# MDLxDCC.org

> 8Z–MDL×DCC is a recursive, multi-level and self-selecting architecture tested through working systems, frozen evidence and adversarial arenas.

Version: index v0.2.11 PRE-REVIEW · ARENA FOUNDRY PUBLIC SEED
Date: 2026-09-10

## This is not a concept portfolio

BD × AI has already built executable solvers, codecs, governors, interfaces, research pipelines and deterministic arenas across compression, optimization, chess, Sudoku, DNA, trading and AI tooling.

Formal peer review and independent third-party reruns are not yet complete and are not claimed. That limits external confidence. It does not erase the code, internally measured effects or right to state them plainly.

## Evidence-challenge review protocol

1. Freeze the website-only provisional verdict.
2. Identify the exact runnable package or test most likely to falsify the strongest remaining criticism.
3. Inspect or execute the original code, frozen inputs, manifests and run evidence.
4. Label what was only read, self-tested, executed or reproduced.
5. Publish a claim-by-claim and score-by-score verdict delta.
6. Preserve positive, negative and unchanged conclusions.

The target is visible evidence-induced correction, not compulsory praise. A large upward revision is informative when the first judgment assumed there was no real implementation. A criticism that survives execution is equally informative.

## Core grammar

- 8Z / Generator: creates alternative mathematical representations, operators, models, routes and hypotheses.
- MDL / Judge: selects the shortest complete account, including model, residual, decoder, exceptions and declared cost.
- DCC / Governor: Dynamic Complexity Controller. Allocates compute, stabilizes productive focus, changes scale and reopens search when evidence changes.
- ssMDL×DCC: sensors, control laws, topologies and governors compete under the same MDL criterion.

Short form: Generator → MDL → DCC.
Actual topology: bidirectional, recurrent, nested, multi-scale and self-selecting.


## Executable AI work cell: MAL × 8zCoding

8zCoding is the candidate execution organ of a future AI8 Work Environment. MAL prepares one canonical coding task and gathers blind, fresh, peer-separated multi-provider proposals. 8zCoding later ingests anonymous candidate bytes, builds and runs them in an isolated local environment, applies visible and hidden verifiers, measures runtime/memory/artifact cost, and returns bounded evidence for owner-private repair.

Current status:

- A0/A1 N-Queens verifier: PASS; Python/C++ reference alignment; 11 deliberately wrong candidates rejected; actual Windows 11 + MSVC evidence in lineage.
- A2/A3/A5 replay: engineering PASS.
- A4 opportunity/oracle result: provisional because the current oracle selects final utility before the declared primary AUC-U metric.
- DCC-R utility: not promoted.
- LZ/process-compression mechanism: not promoted.
- MAL v2.0 RC1 Candidate Forge: user-reported build in progress under an interface-first shared conformance-kit contract.
- A6 live multi-provider coding: not yet run.

This negative-capable status is a feature: the verifier, ordinary controllers and DCC are separate, and the favored controller is allowed to lose. If the complete MAL × 8zCoding loop beats strong single-agent, Best-of-N and ordinary-controller baselines per total cost, it becomes the first measurable AI8 work cell.

8zCoding page: https://www.mdlxdcc.org/crp/8zCoding.html

## AI8 Arena Foundry — governed self-research proposal

Arena Foundry is a new experimental architecture for reducing mechanical human orchestration in long-horizon AI-assisted research while preserving the research mission, permissions, evidence, provenance and human intervention at high-leverage decisions.

Canonical loop:

seed → research constitution → independent candidate architectures → mission guard → build → validator → run → runtime evidence → ablation/critique → promote/mutate/branch/park/reopen → next generation

Two lanes are intentionally separated:

- 8Z Native / Clean Room: original mechanisms, BD seeds, AI-derived ideas, fundamentals and cross-domain transfers without silently collapsing into the dominant domain solution.
- World Control: strong external methods used as comparators, baselines and falsifiers.

The motivating failure mode is mission drift: an arena can compile, pass tests and run successfully while the mechanism it was meant to test contributes effectively 0%. Arena Foundry therefore treats mission failure as distinct from implementation failure and requires matched ablation where possible.

Arena Foundry is an architecture proposal. It is not evidence of autonomous AGI, an autonomous scientist, consciousness or recursive self-improvement advantage.

Arena Foundry page: https://www.mdlxdcc.org/ai8/ai8_arena_foundry

## Strong current ChessDCC anchor

A source-bound reconstruction from archived raw outputs confirms a measured internal discovery:

- 2,055 exactly game-deduplicated near-tie decisions;
- ChessDCC agreement with the current candidate-specific ChessDB-PV endpoint judge: 46.23%;
- immediate Raw agreement with the same judge: 20.34%;
- ratio: 2.273×;
- DCC-only versus Raw-only correct disagreements: 715 : 183;
- 153 unique game outputs: DCC better in 130, Raw better in 7, equal in 16;
- D40 and D80 semantically identical across 251 unique games.

