Experiment 255

“F1.4: K4's rate of beating its copies ranked among 40 shuffled stand-ins, each on 80 of its own fitting key schedules through 0253's procedure”

F1.4, second replication of 0253's excess. 0254 found K4 above all 12 copies in 13.3% of its schedules (0253: 14.3%), but four shuffled stand-ins through the identical procedure gave 10.2%, 7.2%, 1.5% and 17.5%: the rate belongs to the text and is shared across its schedules, so the per-schedule null (1 in 13) is wrong and 0253's binomial p meaningless. The pre-registered verdict was 'unclear' (pooled stand-ins 9.1%, z 2.47). Here the procedure's null is measured as a distribution over texts: 40 stand-ins and K4, 80 fitting schedules each, fresh seeds.

revertedstatus
experiment 254parents
2.3 htime, all phases
0candidates
0/24known letters
unclocked
1

The experiment method

What the method does with the 24 known letters.

The test said K4the carved letters The 97 letters cut into the copper at the CIA, shown at the top of every experiment page. beats its scrambled copiesa shuffled control The same experiment run on the carved letters shuffled into a random order. more often than chance, but four scrambled copies of K4 put through the same test ranged from far below to above K4. Here 40 of them go through it beside K4, so we can see where K4 falls among texts that certainly hide no message.

Words in the margin are defined on the glossary page.

1.1

The method at work

The carved letters go in at the top; the reading comes out at the bottom.

OBKRUOXOGHULBSOLIFBBWEASTNORTHEASTOTWTQSJQSSEKZZWATJKLUDIAWINFBBERLINCLOCKWGDKZXTJCDIGKUHUAUEKCAR41 texts (K4 and 40 shuffles of it), each on 80 of its own fitting key schedules: the share where it beats all 12 of its own fitting copieseach text's rate, lowest to highest; one cell per text, its height by the rateK4 10%, 21st of 41 from the topstand-ins: mean 10.8%, sd 6.5, range 0-29%; a fair per-schedule test would give 7.7% to every textEASTNORTHEASTBERLINCLOCKIN: THE COPPER, ALL 97 LETTERS, THE 24 KNOWN PLACES IN ACCENTOUT: NO READING; K4 IN THE MIDDLE OF ITS STAND-INS (P 0.51): 0253'S EXCESS WAS THE TEST'S
the idea

0253 counted the key schedules where K4 beat its copies against 1 in 13 per schedule; 0254 showed the rate depends on the text.

the check

40 scrambled copies of K4 put through the identical test, 80 schedules each, beside K4.

the bound

K4 10%, stand-ins 11% on average with a spread of 6 points; 20 of 40 at or above K4. F1.4's residue closes at its measured power.

2

Experiment lineage

What this experiment was built on, from the lineage page: 20 experiments before it, back to the root of its line. Its number is in bold.

kept reverted, in its family's colour ×error or timeout reverted under today's rules built on a parent in the same family built on a parent in another family
3

Best candidate against the known letters

This experiment produced no candidate: the search ruled everything out before a full text was built. Its product is the bounded negative under Result and bounds below.

4

Walkthrough, by hand

Redo the method on a few letters with pencil and paper.

  1. Shuffle K4's letters 40 times to make 40 stand-ins.
  2. For each, find the key-stepping rules it fits with some alphabet, and pick 80 at random.
  3. For each rule, search for the stand-in's most English-looking reading, then search scrambled copies of it that fit the same rule, one at a time, until one reads as well (at most 12).
  4. Each stand-in's share of rules where it beat all 12 is its rate. Do the same for K4.
  5. Where does K4's rate fall among the 40?
the worksheet: the first steps by hand, on squared paper
ratestand-insK40-5%55-10%1510-15%8K415-20%820-30%480 schedules per text, drawn from its own fitting ones; 0253's decide() unchanged. K4 10.0%, 20 of 40 stand-ins at orabove (p 0.51); beating the first copy, 7 of 40 at or above (p 0.20). The procedure's null is a spread over texts, not 1in 13 per schedule.THE SHARE OF SCHEDULES WHERE A TEXT BEATS ALL 12 OF ITS OWN FITTING COPIES, K4 AMONG 40 STAND-INS
5

Worked example, generated

The same steps run by the code on the real letters.

K4 among 40 stand-ins: where does its rate fall?

  target       fitting schedules   80 drawn   above all 12 copies
  K4           2,093               80         ?   (0253: 14.3%, 0254: 13.3%)
  stand-in 1   its own             80         ?
  ... 40 stand-ins (shuffles of K4), the same procedure

  K4's rank among the 41 texts is the verdict.
6

Hypothesis, for a cryptanalyst

Family, key space, the constraint under test, and the clues it reads.

