Exact finite-state TMTO paper and reproducibility artifact

[qoosmo]

Summary

This PR transforms the historical Ensimag–UJF M2P SCCI cryptography/HPC project into a modern, reproducible research artifact and adds the focused ePrint manuscript:

Exact Finite-State Coverage of Hellman and Distinguished-Point Time–Memory Tradeoffs: A Reproducible Study on Reduced-DES Mappings

The goal is to preserve the historical work, make the modern implementation auditable, and connect classical TMTO theory with exact finite-state experiments and reproducible Java/Rust implementations.

What is included

• Preserved historical presentation and legacy prototype
• Modern Java implementation with 31/31 tests passing
• Independent Rust implementation with fmt / clippy / tests passing
• Literal shared Java/Rust experiment plan with byte-identical deterministic outputs
• 840-row Hellman DES-vs-random-map coverage sweep
• 540-row distinguished-point parameter sweep
• 279-row exact finite functional-graph reachability analysis
• Reproducible analysis scripts and generated tables
• 21-page research manuscript
• Exact Ma–Hong coverage comparison
• Exact distinguished-point reachability-floor analysis
• Vector-font PDF suitable for archival/ePrint use
• Historical presentation preserved byte-for-byte

Main Hellman result

For $N = 2^{16}$, $m = 1024$, and $t = 64$:

Quantity	Coverage
Ma–Hong prediction	16.6514%
DES-derived mapping median	16.8396%
Random-map median	16.8098%

The tested reduced-DES mapping therefore shows no anomalous Hellman-coverage effect relative to the established random-function model and matched random-map controls.

This also corrects the earlier use of the independent-occupancy heuristic $1-e^{-1}$ as a Hellman coverage reference. The appropriate fixed-map TMTO baseline is the Ma–Hong recurrence.

Distinguished-point result

For difficulty parameter $d = 8$, exact analysis of the complete $2^{16}$-state functional graph gives:

Eventual truncation floor: 29.715%

This means that approximately 29.7% of starting states can never reach a distinguished point, regardless of how large the chain bound becomes.

The observed long-chain plateau is therefore explained by the structure of the finite functional graph rather than by an implementation error.

The exact DES-derived eventual truncation floors measured in the complete state space are approximately:

• $d = 4$: 0.0305%
• $d = 6$: 0.3143%
• $d = 8$: 29.7150%

Java / Rust reproducibility

The Java and Rust implementations consume the same literal experiment plan.

On the shared 30-trial workload:

• Java and Rust deterministic result CSVs are byte-identical
• Hellman recovery: 4/30
• Hellman median exact coverage: approximately 16.825%
• Distinguished-point recovery: 3/30
• Distinguished-point median exact coverage: approximately 15.412%

This is a semantic-equivalence result, not a performance claim.

Reproducibility gates

• ✅ Java: 31/31 tests passing
• ✅ Rust: fmt / clippy / tests passing
• ✅ Java ↔ Rust shared-plan outputs: byte-identical
• ✅ 840 Hellman result rows
• ✅ 540 distinguished-point sweep rows
• ✅ 279 exact DP graph-analysis rows
• ✅ Historical presentation SHA-256 preserved
• ✅ Generated result tables reproducible from committed CSV data
• ✅ LaTeX references resolved
• ✅ No overfull boxes
• ✅ No Type 3 fonts
• ✅ qpdf structural check passes
• ✅ 21-page final manuscript
• ✅ Scientific source files unchanged during final vector-font preflight

Scope of the claims

This work does not claim:

• a break of DES
• a new generic Hellman theorem
• that DES itself is a random function
• a new distinguished-point technique
• a Java-versus-Rust performance advantage

The contribution is instead the combination of:

established TMTO theory + exact finite-state cryptographic mappings + matched random-map controls + complete functional-graph measurements + independent reproducible implementations

Historical provenance

The repository originates from an M2P SCCI master’s research project at Ensimag–UJF.

Original researchers:

• Abdourahmane Sakho
• Ali Mkhida
• Maad El Yadari

The original presentation is preserved unchanged as the canonical historical artifact.

The modern reproducibility layer, Rust implementation, experiment infrastructure, exact finite-state analysis, and current ePrint manuscript were developed during the repository modernization effort.

Publication status

The scientific content is frozen for the current ePrint candidate.

Remaining publication gates:

• Final visual PDF inspection
• Final authorship decision
• Public contact email
• Manuscript license
• Release/tag creation
• Cryptology ePrint submission

Do not merge this PR until the publication audit is complete.