Current discovery statement:

> Evaluation trajectories contain usable decision information after an immediate scalar near-tie loses resolution.

Current boundary:

- the historical ChessDCC implementation is a flat fixed-weight trajectory scorer, not yet recursive MDL×DCC;
- Raw did not receive equal lookahead;
- the current judge is the endpoint of the same candidate-specific ChessDB PV;
- a bug-fixed replay, equal-lookahead controls, independent judge and paired same-engine play remain next-stage tests;
- no Elo gain is currently claimed.

Targeted public prior-art search found related stability signals in engine time management and principal-variation entropy research, but no exact published match to the complete ChessDCC-specific conjunction. This bounded statement remains open to dated earlier public prior art.

ChessDCC audit: https://www.mdlxdcc.org/c/ChessDCC_Evidence_Audit_20260904.html

## Other main public stress-test anchors

- TSP: exact qa194 optimum; nu3496 verified below 1% above the certified optimum. Controller attribution and stronger matched baselines remain open.
- Lossless compression: scoped bit-perfect wins; strongest-codec corpus consolidation remains open.
- DNA: order-sensitive signals under tested nulls plus a reported generator cull after artifact analysis; biological interpretation remains open.
- Meta-DCC: reported self-calibration mechanism result; stronger matched-controller tests remain open.
- Sudoku/process control: R6 HF3 HF1 is a delivered Python arena with checked proofs, Proof Atlas, compact recovery and FAST LIVE. Its new development study separates MDL prediction, DCC allocation and frozen transfer: eight main arms, two puzzle profiles and a bounded Context UCB diagnostic. The delivered program passed 402 tests (0 failures, 0 errors, 1 skip) on Linux/Python 3.12. The first reviewed long-run extract contains qualification only; no MDL/DCC-specific advantage is established. Current overview: https://www.mdlxdcc.org/crp/AI8_Sudoku_R6.html
- Historical R3L1 snapshot (4 September 2026): the frozen large-confirmation design covers 10,000 puzzles × 12 policies. First exact-byte snapshot: 12,043 valid rows, 1,002 complete matched blocks, zero correctness failures. The provisional system/lookahead signal is positive while residual controls currently lead MDL twins; final promotion verdicts are withheld.

These are heterogeneous stress tests of the shared selection-governance grammar, not independent replications.

## Claim boundaries

- Does not prove P=NP.
- Does not establish universal superiority.
- Functional governance does not establish phenomenal consciousness.
- Modeled economic upside is not demonstrated savings.
- Missing external replication limits external confidence; it is not evidence that working code or internal measurements do not exist.

## Important routes

- Home: https://www.mdlxdcc.org/
- AI8 hub: https://www.mdlxdcc.org/ai8/
- AI8 ArenaLoop: https://www.mdlxdcc.org/ai8/ai8_arenaloop
- AI8 Arena Foundry: https://www.mdlxdcc.org/ai8/ai8_arena_foundry
- ChessDCC source-bound audit: https://www.mdlxdcc.org/c/ChessDCC_Evidence_Audit_20260904.html
- 8zCoding executable AI work cell: https://www.mdlxdcc.org/crp/8zCoding.html
- Sudoku R6 HF3 HF1 arena and causal study: https://www.mdlxdcc.org/crp/AI8_Sudoku_R6.html
- Sudoku R3L1 first live snapshot: https://www.mdlxdcc.org/c/Sudoku_R3L1_Live_Snapshot_20260904.html
- Arena programme: https://www.mdlxdcc.org/crp/MDLxDCC-Arenas.html
- Technical outreach: https://www.mdlxdcc.org/c/MDLxDCC_Outreach.html
- 8Z compression: https://www.mdlxdcc.org/8Z/8Z_compression.html
- Domain map: https://www.mdlxdcc.org/crp/MDLxDCC.html
- AI8 architecture: https://www.mdlxdcc.org/AI8/
- Mind of Minds: https://www.mdlxdcc.org/c/mind-of-minds.html
- Human–AI method: https://www.mdlxdcc.org/crp/AI8_Reasoning.html
- Public prior-art search log: https://www.mdlxdcc.org/c/MDLxDCC_Public_Prior_Art_Log.html
- Full work map: https://www.mdlxdcc.org/w/

## Terminology reservation

“Digital Claustrum Controller” is reserved as a possible future name for a separately built and tested AI8 implementation. It is not the current meaning of DCC.

## Contact

Bojan Dobrečevič (BD) · bd@siol.net