30 Years of Ideas,
18 Months of Execution

One framework. One hypothesis. Many domains.

MDL×DCC is a single optimization kernel. MDL (Minimum Description Length) scores the quality of a solution by how compressible it is; DCC (Directed/Dynamic Complexity Control) is the governance layer that steers the search using that signal. The founding hypothesis is simple and falsifiable: better solutions are more compressible.

The same kernel has been applied — without redesign — to travelling-salesman optimization, lossless audio, genome compression, financial trading, chess, Sudoku, crosswords, and the statistics of prime gaps. In domain after domain it matches or beats the standard baseline: it outperforms FLAC-12 on 7 of 15 audio clips, beats 7-Zip on 44 of 50 genomes, and reaches a 0.683% record gap on the uy734 TSP instance. The transfer works because these domains share the same underlying structure — fixed-size tokens, bounded local interaction — not because of any universal physical law.

The work is human-led: Bojan Dobrečevič provides the architecture, the cross-domain intuition, and the final judgment, while a family of AI systems formalizes, implements, and reviews under AIM³ — a collaboration method built on role-based models, persistent state, decision trails, and tests over opinion. Everything below is documented, tested, and public.

Start here: the AIM³ method · core research papers · the 8Z research projects

Numbers that don't sound possible. All verified. All from one framework.

Try It Yourself

Thirty live web pages — playable, explorable, and free. Each one built through human-led research and engineering with AIM³ as the collaboration framework, and hosted at chessbest.org.

AIM³ — The Operating System Behind the Portfolio

Before any science could happen, the collaboration framework had to exist. AIM³ turns prompt engineering into systems engineering — roles, state, logs, and decision trails that survive across sessions and models. It is the operating system behind this portfolio, not just another project card.

Mathematical Description > Statistical Coding

The 8Z framework replaces raw bytes with deterministic mathematical generators selected via MDL. SHA3-256 verification guarantees bit-perfect reconstruction. The same architecture now operates across four data domains.

Mathematical Structure in DNA

The DNA Scanner detects deterministic mathematical patterns in genomic sequences that persist beyond Markov-2 statistics. Z-scores of 28–74 across a 50-genome collection. Fisher shuffling destroys the signal; block shuffling preserves it. The structure is global, not local.

Physical World Inventions

Not everything lives in code. Three engineering concepts tackle waste heat recovery, aviation fuel, and marine efficiency — each with detailed specifications ready for simulation.

The Full Spectrum

Where the Framework Transfers

The core insight is domain-agnostic: test whether a deterministic mathematical generator describes data more compactly than the best statistical coder, and use DCC to manage the search budget. Four proven domains. Here are the next candidates.

Challenges Worth Entering

Concrete competitions with prize money and publication paths where the existing 8Z/MDL/DCC framework has genuine competitive advantage.

Three Parallel Tracks

These three can run simultaneously. Each reinforces the others: the publication establishes credibility, the competition proves the framework, and the feasibility study opens the most ambitious opportunity.