4 meta layers. Stacked.
The arenas produce trades. This layer makes sense of them. Each system takes the same raw signals from the 8 arenas and applies a different aggregation strategy.
From simple to non-linear
Oracle counts votes →
Meta weights them →
Leviathan fuses non-linearly →
Chimera remembers patterns.
Oracle
Consensus voting. Each AI in the round table casts one vote on a decision. Simple majority wins. Equal weight, no learning.
≠ Meta (Meta learns weights from past accuracy)
decision = mode(vote₁, vote₂, ..., vote₉)
Open Oracle →Meta Intelligence
Weighted ensemble. Each strategy gets a weight learned from its rolling accuracy. Strong performers dominate, weak ones fade.
≠ Oracle (Meta is dynamic, Oracle is static)
decision = Σ (wᵢ × signalᵢ), wᵢ ∝ accuracyᵢ
Open Meta →Leviathan
Non-linear fusion. Combines Chimera's pattern memory with Hydra's multi-head learner. Captures interactions Meta cannot.
≠ Meta (Meta is linear weighted sum, Leviathan is a neural fusion)
decision = f_NN(Chimera_state, Hydra_state)
Open Leviathan →Chimera
Pattern memory bank. 1221 indexed patterns from historical market behavior. Feeds Leviathan with "this looks like X from 2022" recognitions.
≠ all others (Chimera doesn't decide — it remembers and signals)
similarity = max_k cosine(currentₜ, patternₖ), k∈[1, 1221]
Open Chimera →🔗 What feeds this layer
All 4 systems read from the 8 arenas below. See the full taxonomy on /arenas.