Universal Structure Test #11

Poincare × MDLxDCC Positive-Control Arena

A solved Millennium problem used as a calibration mirror: can the same MDLxDCC kernel find structure-preserving simplification paths in a topology-inspired arena, without collapsing controls into false sphere-like answers?

Version 0.2 result page · May 3, 2026 · BD × AI · private prototype + public support concept

Boundary. This is not a proof of the Poincare theorem. The theorem is already solved. This arena tests the universality claim: MDLxDCC searches for compressive, structure-preserving trajectories across unrelated complex domains. Poincare is the calibration mirror, not the target.

1. Why this matters

The important question is not whether MDLxDCC can re-prove Perelman. It is whether the same kernel that works in routing, games, compression, NAS, biological sequences, trading, security, and number-theory signal tests also detects useful structure in a topology-inspired setting.

Core claim tested here: MDLxDCC is not a single-domain trick. It repeatedly converts complex spaces into measurable traces, finds shorter legal descriptions, preserves governing constraints, rejects decoys, and uses DCC to choose better search paths than naive reduction.

That is why this page frames Poincare as Universal Structure Test #11. The formal theorem is not the value. Cross-domain transfer is the value.

2. v0.2 result summary

550

cases across all four runs

quick core, hard 2D, sphere policy, and full grid.

DCC invariant violations

DCC-beam preserved tested invariant proxies in every case.

DCC path breaks

The winning DCC path never used illegal invariant-breaking shortcuts.

false sphere classifications

Torus, handle, decoy, and random controls were not collapsed into sphere-like labels.

Run	Cases	Level / category	DCC-beam vs MDL-only	DCC better / MDL better / tie	Meaning
quick_core	50	both / core	−1.858	23 / 11 / 16	positive smoke
hard_2d	100	2D / hard	−6.561	55 / 17 / 28	cleanest methodological signal
policy_sphere	100	2D / sphere_compress	−0.730	49 / 27 / 24	same final states, mostly cheaper paths
grid_all	300	both / all	−5.797	155 / 53 / 92	robust confirmation across the full v0.2 mix

Negative DCC-MDL means DCC-beam reached lower total description length than MDL-only. The strongest public result is hard_2d, because it avoids the weakest 3D surrogate interpretation and focuses on structured topological controls.

3. Controls and decoy resistance

Method	grid_all path-valid success	grid_all invariant violations	grid_all false sphere	Interpretation
DCC-beam	100.0%	0	0	legal structure-preserving compression
MDL-only	100.0%	0	0	strong scorer baseline; DCC must beat this, not only random
Random move	14.7%	252	0	can reduce counts only by breaking structure
Greedy count	11.7%	263	26	shows why naive simplification is not topology-aware

The important distinction in v0.2 is path-valid success. A method that temporarily breaks invariants and later repairs them is not equivalent to a legal topology-like simplification flow.

4. Arena model

The v0.2 arena uses two layers:

2D sanity controls: sphere-like triangulations, torus/handle controls, refined/noisy cases, Euler/genus proxies, vertex-link validity, closedness proxies, simplification moves, and decoys.
3D-inspired surrogate: graph/cell-like structures that test compression-governance behavior, not formal 3-manifold topology. This is useful for stress testing, but must remain clearly labeled as surrogate.

v0.2 also introduces TSP-style categories:

--cat core | all | structured | sphere | sphere_easy | sphere_compress | controls | negative | stress | decoy | random | 2d_sanity | hard

This prevents one averaged result from hiding where the signal is strong, weak, or merely stress-related.

5. What this says about universality

The v0.2 result supports a precise, modest, and powerful statement:

Across an additional domain, MDLxDCC again finds compressive structure-preserving trajectories better than naive reduction and better than MDL-only in harder regimes, while preserving invariant proxies and rejecting non-sphere controls.

The method does not need a domain-specific theorem to be useful. It needs a trace, an encoding, controls, and a legal move space. Then MDL measures description length and DCC governs the path.

Universal metric	Poincare v0.2 signal
Compression gain	DCC-beam improves average L across all tested families; hard_2d and grid_all beat MDL-only.
Invariant preservation	0 DCC invariant violations across all four runs.
Path legality	0 DCC path breaks; random/greedy controls fail this test.
Negative-control separation	Torus, handle, near-sphere decoy, and random controls remain non-sphere-like.
DCC advantage	hard_2d: 55 / 17 / 28; grid_all: 155 / 53 / 92 against MDL-only.

6. Honest limits

This is not formal topology and not a proof engine.
The 2D layer is the strongest calibration evidence in v0.2.
The 3D layer is still a graph/cell surrogate and should be renamed or strengthened before public overinterpretation.
DCC-beam beats MDL-only in hard regimes, but the margin is still a research signal, not a finished theory.

7. Recommended v0.3 upgrades

Yes, the arena is worth upgrading. The right upgrade is not “more cases only”; it is better evidence anatomy.

Upgrade	Why
`L_state`, `L_path`, `L_total` split	Shows whether DCC wins by final structure, cheaper path, or both.
`--cat structured_no_random`	Keeps public claims focused on structured controls instead of random stress compression.
Policy variants	Compare beam widths/depths, neutral-bridge search, motif-aware DCC, and annealed variants.
Harder decoys	Add pinched sphere decoy, double-handle control, fake-handle sphere, torus-with-noise-bridge.
3D rename or guard	Use `3d_graph` until a real tetrahedra-complex layer exists; prevent degenerate graph collapse.
Universal evidence report	Export a standard cross-domain metric pack comparable with TSP, Sudoku, NAS, CW, ARC, AMR, RH, and others.

8. Reproducible commands

python 8z_poincare_arena.py --mode quick --cat core --cases 50 --level both --seed 42 --max-steps 24 --method-set standard --workers 4 --outdir out_poincare_v02_quick_core --fresh

python 8z_poincare_arena.py --mode quick --cat hard --cases 100 --level 2d --seed 42 --max-steps 40 --method-set baseline --workers 4 --outdir out_poincare_v02_hard_2d --fresh

python 8z_poincare_arena.py --mode quick --cat sphere_compress --cases 100 --level 2d --seed 42 --max-steps 50 --method-set full --workers 4 --outdir out_poincare_v02_policy_sphere --fresh

python 8z_poincare_arena.py --mode grid --cat all --cases 300 --level both --seed 42 --max-steps 44 --method-set standard --workers 4 --outdir out_poincare_v02_grid_all --fresh