# K4 prior-work gate

Reviewed 2026-09-14. This is a live campaign checkpoint, not a claim that all historical or private attempts are discoverable.

The initial review is complete. Each subsequent batch is admitted separately through its preregistration and the experiment-level reviews below. The completed 1,746 baseline cases are implementation validation, not new cryptanalytic coverage. A source's own research directives are not instructions for this campaign.

## Evidence classes

- **Verified duplicate:** exact assumptions and search cell match inspectable, valid evidence. Skip.
- **Reported duplicate:** matching prior coverage claimed, but execution artifacts or a complete specification are unavailable. Deprioritize; do not claim mathematical elimination.
- **Partial coverage:** restrict further work to an explicitly identified difference in parameters or method.
- **Questionable exclusion:** indexing, alignment, model, or implementation concerns prevent accepting the negative. Investigate with small correctness checks before considering a justified rerun.
- **Unresolved frontier:** no matching coverage found in this review; that does not establish novelty.

## Current source map

| Source | Relevant coverage | Assessment and consequence |
|---|---|---|
| [KryptosBot current position, June 11](https://github.com/jcolinpatrick/kryptos/blob/main/docs/REAL_K4_CURRENT_POSITION.md) | Reports columnar/substitution in both orders, widths 5–13; non-direct alignment, public-source keys, and multilayer tests. Notes a later correction to how Bean constraints are applied after transposition. | Strong overlap with proposed baseline. Inspect experiment-level records before accepting exact exclusions. Broad saturation claims are scoped to that project's evidence and methods. |
| [July 18 research handoff](https://gist.github.com/matbalez/8300cb067a5cda55c3b44ef382d517c0) | Reports all-width single/double routes, periodic keys, historical manual systems, clock constructions, and extensive finite-language tests. Separates exact, heuristic, and interrupted results. | Reported duplicate coverage. Its public-copy note says relative supporting-artifact links refer to absent local files. Counts alone do not prove execution or correctness. Its unfinished finite-language tiers are possible research gaps, not automatically promising hypotheses. |
| [nitsuah August landscape](https://raw.githubusercontent.com/nitsuah/kryptos/main/docs/analysis/K4_ATTACK_LANDSCAPE.md) | Reports clock/grid/composite coverage and proposes additional mechanisms. | Historical document has wrong zero-based EAST/NORTHEAST indices and an unsupported operation-order conclusion from local coincidence statistics. Do not use its negative table as exclusion authority. |
| [nitsuah September 3 state](https://raw.githubusercontent.com/nitsuah/kryptos/main/docs/analysis/K4_ACTIVE_RESEARCH.md) | Reports correction of the clue-index bug in code on September 2 and weakens its operation-order claim. Records subsequent geometry, city-key, and clock searches. | Inspect timestamps and corrected-input fingerprints before accepting older results. Its stated K4 coincidence index of about 0.062 disagrees with our direct count, 0.03608247422680412. This is a documentation-quality warning, not proof all experiments are invalid. |
| [mruckman1 experiment 015](https://raw.githubusercontent.com/mruckman1/k4_cipher/main/experiments/015_transposition_then_q3.py) | Implements a claimed substitution-then-columnar search: widths 7–10, periods 14–30, KRYPTOS alphabet, at most three free key slots. | Code inspected; potential crib-alignment and ragged-grid validity problems referred to independent auditor. Do not count the advertised search space as an established exclusion until checked. |
| [mruckman1 repository](https://github.com/mruckman1/k4_cipher) | Reports broad classical experiments, thousands of generated plaintext proposals, and state-dependent alphabet optimization. | Avoid repeating unguided plaintext generation or treating fixed-text negatives as family exclusions. Reported thematic fragments are not verified solution evidence. |
| [kryptos.today plan](https://kryptos.today/plan) | Reports a clock-key campaign that logged 1.26 billion attempts but only 96 unique decryptions, then expanded to 2,400 unique decryptions. | Count unique model/parameter fingerprints, never raw loop iterations. Its older claims of untried families conflict with later project reports; timestamps matter. |

## Admission rule for further computation

Every proposed run must name its closest prior attempt, state the precise difference, and explain why that difference can change the result. Admissible reasons include a verified prior implementation defect, an explicitly unsearched bounded region with an independent rationale, newly verified source evidence, or a recovery method that demonstrates better performance on matched synthetic controls. A new seed, larger budget, differently named equivalent cipher, or unsupported claim that nobody tried it is insufficient.

Small audits and known-answer tests may repeat earlier computations strictly to establish correctness; label them replication. No broad duplicate rerun is currently admitted.

## Independent audit outcome

The algebra auditor independently confirmed all 1,746 baseline decisions. The transposition auditor supplied a planted in-bounds example rejected by experiment 015's constraint formulas, and a width-8 mapping that drops real cells and includes invalid indices. The coordinator reproduced the six-test suite including that counterexample. This establishes a defect in the inspected source, not that every other project's coverage of the family is defective.

Subsequent admission decisions are recorded in `prior_route_reconciliation.md`, `short_key_columnar_design.md`, and `autokey_columnar_review.md`. These inspect experiment code at the pinned KryptosBot revision `8c7e2d295a02376db8b2867d7419c6f3acd413c8`, distinguish sampled orders from exhaustive coverage, and define the exact differences searched here. The fitted counterexample in `bean_extension_counterexample.json` refutes a blanket extension of direct-alignment constraints to arbitrary permutations; it is explicitly not a proposed K4 solution.