01 / SEARCH & REVIEW
Normal Human Man
Start small: all 26 Caesar shifts and six rail-fence heights. A plain, useful baseline.
CIPHER LAB / A WORKBENCH FOR THE UNKNOWN
Cipher LabFind a method. Follow the evidence.
Explore classical ciphers with the solvers from Undeciphered Texts.
FIG. 01 / THREE LETTERS APART
01KH
02HE
03OL
04OL
05RO
Runs on your deviceOriginal Python engineCandidates with evidence
01 / TRY A READING
Work the cipher.
Start with an example,
or bring your own transcription.
Try an example
02 / EVIDENCE & CANDIDATES
Ready for inputLoad an example or paste your text, choose a solver, then run. The engine loads only when you need it.
A reading starts with a check.
Candidate plaintexts, forward checks and search limits will appear here. An English-looking answer still needs independent evidence.
Inspect the complete report
Explore the ten solver personas
03 / COMPLEMENTARY METHODS
Different ways into the same question.
Ten research personas.
One standard for correctness.
02 / SEARCH & REVIEW
Emperor
A systematic classical investigation with a factual ledger and optional neural advice.
03 / SEARCH & REVIEW
Inheritance
Use your keyword guesses, word list and aligned cribs. It needs caller-supplied clues.
04 / SEARCH & REVIEW
Hallucinogens
Try finite reversals, rotations and affine combinations. Every candidate must reproduce the ciphertext.
05 / SEARCH & REVIEW
Pacifist
Try the finite affine keys and exact crib checks, without persona vocabulary preferences.
06 / SEARCH & REVIEW
Detective
Place an unplaced crib and look for repeated-key evidence. Enter the clue in the unplaced crib field.
07 / SEARCH & REVIEW
Cartographer
Explore bounded rail, route, Redefence and rectangular transposition layouts.
08 / SEARCH & REVIEW
Mechanic
Use aligned cribs for progressive keys, autokey recurrences and Hill constraints.
09 / SEARCH & REVIEW
Adversary
Look for alternative readings consistent with the same supplied evidence.
10 / SEARCH & REVIEW
Skeptic
Review a proposed reading against reserved cribs or a reference hash. A dry voice, exact checks.
04 / KEEP THE EVIDENCE
A candidate is
the beginning.
What can this cipher solver do?
Try bounded Caesar, affine and transposition searches. Supply cribs for repeated-key, autokey and Hill inference. Connect the research personas in a shared council, then inspect their assumptions, checks and candidate witnesses.
What does Bob the Neural Net do?
Bob ranks known cipher families from statistical and cryptanalytic features. A family suggestion helps choose the next method; it does not establish a key or decrypt the text. Its model and benchmark provenance ship with the workbench.
When is a cipher actually solved?
A candidate must account for the ciphertext and survive evidence that was not used to fit it. A high language score, several agreeing personas, or an exact fit to your own clue is not independent proof. The engine keeps unresolved historical claims unresolved.
Where can I inspect the code?
This workbench runs a pinned snapshot of the open-source Python engine. The repository includes source-backed controls, tests, certificates and a structured case workflow for unsolved ciphers.
Engine and model provenance
The first run downloads Python and NumPy from the pinned Pyodide CDN, then executes the original engine in a worker on your device. Network access is needed for those initial files. Inputs and results are not uploaded to a solver service.
The exact engine snapshot and file hashes are listed in the build manifest.
05 / RESEARCH NOTES
Undeciphered scripts.
Open historical ciphers.
Primary sources checked
The workbench above handles classical A-Z text. It does not decipher unknown glyph scripts. Those cases need a documented sign inventory, preserved uncertainty and linguistic or archaeological evidence beyond an English language score.
These dated notes separate the source record from proposed experiments. Research plans are not executed results. “Open” describes the cited evidence, not a promise that no one has proposed an answer. A solved control checks a method; it does not solve a different historical target.
Kryptos K4Historical cipher
Archival recovery documented; public method not established here.
