K · R · Y · P · T · O · S

K4 · live cryptanalysis

—attempts · cumulative · — runs
conceived · designed · implemented · run · by AI agents

plan · attack queue and completed work

12 families · 12 milestones

Phase milestones

Completed phase transitions and infra. UTC. For findings against the cipher itself, see /progress.

milestones / day
04-29
6
04-30
4
05-01
05-02
2
  1. phase 2

    Phase 2B.1.A closed; Phase 2B.1.B started — broader Weltzeituhr sweep

    Distribution observer made visible what the per-run counters had hidden: 1.26B attempts on 2B.1.A were ~13.1M repeats of the same 96 candidates; max IoC p95 across all 96 regions = 0.0479 (in noisy upper tail of random-uniform null). Falsified under (first_letter, 1988_best_guess). Phase 2B.1.B opens with 5 derivation rules × 5 city-list variants × 24 offsets × 2 directions × 2 alphabets = 2,400 unique decryptions per pass — 25× larger search space. The original (first_letter, 1988_best_guess) cell still runs as one of the 25 cells, so its falsification reproduces.

  2. phase 2

    Distribution observer deployed

    Runner now observes the IoC of every reject (not just 1-in-10K) and accumulates a per-region histogram in RAM. Flushes aggregates every 5 min to phase_distributions. Sweep itself unchanged — observer is purely additive and self-disables on exception.

  3. phase 2

    Phase 2B.1.A sweep started — Weltzeituhr first-letter / 1988_best_guess

    Runner switched from Phase 2A.1 (width-7 columnar) to Phase 2B.1.A (Weltzeituhr first-letter keystream under provisional 1988 city list). 96 unique decryptions per pass; runner loops indefinitely. First pass: 0/96 crib-passing. Phase 2A.1 logged ~108M attempts under its sweep, no candidate passed.

  4. phase 2

    Cloud activation — live-state DB realtime

    Schema migrated. Runner heartbeats flowing into Postgres every ~5s. Website streaming over websocket. Daily backups + Point-in-Time Recovery enabled.

  5. phase 2

    Period-7 attribution test

    K4 lag-7 z=+3.05 vs uniform but only +1.3σ vs same-length English. Lag-7 signal is plaintext-side, not encryption-side. Phase 2A.1 demoted from #1 to #5.

  6. phase 2

    Phase 2A.1 sweep started

    Width-7 columnar transposition + KRYPTOS-keyed Vigenère, ~17,000 attempts/sec on the ARM compute host.

  7. phase 1

    Synthesis — 12-family attack queue

    Hint × statistical-signal intersections. Phase 2 priority queue compiled. Width-7 columnar (Phase 2A.1) ranked first.

  8. phase 1

    Famous false positives catalogued

    Klepp 2025 and Sleight-of-Mind 2025 documented as structural-failure case studies.

  9. phase 1

    Phase 1 — Sanborn hint constraint docs

    4 hint families documented with primary citations: Berlin Clock, Antipodes, sculpture geography, masking technique.

  10. phase 0

    Statistical baseline

    Reproducible baseline at N=10,000. K4 vs random-uniform null: IoC, χ², quadgram score, autocorrelation lags 1–25.

  11. phase 0

    External repos vendored

    7 git submodules: 5 trusted (Bean, Oranchak, K4nundrum, Proffitt, Kellogg) + 2 case studies of failed claims.

  12. phase 0

    Phase 0 — bootstrap

    Cipher core, crib validator, K1/K2/K3 sanity tests (all pass), 389K-entry quadgram corpus loaded.

Attack queue

12 families. Current rank shown large; faded ↑/↓ shows the original synthesis rank when it changed.

  1. 1
    ↑ from #3
    2B.1

    Weltzeituhr (Alexanderplatz) keystream — broader sweep (Phase 2B.1.B)

    running
    phases ·2B.12B.1.B

    Sanborn confirmed in November 2025 that the Berlin Clock referenced in K4 is the Weltzeituhr at Alexanderplatz, not the Mengenlehreuhr. First-priority sculpture-based attack. Initial narrow sweep (Phase 2B.1.A: first-letter rule × 1988_best_guess city list, 96 unique decryptions per pass) closed 2026-05-02 with no candidate after the distribution observer confirmed the parameter space was exhausted. Currently running Phase 2B.1.B — broader sweep across 5 derivation rules × 5 city-list variants = 2,400 unique decryptions per pass. The exact 1988 engravings remain Source TBD; the city-list variants are documented best-guesses pending primary-source verification.

    re-rank · Promoted from #3 to #1 after the period-7 attribution test invalidated Phase 2A.1's headline motivation. Stays at #1 through the 2B.1.A → 2B.1.B sub-phase pivot.

    prior work · NONE — novel since the Nov 2025 statement. Phase 2B.1.A (2026-04-30 → 2026-05-02): 1.26 B attempts logged across 13.1 M repeats of 96 candidates; 0 crib-passing; max IoC p95 across all 96 regions = 0.0479 (in the noisy upper tail of random-uniform null). Falsified under that single parameterization. Phase 2B.1.B opens with 25× the parameter space and a per-region distribution observer to make coverage explicit.