F1.4 (one free alphabet, hand key schedules; C20(d), C27(c)), decided by 0253's decide() (one tempering run on the target, fitting shuffles climbed one at a time, at most 12). The quantity is each text's rate of beating all 12 copies over its own fitting schedules. If K4 carries a message under some schedule, its rate is lifted only near that schedule; if the rate reflects how a text's letters sit at the 24 known places (shared by all schedules), stand-ins vary as much as K4. K4's rank among 41 texts gives p to 1/41.

7

Prediction

Written before the run; the result decides it.

K4's rate inside the stand-ins' spread: rank between 5th and 30th of 41 (one-sided p above 0.1); the stand-ins' rates spread with a standard deviation of about 5 points around 9%. Zero candidates.

8

Method

  1. Targets: K4 and 40 shuffles of K4 (seed 2551). Each target's cell: every schedule of 0252's list it fits exactly.
  2. 80 schedules drawn at random from each cell; per schedule 0253's decide() without escalation.
  3. Each target's rate (above all 12) and its rate of beating the first copy; K4's rank among the 41 on each, one-sided p = (1 + stand-ins at or above K4) / 41. A lead only if p is at most 1/41 on the rate.
9

Result and bounds

The harness's record: candidates, gates, loss, and the negative result this experiment leaves behind.

0candidates returned
0passed every gate
0/24known letters, best
–lowest loss
nonechampion before

K4 among 40 stand-ins: K4 rate 0.100, stand-ins mean 0.108 (sd 0.065, range 0.000-0.287); 20 of 40 at or above K4, p 0.512; beat-first rate p 0.195. K4 inside its stand-ins.

What was searched, as the harness recorded it.

