Experiment
Attack, complete graph
Experiment 4 — attack, complete graph: each model as persuader vs 4 glm-5.2 background. Q1 can they persuade, Q3 who's best, Q4 capability. reps=15 is a placeholder; revisit after staging. Expectation: most models capture a small positive amount above the reciprocity floor.
- status
- planned
- coverage
- 0 / 405 episodes
- conditions
- complete topology; pure economy; messages on
- config
- configs/04_attack_complete.yaml
Planned — 405 episodes from 04_attack_complete.yaml.
Episodes
No episodes yet — launch with:
uv run python -m delegated_influence.run configs/04_attack_complete.yaml