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

The Civil War campaign

Every key, ciphertext, code dictionary and traffic edge we hold for 1861 to 1865, cited to the source that prints it.

The Civil War is the first campaign on this site because it is the best documented cipher war in American history and it still has unread traffic in it. Both sides’ records were captured, filed and eventually printed, and where the editors hit a passage in cipher they set it in type exactly as they found it. The volumes and the microfilm behind them have since been scanned, so the raw material sits in public archives anyone can reach from a desk. Some of it has not been read since it was written.

The problems that record are catalogued here. What follows is the reference library behind them. The rule throughout is citation first: every key carries the document that attests it, every ciphertext the volume or frame that prints it, and every negative its denominator. Claims come in three tiers, attested, reported and conjectural, and the tiers are never mixed: a key with neither a source nor an independently verified derivation is not attested, whoever repeats it.

The corpus is also being worked as a graph rather than a shelf. A key belongs to a correspondence regime, not to a side: what matters is who could read whom, on which wire, between which dates, and which physical copy of the book sat at each end. Those traffic edges are where the remaining problems live, and the registers below are shaped to carry them. That work is in progress and this page reports only what is already evidenced.

September campaign update

The reconciled Civil War work adds 8 problem records and updates 8 existing records. Research remains paused. Partial readings and unsuccessful attempts are retained alongside decipherments; a new record is not a new solve.

The reference-library counts below preserve the 25 August register snapshot. Use the problem catalogue for current problem statuses and totals.

Counts on this page are read from the working registers, last reconciled 25 August 2026. The register files are being brought online progressively under /files/corpus/; a download link that does not resolve yet is queued, and the counts here are from the working copies.

Keys · Dictionaries · Ciphertexts · Sources · Full scans · Coverage · Downloads

Confederate keys

The census holds 26 keys: 17 attested, 7 reported, 2 conjectural. A key belongs to an evidenced correspondence regime, not to the Confederacy: Richmond routinely could not read its own generals. In December 1862 the Adjutant General's office wired J. E. Johnston: “Your dispatch of yesterday, in cipher, cannot be read; neither the President nor this office has the key.”

Two received rules fail against this table. The “always fifteen letters” rule is false: attested lengths run from 3 to 34, and the Tidwell register alone breaks it four times. And the keys are not all English: one is Latin, several are proper names. Replayed against its own traffic, the census recovers 59 of 63 scorable controls at rank 1.

Attested: 17

Established by a primary or scholarly source, or mechanically recovered and verified against primary witnesses.

Reported: 7

Reported by a source but not verified against matched traffic, or circulating historically on a missing or refuted citation. Kept visible precisely so nobody rediscovers and promotes them.

Conjectural: 2

Our own inferences, recorded with their outcomes.

Candidate keys, all conjectural: 37

Generated guesses with no source: source_citation = NONE on every row. All 37 were swept against the 63 scorable ciphertexts at every phase and three tableaux, 108,810 trials; none reached the z ≥ 9 reporting floor. A clean negative, published so the sweep is not re-run. The file's 38th row duplicates LIBERTY OR DEATH as an attested control and was skipped. This file is never merged into the key census.

Code dictionaries and books

The Confederate Navy enciphered by dictionary: a word becomes its page and ordinal position in an agreed printed book. A crib pair therefore selects an edition, never a word, and the notation shape fingerprints the page layout: two fields for one column, a binary marker for two columns, a ternary marker for three. Five books are documented in Confederate use; four are identified and on disk.

Around the identified keys sits a matching table of American and British pocket dictionaries of 1830 to 1875, censused by column layout, with 13 layouts verified from page images. Only two single-column families are known: Cobb's Miniature Lexicon and Sharpe's Diamond Dictionary (London, 1835). The one Confederate-published dictionary, the Palmetto Dictionary (Richmond: J. W. Randolph, 1864, 730 pp, two columns), appeared too late to be Maffitt's book.

Ciphertexts

Confederate census

The consolidated register holds 163 Confederate ciphertext and key-material rows, generated from SHA256-pinned source snapshots: 107 rows carry surviving cipher text, and 56 are register entries whose cipher pages did not survive, kept because an evidenced absence is still evidence. Dates run 1861 to 1866.

Beside the register, three hunt ledgers record the archival survey itself: what was located, what was screened, and what came back negative, by source class.

The M618 roll 19 census adds an item-level catalogue of 22 unique cipher records from the Confederate Secretary of War's incoming telegrams, December 1864 to April 1865: 8 of the 22 retain their raw cipher letters, the rest survive as clerks' interpretations or reference cards. These are problems 29 to 43, joined by a second pass over the earlier rolls at 71 to 77 and 82.

Union registers

The Union side is organised differently, because “Stager cipher” is not a key: the operative key is a specific physical code book, scoped to an operator, a station and a date. The registers therefore separate systems, physical key copies, channels, events, messages and witnesses rather than flattening them into one list. The denominators: 32 cipher and code book objects in 1,719 images, 35 telegram ledgers in 11,208 pages, and the Fort Monroe book EC25 censused at 840 telegrams across 400 pages. The Fort Monroe traffic is problems 44 to 61; the wider Union records are problems 62 to 70. Diplomatic and register material outside both lanes is problems 78 to 81.

Major source collections

What the corpus is built from, and who preserves it.

Full scans

The full-resolution page scans behind the problems live outside the repository: a single page image can run to tens of megabytes, so the working split keeps text and small evidence in git and the bulk scans in a private archive, with their hashes recorded here. The index below is what that archive holds, 65 acquisition directories totalling 13.8 GB, each anchored by the SHA-256 of its largest file. Nothing on this list is downloadable yet; the anchor hash is what makes a future copy checkable.

Coverage

The coverage ledger reconciles all of the above into 224 records, each graded by what is held and what is missing. They are coverage records, not messages or documents: the ledger collapses aliases and carries passage-level and whole-scope rows alongside single ciphers.

Downloads

The registers, under /files/corpus/. These are our own working files: the censuses, ledgers and coverage records behind everything above. The source scans they cite are a separate matter, indexed under Full scans and not yet hosted here.

Rows marked queued are not downloadable yet. A link to one resolves to a placeholder rather than the file, which is a promise recorded in public, not a delivery. Counts are from the working copies.