target cases (seeded)
# and count cases the kept rules cannot handle, and hard-limit misses.
# Rule sizes in bytes and all weights are illustrative.
import math
import random
SIZES = [34, 41, 29, 52, 38, 47, 33, 44, 61, 36, 58, 49]
HARD = {9, 11, 12}
def weights(boost):
w = []
for r in range(1, 13):
x = 1.0 / r
if r in HARD:
x *= boost
w.append(x)
t = math.fsum(w)
return [x / t for x in w]
if __name__ == "__main__":
rng = random.Random(378012)
src = weights(1.0)
tgt = weights(6.0)
cases = rng.choices(range(1, 13), weights=tgt, k=20000)
n_hard = 0
for c in cases:
if c in HARD:
n_hard += 1
print("source share of hard rules {:.4f}".format(
math.fsum(src[r - 1] for r in sorted(HARD))))
print("target share of hard rules {:.4f}".format(
math.fsum(tgt[r - 1] for r in sorted(HARD))))
print(" k bytes ratio src-cover tgt-miss hard-miss")
total_bytes = math.fsum(SIZES)
for k in (12, 10, 8, 6, 4, 2):
kept = set(range(1, k + 1))
nbytes = math.fsum(SIZES[:k])
cover = math.fsum(src[r - 1] for r in sorted(kept))
miss = 0
hard_miss = 0
for c in cases:
if c not in kept:
miss += 1
if c in HARD:
hard_miss += 1
print("{:>2} {:>5.0f} {:>5.3f} {:>9.4f} {:>8.4f} {:>9.4f}"
.format(k, nbytes, nbytes / total_bytes, cover,
miss / len(cases),
hard_miss / n_hard if n_hard else 0.0))Single source and citation
Fan, Y. (2026). Beyond Distillation: Executable Business-Organ Synthesis, Generative Code Crystallisation and Living-System Bootstrapping (Version 1.0.0), Chapter 1: The Ceiling of Distillation: Compression Keeps, Synthesis Generates. 5D Library. DOI pending
DOI pending · monograph page goes live after publication · 5DL-B-378 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026378,
author = {Fan, Yafu},
title = {Beyond Distillation: Executable Business-Organ Synthesis, Generative Code Crystallisation and Living-System Bootstrapping},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}Other chapter answers in this book
- Chapter 2 · The Pricing Engine: Constraint Solving, Invariants and Property Tests
- Chapter 3 · The Front-End Statechart: Reachability, Deadlock and an Interface Gate on Every Path
- Chapter 4 · The Objection Tree: Coverage, Depth and the Order of Repairs
- Chapter 5 · The Live Probe: CUSUM, False-Alarm Run Lengths and Writing Back into the Mother Book
- Chapter 6 · Crystallinity and Bootstrapping: Mutation Testing, a Composite Index and the Limits of Self-Fingerprinting
- Chapter 7 · Classical Sources East and West: the Kaogongji's Four Conditions, the Laozi's Generative Sequence and Aristotle's Four Causes
- Chapter 8 · The Empirical Loop: Field Tools and Checker Specifications