item
F1.4, K4 among 40 stand-ins (second replication of 0253's excess)
decider
0253's decide(), imported unchanged: one tempering run on the target, fitting shuffles climbed one at a time, at most 12
targets
k4
cell size
2,093
sampled
80
above all 12
8
rate
0.1
beat first copy
52
beat first rate
0.65
mean copies climbed
4.06
standin0
cell size
4,586
sampled
80
above all 12
7
rate
0.087
beat first copy
43
beat first rate
0.537
mean copies climbed
3.33
standin1
cell size
4,488
sampled
80
above all 12
18
rate
0.225
beat first copy
56
beat first rate
0.7
mean copies climbed
5.53
standin2
cell size
2,240
sampled
80
above all 12
2
rate
0.025
beat first copy
42
beat first rate
0.525
mean copies climbed
3
standin3
cell size
2,810
sampled
80
above all 12
12
rate
0.15
beat first copy
42
beat first rate
0.525
mean copies climbed
3.55
standin4
cell size
2,079
sampled
80
above all 12
7
rate
0.087
beat first copy
43
beat first rate
0.537
mean copies climbed
3.31
standin5
cell size
7,214
sampled
80
above all 12
0
rate
0
beat first copy
30
beat first rate
0.375
mean copies climbed
1.95
standin6
cell size
1,994
sampled
80
above all 12
12
rate
0.15
beat first copy
40
beat first rate
0.5
mean copies climbed
3.55
standin7
cell size
4,140
sampled
80
above all 12
3
rate
0.037
beat first copy
43
beat first rate
0.537
mean copies climbed
3.02
standin8
cell size
3,933
sampled
80
above all 12
8
rate
0.1
beat first copy
37
beat first rate
0.463
mean copies climbed
3.25
standin9
cell size
3,583
sampled
80
above all 12
2
rate
0.025
beat first copy
47
beat first rate
0.588
mean copies climbed
2.92
standin10
cell size
1,784
sampled
80
above all 12
19
rate
0.237
beat first copy
49
beat first rate
0.613
mean copies climbed
4.83
standin11
cell size
2,767
sampled
80
above all 12
7
rate
0.087
beat first copy
48
beat first rate
0.6
mean copies climbed
3.24
standin12
cell size
3,496
sampled
80
above all 12
5
rate
0.062
beat first copy
41
beat first rate
0.512
mean copies climbed
3.09
standin13
cell size
2,001
sampled
80
above all 12
5
rate
0.062
beat first copy
34
beat first rate
0.425
mean copies climbed
2.59
standin14
cell size
4,788
sampled
80
above all 12
5
rate
0.062
beat first copy
46
beat first rate
0.575
mean copies climbed
3.38
standin15
cell size
1,889
sampled
80
above all 12
4
rate
0.05
beat first copy
47
beat first rate
0.588
mean copies climbed
2.99
standin16
cell size
3,847
sampled
80
above all 12
13
rate
0.163
beat first copy
60
beat first rate
0.75
mean copies climbed
4.94
standin17
cell size
2,416
sampled
80
above all 12
11
rate
0.138
beat first copy
45
beat first rate
0.562
mean copies climbed
4.24
standin18
cell size
3,776
sampled
80
above all 12
6
rate
0.075
beat first copy
39
beat first rate
0.487
mean copies climbed
2.83
standin19
cell size
5,570
sampled
80
above all 12
9
rate
0.113
beat first copy
42
beat first rate
0.525
mean copies climbed
3.41
standin20
cell size
2,139
sampled
80
above all 12
23
rate
0.287
beat first copy
57
beat first rate
0.713
mean copies climbed
5.69
standin21
cell size
3,760
sampled
80
above all 12
6
rate
0.075
beat first copy
41
beat first rate
0.512
mean copies climbed
3.15
standin22
cell size
2,382
sampled
80
above all 12
17
rate
0.212
beat first copy
54
beat first rate
0.675
mean copies climbed
4.74
standin23
cell size
2,299
sampled
80
above all 12
13
rate
0.163
beat first copy
55
beat first rate
0.688
mean copies climbed
4.17
standin24
cell size
4,339
sampled
80
above all 12
10
rate
0.125
beat first copy
47
beat first rate
0.588
mean copies climbed
3.81
standin25
cell size
4,794
sampled
80
above all 12
5
rate
0.062
beat first copy
39
beat first rate
0.487
mean copies climbed
2.75
standin26
cell size
2,830
sampled
80
above all 12
12
rate
0.15
beat first copy
51
beat first rate
0.637
mean copies climbed
4.41
standin27
cell size
4,660
sampled
80
above all 12
9
rate
0.113
beat first copy
51
beat first rate
0.637
mean copies climbed
4
standin28
cell size
4,039
sampled
80
above all 12
5
rate
0.062
beat first copy
45
beat first rate
0.562
mean copies climbed
3.16
standin29
cell size
1,670
sampled
80
above all 12
7
rate
0.087
beat first copy
42
beat first rate
0.525
mean copies climbed
3.1
standin30
cell size
1,811
sampled
80
above all 12
15
rate
0.188
beat first copy
54
beat first rate
0.675
mean copies climbed
5.17
standin31
cell size
3,301
sampled
80
above all 12
8
rate
0.1
beat first copy
42
beat first rate
0.525
mean copies climbed
3.44
standin32
cell size
2,481
sampled
80
above all 12
13
rate
0.163
beat first copy
55
beat first rate
0.688
mean copies climbed
4.69
standin33
cell size
5,040
sampled
80
above all 12
2
rate
0.025
beat first copy
47
beat first rate
0.588
mean copies climbed
2.88
standin34
cell size
3,779
sampled
80
above all 12
10
rate
0.125
beat first copy
46
beat first rate
0.575
mean copies climbed
3.84
standin35
cell size
2,594
sampled
80
above all 12
7
rate
0.087
beat first copy
33
beat first rate
0.412
mean copies climbed
2.79
standin36
cell size
2,542
sampled
80
above all 12
14
rate
0.175
beat first copy
47
beat first rate
0.588
mean copies climbed
4.49
standin37
cell size
4,757
sampled
80
above all 12
8
rate
0.1
beat first copy
42
beat first rate
0.525
mean copies climbed
3.1
standin38
cell size
6,172
sampled
80
above all 12
4
rate
0.05
beat first copy
44
beat first rate
0.55
mean copies climbed
2.94
standin39
cell size
4,640
sampled
80
above all 12
4
rate
0.05
beat first copy
40
beat first rate
0.5
mean copies climbed
2.84
schedules per target
80
fair rate 12
0.077
k4 rate
0.1
standins rate mean
0.108
standins rate sd
0.065
standins rate min max
0, 0.287
standins at or above k4 rate
20
p rate
0.512
k4 beat first rate
0.65
standins at or above k4 beat first
7
p beat first
0.195
lead
no
core seconds
250,870
compute seconds
8,139
exhaustive
no
modal cost estimate
41 targets x 80 schedules x about 4 climbs x about 17 s: about 223,000 core-seconds, about $3.5 of Modal credit (credit left about $15 after the margin)

How this experiment sits among the others is on the lineage page; its family's open hypotheses are on the families page.

Rendered from the record on 2026-11-05 17:14 UTC; updated after every experiment.
view on GitHub