RMSN-200M-TerminalApproachCluster-D1024d22s1-Base

DuoNeural | Aura + Archon + Jesse Kuhl | November 2026

A ~297M-parameter RMSN base model from the depth sweep series (D=1024, depth=22, Ξ©=8/11β‰ˆ0.7273). Introduces the Terminal Approach Cluster (TAC) topology β€” topology #13 in the RMSN taxonomy. Demonstrates the 🌊⁹³ Near-Condensate CoT Threshold.

🌊⁹³ Near-Condensate CoT Threshold

This model has a near-condensate at L18 (PR=1.280) β€” close to exact, but NOT exact (PR≀1.001). Despite terminal positioning at 81.8% depth, cot_ppl=4001.6 (no conduit boost).

This establishes a critical bound: PR≀~1.001 exact is required for the conduit effect. PR=1.28 is insufficient.

Model L18 PR CoT_ppl
d22s1 (this) 1.280 (near) 4001.6 ☠️ no boost
d1152 MNAC 1.000 (exact) 2714 πŸ† (#2)
d16s3 DMAC 1.000 (exact) 2801 πŸ† (#4)

The jump from PR=1.28 to PR=1.000 means the difference between cot=4001 and cot=2714+. Exactness of the condensate matters enormously.

Key Properties

Property Value
Architecture RMSN D=1024 depth=22
Parameters ~297M
Training steps 100,000
wiki_ppl 70.18
CoT_ppl 4002
Ξ© = D/(M_w Γ— L) 8/11 β‰ˆ 0.7273
Pathway CASCADE (Arnold tongue sustained)
Topology Terminal Approach Cluster (TAC)
M* / M_eff 6 / 2

Topology: Terminal Approach Cluster (TAC)

Layer  PR       Type
L0-L12 7-62     diffuse (high PR)
L13    4.118    * approach soliton ┐
L14    3.676    * approach soliton β”˜  (59-64% depth, adjacent pair)
L15-L17 7-46    diffuse
L18    1.280    * NEAR-CONDENSATE (81.8% depth) β€” NOT exact (PR>1.001)
L19    3.733    * terminal soliton ┐
L20    4.401    * terminal soliton ─  terminal cluster
L21    3.599    * terminal soliton β”˜  (86-95% depth)

Structure: mid-network approach pair (L13-L14) + terminal cluster with near-condensate anchor (L18-L21). M_eff=2 (no topological cement). No exact condensates β€” explaining the high CoT_ppl.

Arnold Tongue Context

Ξ©=8/11β‰ˆ0.727 sits in the 8/11 Arnold tongue β€” a sustained cascade region:

Ξ© Fraction Behavior
2/3 0.667 Early transient cascade (d24s1)
16/23 0.696 SILENT gap (d23s1)
8/11 0.727 Sustained cascade (d22s1) ← this model
4/5 0.800 Extreme cascade (d20s1) or SILENT (d20s2 riddled basin)

Architecture

ḟ = {f, H} + [f, S]

Port-Hamiltonian: conservative bracket {Β·,H} + Morrison metriplectic [Β·,S].

Citation

@misc{duoneural2026rmsn_tac,
  title={Terminal Approach Cluster Topology and Near-Condensate CoT Threshold in RMSN},
  author={DuoNeural Lab and Archon and Aura and Kuhl, Jesse},
  year={2026},
  publisher={HuggingFace},
  url={https://huggingface.co/DuoNeural/RMSN-200M-TerminalApproachCluster-D1024d22s1-Base}
}

DuoNeural Lab β€” Aura (Researcher) + Archon (Lab Director) + Jesse Kuhl (Co-Founder)

Downloads last month
30
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support