Productive Value-Productive Power in a Dynamic Multi-Agent Productive System
About the White Paper
This paper reports the primary results of Asterion Benchmark V2, an internally constructed stress benchmark designed to exercise a large portion of the frozen PV-PP architecture simultaneously.
The benchmark models a tightly coupled productive system containing a manufacturer, suppliers, a customer, a lender, logistics, inventories, contracts, financing constraints, supplier deterioration and recovery, quality failures, alternate productive capacity, delayed field consequences, incomplete information, and persistent productive-power states.
The governing question is whether PV-PP can maintain coherent productive-system behavior under dynamic multi-agent stress, and how that behavior differs from specified competent comparison controllers operating over the same world, information boundary, action surface, and stochastic conditions.
The M14 Primary Ensemble
Principal Comparison
Against the balanced scalar receding-horizon controller B0, the PV-PP regime produced the following mean paired differences across 1,000 common seeds:
- +185.6 delivered usable units
- -178.4 physical-shortage unit-weeks
- +$380.6k terminal manufacturer cash
- -$71.3k adaptation expenditure
- +34.8 terminal S1 line productive power
- +45.1 terminal S1 quality productive power
PV-PP delivered more usable product in 999 of 1,000 matched stochastic scenarios and preserved greater terminal S1 line and quality productive power in 1,000 of 1,000.
Interpretation
The results establish strong behavioral separation within the frozen Asterion benchmark. Trace-supported analysis indicates that the separation is associated with differences in how the architectures sequence recovery, diversification, renegotiation, conservation, logistics, financing, and productive-capacity preservation.
The result is intentionally benchmark-relative. The paper does not claim universal superiority of PV-PP over scalar optimization, model-predictive control, multi-objective optimization, or other decision architectures.
Instead, the benchmark asks a narrower question: when a productive system begins to degrade, does explicitly representing productive capacity, adequacy, recovery paths, and future productive options change decision behavior in consequential ways?
Repository Contents
The associated GitHub repository is intended to preserve the publication-facing research record for the white paper, including the current manuscript and related publication artifacts.
The full Asterion benchmark execution package, benchmark-definition documents, banking records, and final M14 evidence are maintained in the main PV-PP benchmark repository.
Citation and Status
Status: Publication candidate / public research white paper.
Repository DOI, archival identifier, and final citation metadata can be added here after deposition to Zenodo, OSF, or another persistent archive.
Suggested title: Productive Value-Productive Power in a Dynamic Multi-Agent Productive System: The Asterion Benchmark V2.