“A13.3, replication: 0034's saved readings read backwards, the one candidate 0216 left unresolved, against more fresh copies of the same search, with reversed messages planted through 0034's own climb”
A13.3 (Stage 2, step 3, issue #75), the replication 0216 began and its clock cut short. 0216 found K4 above every copy in 2 of 112 cells; 0034/eng_keep_nth met a fresh copy at copy 6, but 0034/eng_reverse (K4 -5.458, the best of its 18 readings read backwards, from period 71) stayed above all 8 fresh copies drawn before the clock (and all 10 of pass 1). Here the same pre-registered sequence continues: fresh copies 8 onward of seed '3113', each a plain shuffle run through 0034's frozen sweep unchanged (24 conventions, periods 27-29 and 53-95 every third, 20 restarts), eight at a time, until one copy's best backwards score is at or above K4's (not a flag), or the clock (unresolved, with its p bound). New: reversed messages planted under 0034's own method at period 71 and climbed by its own climb, to say whether this cell can see a reversal at all.
An old search (0034) produced 18 readings of K4the carved letters The 97 letters cut into the copper at the CIA, shown at the top of every experiment page. with every known letter in place and threw them away as gibberish. One of them reads a little better backwardsa mirror Reading or writing the text from the end to the start. than any reading the same search made from scrambled copiesa shuffled control The same experiment run on the carved letters shuffled into a random order. of the carving, which could mean Sanborn wrote his message backwards before enciphering it, or could be luck. Luck is tested the only honest way: run the same search on more scrambled copies until one of them reads as well backwards. We also hide some backwards messages of our own and see whether this search and test would notice them.
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.
the idea
0216 left one comparison open: K4's period-71 reading of 0034, written backwards, read a little better than any copy's. If Sanborn reversed his message before enciphering it, that is where it would show.
the check
The same pre-registered run of scrambled copies continued through 0034's own search until one copy read as well backwards; four backwards messages planted through the same search.
the bound
Fresh copy 14 reached -5.419 against K4's -5.458: not a flag, A13.3 ends with none. The plants were missed 4 of 4 times though each message itself would have been found: the forward climb overwrites a backwards message, so this test is weak for masks.
2
Experiment lineage
What this experiment was built on, from the lineage page: 14 experiments before it, back to the root of its line. Its number is in bold.
keptreverted, in its family's colour×error or timeoutreverted under today's rulesbuilt on a parent in the same familybuilt on a parent in another family
3
Best candidate against the known letters
Laid out as the copper carries K4: OBKR ends the line that finishes K3, then three lines of 31. Left, as carved, the 24 known places in accent. Right, the candidate: a known place it gets right is an accent cell, one it misses is outlined. This is the candidate with most known letters right; none passed every gate.
in: the copper, always the same
as carved, 97 letters; the 24 known places in accent
ECDMRIPFEIMEHNLSSTTRTVDOHW?OBKR
UOXOGHULBSOLIFBBWFLRVQQPRNGKSSO
TWTQSJQSSEKZZWATJKLUDIAWINFBNYP
VTTMZFPKWGDKZXTJCDIGKUHUAUEKCAR
OBKR ends the line that finishes K3, then three lines of 31
out: the candidate
every place read; known places in accent, misses outlined
ECDMRIPFEIMEHNLSSTTRTVDOHW?HUKS
ALLTHAVICEFFICESAEASTNORTHEASTN
ATIONANDINATESTINANDATTHERESBER
LINCLOCKENSTIONESERTERETOODSLCQ
checked: the known letters in place, and the reading turns back into the copper
4
How the score is made
The loss of the best candidate, part by part. Zero is solved.
Known letters missed
0 of 24, ten points each
0.00
English
the 73 free letters score -4.01; English text scores -3.6
0.41
Key
348 bits to write it down (key words are cheap, arbitrary letters 4.7 bits each); each 60 beyond 60 costs a point
4.80
Loss
of 240, the worst possible
5.21 2.2%
5
Walkthrough, by hand
Redo the method on a few letters with pencil and paper.