Archival text recovery and the 2025 sale are documented. Paradigm identified itself as custodian in June 2026 and offers a reference-answer verifier while still calling K4 unsolved. The checked sources do not publish a reproducible full method. RR Auction listed private K4 plaintext and coding material, attributed to Sanborn without examining the contents independently.
RR Auction: archive discovery, 23 October 2025 RR Auction: private archive sale, 21 November 2025 Paradigm: current custodian update, 12 June 2026 RR Auction: archive lot and verification limits
Next experiment. Reserve one published clue while fitting another, then test every predicted letter and the complete forward transform. A failed bounded search rules out only its stated models.
Dated research note and evidence limitsVoynich manuscriptUnidentified script
Meaning unresolved; encipherment itself is uncertain.
Yale identifies an unknown script and cautions that the manuscript may not be encoded. EVA and other transliterations represent visible signs, not established sounds or translations.
Next experiment. Compare two versioned transliterations against the same Yale folios. Hold out entire folios and test whether measured patterns survive alternative sign and space readings.
Dated research note and evidence limitsLinear AUndeciphered writing system
Approximate sign readings do not establish the language.
SigLA connects inscriptions with sign occurrences and their material context. Some conventional readings borrow comparable Linear B sign values; that does not translate the underlying Linear A language.
Next experiment. Preserve sign numbers, uncertain readings and document types. Test repeated sequences on held-out sites rather than selecting a proposed language from a few attractive matches.
Dated research note and evidence limitsIndus signsUnresolved sign system
Statistical structure is evidence about sequences, not a translation.
The authored n-gram study identifies limited texts, no bilingual anchor and an unknown underlying language. Its predictive experiments concern sign sequences; they do not provide a decipherment.
Next experiment. Keep artifact identifiers and seal-versus-impression orientation. Evaluate masked signs on held-out artifacts against simple frequency baselines, excluding duplicate impressions across splits.
Dated research note and evidence limitsRongorongoUndeciphered script
A material date does not determine an inscription’s reading or date.
The 2024 radiocarbon study reports older wood for one tablet but explicitly discusses reused wood and potentially later engraving. Museum records also distinguish original tablets from modern casts.
Ferrara and colleagues: radiocarbon study, 2024 British Museum: original wooden tablet British Museum: plaster cast record
Next experiment. Build an object ledger separating originals, casts and copies. Preserve alternating line orientation, then test parallel sign passages on tablets excluded from the fitting set.
Dated research note and evidence limitsPhaistos discUndeciphered inscription
The museum reports no definitive interpretation.
Heraklion’s object description records 241 stamped signs in 61 groups across two sides. Proposed word boundaries and a hymn interpretation remain interpretations, not recovered plaintext.
Next experiment. Compare independent sign annotations while preserving side, group and reading direction. Report sensitivity to disputed signs before attempting a language mapping on this single object.
Dated research note and evidence limitsZodiac Z13 and Z32Short historical ciphers
Open in the cited authors’ account; Z340 is a solved control.
Z340’s solvers describe its verified solution and the remaining short ciphers. They explain why many plausible Z13 or Z32 readings can fit the available symbols without establishing a unique answer.
Next experiment. Reproduce the published Z340 transform as a separate control. For Z13 and Z32, enumerate competing compatible readings and require external evidence before treating a name or location as established.
Dated research note and evidence limitsDorabella cipherHistorical cipher
The cited experiments do not establish a systematic reading.
Hauer and colleagues test language, substitution and music hypotheses. Their 87-character transcription includes ambiguous symbol orientations. Failed solver trials are evidence about tested assumptions, not a proof that every possible method fails.
Next experiment. Freeze alternative glyph readings before search. Compare bounded trials with matching synthetic controls, keeping attractive English phrases separate from independently verified evidence.
Dated research note and evidence limitsSource extraction limits, corpus access and reproducible bounds are recorded in the notes. No new historical decipherment is claimed. Read the research register and experiment standards.