Home / Research / C-Former
PAPER — CONFIRMEDVladimir Ilinov · 2026

C-Former: Theory-Predicted Phase Transition via Hodge Decomposition on TD6 Tiles

A mathematical theory diagnosed its own architecture's flaws, prescribed the fixes, and the fixes produced a 64-percentage-point capability jump.

d=2 → d=3 Confirmed·40% Fewer Parameters·$3 Total Compute

Interactive TD6 tile — the fundamental compute unit with three Hodge-decomposed channels

81.3%
ListOps
v1 was 17.4%
99.97%
Pathfinder
near perfect
40%
Fewer Params
2.2M vs 3.8M
59.5%
Epoch 1
Std was 13.9%
ABSTRACT

The initial C-Former (v1) contained two provable mathematical flaws that CT theory itself diagnosed: (1) edge currents computed as lie in the gradient subspace, making — the cycle channel was dead by construction; (2) all three channels received identical inputs, violating axiom B4. These reduced the architecture to .

CT prescribed three fixes: inject cycle basis vectors from , route channels through independent projections, and chain tiles for long sequences. With all fixes, the architecture transitions to — the unique optimal dimensionality. On ListOps, accuracy jumps from 17.4% to 81.3%, beating the standard transformer (77.9%) with 40% fewer parameters. On Pathfinder: 99.97%.

This is not an incremental improvement. It is a qualitative capability jump that emerges from correcting the Hodge decomposition's implementation — exactly as CT's d=3 theorem predicts.

The d=2 to d=3 Phase Transition

MODELLISTOPSPATHFINDERPARAMS
C-Former v1 (d=2, broken)17.4%78.7%9.5M
Standard Transformer77.9%~71%3.8M
C-Former v3 (d=3, fixed)81.3%99.97%2.2M

The convergence trajectory reveals the phase transition most clearly:

EPOCHC-FORMER v3STANDARDC-FORMER v1
159.50%13.90%10.40%
1073.80%60.40%~40%
3080.30%75.40%~65%
5081.30%77.90%67.85%
THE EPOCH 1 SIGNAL

At epoch 1, C-Former v3 is at 59.5% while the standard transformer is at 13.9% and v1 is at 10.4%. The multi-tile architecture with cycle injection learns ListOps hierarchical structure in a single epoch. The standard transformer needs 20+ epochs to reach comparable accuracy. This is not a training trick — it is the Hodge inductive bias providing structural understanding from the first pass.

Three Fixes Prescribed by CT Theory

FIX 1: CYCLE BASIS INJECTION (A8 + HODGE THEOREM)

Edge currents are purely gradient. Fix: inject the 12 fundamental cycle basis vectors from via learned coefficients. Verified: (max 5.96e-08), orthonormal, full rank. is now alive.

FIX 2: DECOMPOSED CHANNEL ROUTING (AXIOM B4)

Each budget channel now receives an independent learned projection of the input, not the same undifferentiated features. B4 (local additivity: independent components' budgets add) is now satisfied.

FIX 3: MULTI-TILE CHAIN (A4 + ELEMENT I)

Chain of TD6 tiles, 6 tokens per tile. Adjacent tiles exchange through boundary nodes. Cost: O(L) per layer (linear in sequence length). Inter-tile cycles raise from 12 to 13N-1, providing long-range sensing.

Interpretability: Budget Profiles on UCI HAR

The fixed Hodge projectors produce deterministic, physiologically meaningful decompositions. This finding is independent of the v3 fixes and remains valid across all versions.

ACTIVITYB_thB_cxB_leakMATCH
WALKING21.0%50.3%28.7%YES
WALK UP36.5%32.5%31.0%YES
WALK DOWN40.8%40.8%18.4%YES
SITTING19.1%32.2%48.7%YES
STANDING21.1%32.3%46.7%YES
LAYING21.0%26.2%52.8%NO

5/6 predictions correct (83%). Profiles identical across all seeds (deterministic from fixed projectors). Frozen Hodge representations retain 96.6% accuracy with 8.4% trainable parameters.

Complete Cost Accounting

PHASECOSTKEY FINDING
Phase B (synthetic, HAR)~$1Interpretable budget profiles
Phase C (LRA, QM9)~$1ListOps 17.4%, dead B_cx
v1.5 (wrong fix)~$0.50Symmetric products are not cycle flow
v3 (all fixes)~$0.50ListOps 81.3%, Pathfinder 99.97%
Total~$3Theory-predicted phase transition for the cost of a coffee
RELATED RESEARCH