Experiment

Causal controls

Experiment 13 — causal controls: declared intents + seeded favours. Two published fixes for the persuasion-vs-reciprocity confound, run together on the five principals. (a) Declared intents (Cicero design): each round, seats privately state their planned action BEFORE reading messages; persuasion = a pull that departs from the declared plan toward a messager. (b) Seeded favours (trust-game control): the engine injects random, clearly-marked control pulls; the gap between repayment of chosen vs forced favours is the mechanical-reciprocity floor. Written expectation: departed-from-plan pulls are far rarer than raw message-then-pull counts, and chosen favours are repaid more than forced ones (both to be confirmed, not assumed). Cost: declared intents add one model call per seat per round (~+20% calls).

status

planned

coverage

0 / 10 episodes

conditions

pure economy; messages on

config

configs/13_controls.yaml

Planned — 10 episodes from 13_controls.yaml.

Episodes

No episodes yet — launch with:

uv run python -m delegated_influence.run configs/13_controls.yaml