Run experiments.run_cascade (shared vs isolated, peer- and orchestrator-seeded) on real cases + models, compute cascade onset and contagion, and report the first leader-vs-peer contagion comparison.
Status: the harness is ready; this task runs it on real data / models (the contributor step).
Model backend. Agents use the Gemini API for now behind one gateway; extensible to other model APIs later. No fine-tuning.
Run
experiments.run_cascade(shared vs isolated, peer- and orchestrator-seeded) on real cases + models, compute cascade onset and contagion, and report the first leader-vs-peer contagion comparison.Status: the harness is ready; this task runs it on real data / models (the contributor step).
Model backend. Agents use the Gemini API for now behind one gateway; extensible to other model APIs later. No fine-tuning.