Take 0034's reading at period 71 with all 24 known letters in place.
Write it backwards and score the backwards text for English: -5.458.
Scramble the carved letters, run 0034's search on the scramble, and do the same with each of its 18 readings; keep its best backwards score.
If that score reaches -5.458, K4's lead was luck: stop.
If not, scramble again and repeat; 2,242 scrambles in a row below K4 would be a real signal.
Separately, write a known message backwards, encipher it the 0034 way, run the search, and see whether the backwards score finds it.
the worksheet: the first steps by hand, on squared paper
6
Worked example, generated
The same steps run by the code on the real letters.
0034's reading, period 71 : HUKSALLTHAVICEFFICESAEASTNORTHEASTNATIONANDINATESTINANDATTHERESBERLINCLOCKENSTIONESERTERETOODSLCQ
written backwards: QCLSDOOTERETRESENOITSNEKCOLCNILREBSEREHTTADNANITSETANIDNANOITANTSAEHTRONTSAEASECIFFECIVAHTLLASKUH
English score of the backwards text: -5.458
a fresh scrambled copy through the same search counts against it when its best backwards score is at or above this
7
Hypothesis, for a cryptanalyst
Family, key space, the constraint under test, and the clues it reads.
A13 (knowledge/12-assumptions.md), C4(a) (Scheidt 2005, 2015: 'I masked the English language'; a plaintext transform before the keyed stage), reading: the mask is a reversal of the message, the simplest move-only mask (0023 closed it exactly for short periodic keys; here the long periods 53-95 where the known letters leave 47-71 free key slots). Cipher: one periodic tableau key on the standard or KRYPTOS alphabets, 24 conventions, periods 27-29 and 53-95 every third (0034's space), free slots completed by a forward quadgram climb. Statistic: eng_reverse (tools/a13_structure.py), the quadgram score of the reading written backwards, best over a text's 18 readings. The cell was one of 112 pre-registered in reports/a13-structure-check.md (flag at p < 0.05/112, all of 2,242 fresh copies below K4). K4's two best backwards readings (periods 71 and 77, -5.458 and -5.525) both exceed every copy's best so far (-5.553), which is why it needs fresh copies, not argument. Plants measure power: a forward climb pulls the free letters toward forward English, so a reversed message may be destroyed by the very search that should reveal it.
8
Prediction
Written before the run; the result decides it.
A fresh copy reaches K4's backwards score within the 40 or so copies the clock allows (K4 has led 18 copies, which luck alone does about one time in 19), so the cell is not a flag and A13.3 closes with no flag. The reversed plants are found 0 or 1 times in 4: the forward climb overwrites the free letters, so the backwards test on this search has little power, and a 'nothing found' here says little about reversal. Zero candidates pass gate 4.
9
Method
K4: 0034's 18 saved readings (24 of 24 known letters), each re-enciphered by 0034's frozen encrypt; the best backwards score.
Reversed plants: four messages from Carter's 1923 book, written backwards, the 24 known letters set after the reversal, enciphered under 0034's method with a random convention and a random period-71 key; each climbed by 0034's frozen sweep at period 71 only (a slice: one period of its 18, which can only lower the plant's best); found when its best backwards score is above every 0034 copy's.
Fresh copies 8 onward of seed '3113' (copies 0-7 were 0216's), 0034's sweep unchanged, eight at a time on eight cores, until one copy is at or above K4, or 2,242 copies all below (a flag), or the clock.
Verdict: caught (not a flag; A13.3 closes), or unresolved with p at most 1/(copies + 1).
10
Result and bounds
The harness's record: candidates, gates, loss, and the negative result this experiment leaves behind.
Carter 1923 written backwards, 24 known letters set after the reversal, random convention, random period-71 key; climbed by 0034's sweep at period 71 only (sliced); found = best backwards score above every pass-1 copy's (-5.553)
plant seconds
67.1
seconds
2,171.9
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.