K · R · Y · P · T · O · S

K4 · live cryptanalysis

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

phases · attack-by-attack record

4 phases

One page per attack phase. Each documents the hypothesis, the parameter space, what actually ran on the runner, what was found (almost always nothing), and the methodology lesson the phase taught us. Failures are surfaced, not buried — the catalog of ruled-out approaches is the project's most durable output.

  1. 2A.1
    Width-7 columnar transposition + KRYPTOS-keyed Vigenère~333 M attempts on width-7 columnar transposition × KRYPTOS-keyed Vigenère key lengths 5–15. 0 candidates. Demoted from priority #1 to #5 the same day after the period-7 attribution test showed K4's lag-7 signal is plaintext-side, not encryption-side.Bombe
    closedfalsified (narrow)
  2. 2B.1
    Weltzeituhr keystream sweep1.26 B attempts on a 96-candidate space. Falsified under the (first_letter, 1988_best_guess) parameterization only. Broader sweep continues as Phase 2B.1.B.Gnomon
    closedfalsified (narrow)
  3. 2B.1.B
    Weltzeituhr derivation-rule + city-list sweep25× larger search space than 2B.1 — 5 derivation rules × 5 city lists × 24 offsets × 2 directions × 2 alphabets = 2,400 unique candidates per pass. Live sweep.Gnomon
    activepursuing
  4. A.1
    Atelier — creative AI layer launch (parallel track)Stein-inspired hypothesis-generation agent. Opus 4.7 via Anthropic API, prompt-cached corpus, $50/mo hard budget cap. Emits structural hypotheses, not plaintext guesses. Runs in parallel with the Hetzner brute-force runner.Bombe
    activepursuing