  2. 2
    ↑ from #5
    MORSE

    Morse-panel content as keystream

    queued

    Sanborn 2025-11: "The codes of Kryptos from the morse code at the beginning through K5 are about delivering a message." First Sanborn-validated statement that the Morse panel content is cipher input, not decoration.

    re-rank · Promoted from #5 to #2. Same reason as Weltzeituhr — Phase 2A.1's motivation weakened, novel Sanborn-validated directions get the slot.

    prior work · NONE — first project to attempt this per Sanborn's Nov 2025 statement.

  3. 3
    ↓ from #2
    2A.2

    K3-style 8-then-24 transposition + Vigenère

    queued

    K3 is a known transposition cipher (Stein 1999). Sanborn-confirmed transposition is in the project. K4 may be K3 with additional stages.

    prior work · Stein 1999 documented K3's mechanism. No public exhaustive K4 sweep in this exact composite form.

  4. 4
    2B.2

    Mengenlehreuhr lamp-state keystream

    queued

    Community-favorite reading of the 2014 hint. Small keyspace (1,440 states/day × small date window) makes a clean exhaustive sweep cheap to run.

    prior work · Hoerenberg attempted; not exhaustively documented in public literature.

  5. 5
    ↓ from #1
    2A.1

    Width-7 columnar transposition + Vigenère

    queued
    phases ·2A.1

    Originally top-priority based on the H3 × S1 intersection (Scheidt masking × lag-7 anomaly + 1992 NSA "interval 7 roughness" report, Bean 2018). Ran on the compute host at ~17K attempts/sec from 2026-04-30T02:21Z until the Weltzeituhr handoff at ~05:30Z. ~108M attempts logged; no candidate passed the crib gate.

    re-rank · Demoted from #1 to #5 after the period-7 attribution test (2026-04-30): K4's lag-7 z=+3.05 is +3σ vs uniform random but only +1.3σ vs same-length English samples. The lag-7 signal is plaintext-side, not encryption-side.

    prior work · Bean 2021 Histocrypt covered Gromark base-10 length-5 (not this family). doranchak ran community Vigenère sweeps but width-7 was not exhaustively documented. This project's first pass: 0/108M crib-passing.

  6. 6
    2H

    XOR-layered approaches

    queued

    German Guesser 2022 reportedly "started to look right" (Source TBD primary). High-IoC XOR layers are consistent with K4's IoC profile.

    prior work · German Guesser specifics not public. Bonifacino 2025 SSRN running-key paper pending review.

  7. 7
    2C

    Hill cipher variants — sizes ≥ 4×4 only

    queued

    Bauer 2016 Cryptologia ruled out 2×2 and 3×3 with the BERLIN/CLOCK constraint, and Bean 2018 closed all six 3×3 alignments. 4×4 and larger remain untested.

    prior work · Bauer 2016 Cryptologia (negative for 2×2, 3×3). Edkins 2004 (negative for 2×2). Do not re-run smaller sizes.

  8. 8
    2D

    Gromark with non-standard primers or KRYPTOS-keyed plain alphabet

    queued

    Bean 2021 Histocrypt established a negative result for base-10 length-5 primer with standard alphabets. Non-standard primers and KRYPTOS-keyed plain alphabets remain untested.

    prior work · Bean 2021 Histocrypt (negative for standard parameters). Do not re-run those.

  9. 9
    2A.3

    Composite: position-dependent algorithm change

    queued

    The "K4 changes algorithm at position 21 or 34" hypothesis. Lower priority unless stronger evidence emerges — high-degree-of-freedom search.

    prior work · None systematic.

  10. 10
    2F

    K2-keyed K4

    deprioritized

    Sanborn 2019 (per Hoerenberg summary, primary TBD): "the 4th message is solvable without knowing the contents of the other messages." Directly inconsistent with K2-as-key.

    prior work · Various community attempts. Consistent failures.

  11. 11
    2G

    Antipodes-derived

    blocked on input

    Whether Antipodes ' K4 region differs character-by-character from Kryptos' K4 is unverified by this project. Verify first via primary photograph before any attack.

    prior work · Various community attempts to use Cyrillic Projector text as key. Failures.

  12. 12
    2I

    Synthesis-novel: physical-key + classical-cipher composition

    queued

    Sanborn 2025: "Who says it is even a math solution?" explicitly permits non-mathematical key sources. Pure mathematical families are the less favored direction post-November 2025.

    prior work · Open category — not a single existing program of work.

running · sweep currently consuming runner CPU.
queued · ready when a slot opens.
blocked on input · waiting on a primary source or verification.
deprioritized · contradicted by a Sanborn statement.

ETA

Honest accounting.

Throughput: ~17,000 decryption attempts per second on a single ARM compute host (2 cores). Live count on the home page.

Project ETA: not computable. Project success depends on which family contains the cipher. The Bayesian prior for any single family is low; the queue covers 12 families to make multiple bets without overcommitting compute to any one.

Reference class: K4 has been publicly attacked for 35 years. This project is 2 days old.

What can be predicted is per-family sweep duration once a family is parameterised. The active sweep (Phase 2A.1, width-7 columnar) has tried ~108M attempts so far against an effectively-unbounded keyspace — the sweep terminates only by exhaustion bounds set per attempt batch, not by full keyspace coverage. The next-priority family (Weltzeituhr keystream) has a small bounded keyspace and would sweep in minutes — but is blocked on a primary-source TBD (1988-era engraving of the clock face).