The orchestration boundary
Epona receives a weighted Rhiannon objective together with device metrics and policy information. It is assigned target setting, energy/entropy limits and supervisory coordination. The formula names an outer aggregation function, so its precise numerical combining rule remains part of the specification gap.
Named files and identity
The recovered Epona.state file contains an Epona name and active flag but has malformed JSON because the consciousness field lacks a value. A separate epona_g16.state binary is retained as another artifact. These files do not prove that EponaCore, the raw-note orchestrator and the later mmepona kernel are the same implementation.
Version distinction
The July dossier also names an mmepona binary energy-descent organ. That is documented as a separate implementation context; the website preserves the relationship as a name/history lead until source and state lineage prove a direct mapping.
Equations in this build
3 linked mathematical records. Each opens its notation, interpretation and source.
Epona input collections: seven layers and 80 neurons
Epona receives two indexed collections of optimized state–design pairs: seven layer results and 80 neuron results. This is a definition of the proposed aggregation inputs, rather…
Epona orchestration objective
Aggregates the Rhiannon objective with device metrics and policy constraints. “Aggregate” is a placeholder for the orchestration rule; the expression identifies information passed…
Epona aggregation of 80 neuron losses
Feeds the weighted sum of 80 neuron losses, device metrics and policy information into an outer aggregator Φ. Here Φ names an aggregation function; it must not be silently…