- Rebuild the same readings as before for K4 and its 300 shuffles.
- Show all 301, one at a time and unlabelled, to the judge, who scores how much reads as English.
- If K4's reading is not among the top 5%, the excess was letter texture, not words.
Experiment 239
0237 flagged K4 at table-then-move, key length 2 (p 0.0099 against 100 shuffles) and 0238 replicated it (p 0.023 against 300 fresh shuffles; in that cell 1 of 300 at or above). K4's reading there is junk with 3 of 24 known letters in place; its move reads the carved text down a width-19 grid. Here 0238's search is rerun with the same seeds for lengths 1-2 only, which reproduces the length-2 readings exactly, and K4's and the 300 shuffles' best B2 readings go to the blind judge in one call (F13.3: the judge reads the survivors).
What the method does with the 24 known letters.
K4the carved letters The 97 letters cut into the copper at the CIA, shown at the top of every experiment page. really did score better than its shuffles in one setting, twice. But the scorer counts letter patterns, not words, and K4's letters are known to sit in slightly unusual patterns. So we show K4's reading and the 300 shuffles' readings to the language-model judge, blind, and ask whether any of them contains English words.
Words in the margin are defined on the glossary page.
The carved letters go in at the top; the reading comes out at the bottom.
Is 0237-0238's replicated excess words, or the texture of K4's carved letters?
The same readings regenerated for K4 and 300 shuffles and judged blind in one call.
K4 12, 14 shuffles at or above (max 15): words at the edge of chance, the known letters out of place. A lead for A2, not a solution.
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.
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
OBKR ends the line that finishes K3, then three lines of 31
out: the candidate
checked: the known letters in place, and the reading turns back into the copper
The loss of the best candidate, part by part. Zero is solved.
| Known letters missed | 20 of 24, ten points each | 200.00 |
| English | the 73 free letters score -6.99; English text scores -3.6 | 3.39 |
| Key | 669 bits to write it down (key words are cheap, arbitrary letters 4.7 bits each); each 60 beyond 60 costs a point | 10.15 |
| Loss | of 240, the worst possible | 213.54 89.0% |
Redo the method on a few letters with pencil and paper.
The same steps run by the code on the real letters.
Order A on a planted message: K3's rule with step 7 (letter i goes to place 7i mod 97), then the KRYPTOS table with SHADOW:
message THEHOTAIRESCAPINGFROMTHECHAMBERCAUSEDTHEFLAMETOFLICKERBUTPRESENTLYDETAILSOFTHEROOMWITHINEMERGEDTH
carved DPILCFDYKUWKWYMWSBAQVXLSDZFPDXGNGVJRECNUCIGSGRXELPEUQRLIBXISWIQUJRSJFUESQZLIISQZBUWRGUNMCQQLZRKTU
undone THEHOTAIRESCAPINGFROMTHECHAMBERCAUSEDTHEFLAMETOFLICKERBUTPRESENTLYDETAILSOFTHEROOMWITHINEMERGEDTH
The search tries all 27,852 moves in both orders, every length 1-20 and the 8 ways of using the table.
Family, key space, the constraint under test, and the clues it reads.
The excess is measured by MIX8m, a letter n-gram model; K4's carved order is known to be non-random at widths 7 and 21 (0056: corrected p 0.005 and 0.010), and a move that reads it at width 19 can turn such structure into n-gram-friendly letter pairs without any words. If the judge, which reads for words in order and ignores n-gram texture (reports/judge-bench.md: it beats 4-gram soup, 0233), puts K4's reading above all but 5% of the shuffles' readings, the excess is word-level: a lead for A2 (the known letters defined on an intermediate text). Otherwise it is recorded as one more statistic of the carved order and F13.4d closes on 0237's power. Pre-registered: word-level when K4's judge score has at most 15 of 300 shuffles at or above it; None counts as 0.
Written before the run; the result decides it.
The judge scores K4's reading low (under 20) and inside the shuffles' (most 0-20): not word-level. F13.4d closed for lengths 1-12 with the B2 excess written down as a statistic of the carved order.
The harness's record: candidates, gates, loss, and the negative result this experiment leaves behind.
All 4 candidates, lowest loss first.
judge: K4's B2 reading 12, shuffles at or above 14 of 300 (max 15): WORD-LEVEL; K4 statistic 4.62 at B2, p 0.007 against 300 shuffles; plant power ; decided lengths []
What was searched, as the harness recorded it.
NBENDJZERANDBCKNFILLBAADWRONNWIIHLGARTNRHHKORCULLHCNCSIDEPXMEDFCKRHALORKYXQTRGROWDOJOWSMSHDVNLYATHow this experiment sits among the others is on the lineage page; its family's open hypotheses are on the families page.