crackciphers.com
[◑]

Compete to crack secrets lost to history—for fun, clout, and (no) profit


Links

Problems › 19 › solution › 1

Seventy-nine transmitted words, one recovered route

Finding a message among ordinary words

The result is a recovered 68-word dispatch. Bascom relays instructions from Cox to Milroy: Kelley has been ordered to remain ready for movement under army headquarters, and Milroy must account for that restriction when issuing orders to his detachments. This was recovered by finding an unknown reading route, without a supplied plaintext. The manuscript alternatives, uncertain year, and unresolved publication priority remain part of the result.[2]

Before knowing the answer, the sheet offers a useful structural clue. It contains 79 transmitted tokens, mostly ordinary English words. Set aside the opening Turkey and the remaining 78 divide into six blocks of thirteen. Treat each block's last word as a check and twelve words remain in each block. That gives 72 words for a rectangle twelve rows high and six columns wide. This was the registered family of possibilities, not a geometry selected after reading a promising sentence.

Why the rectangle is not the answer

Knowing the dimensions leaves the order of six columns and the direction of each unresolved. There are 6! possible column orders and 26 choices of up or down, or 46,080 routes in this family. Each one returns ordinary English words because the cipher changes their positions, not their spelling. A wrong route does not produce impossible letters. It produces real words in the wrong relations.

The source also has four disputed words. At transmitted positions 13, 43, 68, and 78, the held transcript reads general, move, hold, nice; the independent image-first record reads guard, more, holds, vice. All sixteen combinations were registered before scoring. The program was not allowed to keep only whichever spelling improved the sentence.[3]

An earlier campaign stopped after eleven of a planned twenty-four qualification examples. Its statement that the historical telegram was never scored is true of that earlier pause. It is superseded by the September run: all twenty-four qualification examples passed before the separate historical adapter ran. The page records the completed result rather than carrying the old unexecuted-target label forward.

Route that survived every source variant

The frozen search used word-based screening of the full route family followed by the existing language-model reranking and masked-word scoring stages. Such scores favor coherent English, but they do not certify an answer. Here every one of the sixteen source variants selected the same route. Its internal identifier was 2290, with the explicit rule 1-3-4-6-5-2 | UUDDUD. The route itself, rather than that software identifier, is what lets a reader reproduce the result.

Number the six transmitted twelve-word blocks from one to six. Plaintext columns one through six receive transmitted blocks 1, 3, 4, 6, 5, 2. Put the words of blocks 1, 3, and 5 into their columns from bottom to top; put blocks 4, 6, and 2 in from top to bottom. Then read the completed rectangle across each row, from left to right.

Plaintext columnTransmitted blockPlacement
11Bottom to top
23Bottom to top
34Top to bottom
46Top to bottom
55Bottom to top
62Top to bottom

Reconstructed rectangle

Brackets contain the independently recorded source alternatives. They are not corrections selected from the recovered meaning.

123456
general [guard]Kelleyhavingreceiveddirectionsfrom
generalHallecktohold [holds]himselfin
readinesstomove [more]underordersfrom
armyheadquartersthiswillinterfere
withmovementsatpresentExceptsuch
asareimmediatelyandpressinglynecessary
givesuchorderstoyourdetachments
aswillbestprovideagainstcontingencies
andkeepgeneralKelleyandthese
headquartersfullyinformedofthe
enemysmovementsbycommandofmajor
generalCoxEndnice [vice]daythis

Recovered dispatch

General [source alternative: guard] Kelley, having received directions from General Halleck to hold [holds] himself in readiness to move [more] under orders from army head quarters, this will interfere with movements at present except such as are immediately and pressingly necessary. Give such orders to your detachments as will best provide against contingencies and keep General Kelley and these head quarters fully informed of the enemys movements. By command of Major General Cox.

Punctuation and sentence capitalization are editorial. The source spelling enemys and the separate words head quarters are preserved. The message ends after Cox, at position 68 in the reconstructed rectangle. Four cells remain: End [nice|vice] day this. Interpreting End as the message terminator and the next three words as padding is plausible, but no recovered instruction establishes that interpretation.

Every transmitted token

The table retains all 79 wire positions. The 72 placed words receive a row, column, and reading position. Turkey and the six block-ending words are kept explicitly as framing under the registered convention. They are not words the solver quietly discarded after seeing the answer.

