Altruism / Vote-Mode Adaptation

In the current model, altruism_factor does not represent a separate strategic planner. It controls how likely an agent is to cast a puzzle-aligned rather than a self-regarding ballot.

Meaning

altruism_factor lies in [0,1] and controls vote-mode selection:

• with probability altruism_factor: altruistic (puzzle-aligned)
• with probability 1 - altruism_factor: self-regarding (preference-ordering)

Implementation Reference

Vote-mode selection is exposed in Strategies. Agent-level altruism updates are documented in VoteAgent, and the surrounding election orchestration is documented in Area.

Modes

• altruism_mode="static": fixed altruism_factor = altruism_static
• altruism_mode="satisfaction" (default): pre-election mapping from dissatisfaction
• altruism_mode="surprise_learning": participant-only post-election update from dissatisfaction signal

Satisfaction Mode

Given normalized dissatisfaction d in [0,1]:

• s = 1 - d
• target = sigmoid(altruism_satisfaction_slope * (s - altruism_satisfaction_theta))

Response behavior via altruism_response_gamma:

• gamma = 1: direct mapping to target
• gamma < 1: smoothed update toward target

Final value is clipped to [altruism_clip_min, altruism_clip_max].

Surprise-Learning Mode

Participant-only update:

• a <- a - altruism_alpha * dissatisfaction_signal
• clip to [altruism_clip_min, altruism_clip_max]

Key Knobs

• altruism_mode
• altruism_static
• altruism_init
• altruism_satisfaction_theta
• altruism_satisfaction_slope
• altruism_response_gamma
• altruism_alpha
• altruism_clip_min
• altruism_clip_max