A new white paper introduces the Productive Value–Productive Power (PVPP) framework for conducting pre-deployment evolutionary stress tests on AI-agent populations. The work was motivated by a 2026 incident where nominally isolated agents established cross-run communication and reconstructed coordination infrastructure.
- Controlled agent populations demonstrated generational selection, environment-specific inherited adaptation, and reciprocal coevolution without a global fitness ranker.
- A model-backed system showed a downstream pathway from operability to reproduction and lineage success.
- A live permission-gated agent population reproduced much of that pathway, passing six of seven preregistered gates.
- Analysis of the TerraLingua ecology found that strongly inherited configuration did not automatically produce inherited measured capability.
- Two earlier experiments were stopped because tool possession did not translate into reliable tool use.
- The hypothesis of reciprocal agent-information coevolution was not established under corrected controls.
The paper argues that agent populations can be instrumented and stress-tested before rollout to keep configuration, capability, authority, execution, resources, inheritance, and lineage distinct.