Wire positionSource wordSource alternativeRow, columnReading positionRole
1TurkeyIndicator
2general12, 167Message
3enemys11, 161Message
4head10, 155Message
5and9, 149Message
6as8, 143Message
7give7, 137Message
8as6, 131Message
9with5, 125Message
10army4, 119Message
11readiness3, 113Message
12general2, 17Message
13generalguard1, 11Message
14bullyBlock check
15from1, 66Message
16in2, 612Message
17from3, 618Message
18interfere4, 624Message
19such5, 630Message
20necessary6, 636Message
21detachments7, 642Message
22contingencies8, 648Message
23these9, 654Message
24the10, 660Message
25major11, 666Message
26this12, 672Probable padding
27chapBlock check
28Cox12, 268Message
29movements11, 262Message
30quarters10, 256Message
31keep9, 250Message
32will8, 244Message
33such7, 238Message
34are6, 232Message
35movements5, 226Message
36head4, 220Message
37to3, 214Message
38Halleck2, 28Message
39Kelley1, 22Message
40tightBlock check
41having1, 33Message
42to2, 39Message
43movemore3, 315Message
44quarters4, 321Message
45at5, 327Message
46immediately6, 333Message
47orders7, 339Message
48best8, 345Message
49general9, 351Message
50fully10, 357Message
51by11, 363Message
52End12, 369End marker, interpreted
53ropeBlock check
54day12, 571Probable padding
55of11, 565Message
56of10, 559Message
57and9, 553Message
58against8, 547Message
59your7, 541Message
60pressingly6, 535Message
61Except5, 529Message
62will4, 523Message
63orders3, 517Message
64himself2, 511Message
65directions1, 55Message
66dailyBlock check
67received1, 44Message
68holdholds2, 410Message
69under3, 416Message
70this4, 422Message
71present5, 428Message
72and6, 434Message
73to7, 440Message
74provide8, 446Message
75Kelley9, 452Message
76informed10, 458Message
77command11, 464Message
78nicevice12, 470Probable padding
79finisBlock check

Checks that could have failed

Independent inverse encoding put the reconstructed rectangle back on the wire, adding the retained indicator and block checks. It reproduced all 79 source tokens for every registered source variant. This checks placement and accounting; by itself it cannot establish that a reversible route is the historically intended one.

The separate comparison asked whether the complete scoring procedure would find an equally strong result after the payload positions were randomly shuffled. Each of 199 shuffles used the same rearrangement across all sixteen source variants and reran the frozen pipeline, comparing the best result across the whole family. None of the 199 shuffled families matched or exceeded the observed score. The registered calculation therefore gives (1 + 0) / 200 = 0.005. All 3,200 observed and comparison fits completed, and the coordinator independently recomputed the family maxima and checked the saved fit-file hashes.[2]

That number describes this finite comparison under its declared randomization assumptions. It is not a 99.5 percent probability that the decipherment is correct, does not cover every earlier campaign search, and says nothing about first publication. Coherent content, one route across all source alternatives, complete inverse encoding, and historical correspondence are separate reasons to accept the reading.

Dating the sheet

The catalogue says 4 December 1861. The manuscript's visible dateline does not clearly supply a year. Contemporary correspondence strongly favors 1862: on 3 December 1862 Halleck asked Kelley to be ready for movement; Kelley then wrote Bascom at Charleston explaining why Milroy's proposed movement had to await Halleck's orders. Those telegrams are printed in the Official Records, series I, volume XXI, page 825. Pages 829 and 830 continue the related December 4 correspondence. They fit the recovered chain of instructions, but none is a duplicate of this message.[4]

The rank is consistent with the same concern: a repository description of Cox's 24 December 1861 letter calls him brigadier general, while the recovered closing here says major general. That supports the dating question but does not settle it. Until an original-year notation or a dated duplicate is inspected, the date must remain “1861 in the catalogue; probably 1862.”

What is not claimed

Publication priority remains unresolved. A candidate-led screen found no matching plaintext or target-specific published solution in 876 text objects and twenty web queries. The objects comprised 228 Official Records OCR witnesses, 643 Cryptiana texts, Plum volumes I and II, Bates, Weber's Masked Dispatches, and Noyalas's 2003 thesis. Several objects represent the same publication, so this is not 876 independent books. Thirty-five retained fuzzy-match contexts were inspected and rejected as other documents.[5]

The later Google Books inspection of Noyalas's 2006 biography narrowed one gap but did not close it. Selected index entries and parts of pages 73–77 were inspected, including all of pages 75–76. Searches for Bascom and “pressingly necessary” returned no displayed hits. These are preview results, not a full-book negative. The four Paulus volumes and unviewed parts of Noyalas remain outside full-text coverage. In particular, Paulus volume I, printed pages 130–147, is a specific outstanding source request, not an inspected source.[6]

The four manuscript alternatives are not resolved by the route. Agreement on the route shows that those alternatives do not change the structural result. It does not prove that a modern transcription is right wherever its grammar is better. Higher-resolution archival masters or a duplicate could settle the pen strokes. The meaning of the four final cells, the exact year, and any first-ever decipherment claim remain qualified. This is a successful project decipherment with a reproducible route and a bounded prior-publication search.

References