Skip to content

Core Mechanics: Election Loop, Quality, Fees, Rewards

This page focuses on the shared mechanics that surround the voting rule: participation, quality evaluation, fees, rewards, and the no-participation case.

For how ballots are aggregated into a collective ordering, see voting_rules.md. For the default outcome quality target, see puzzle_quality_gate_concept.md. For the later update of participation propensity, see participation_learning.md.

Election Step

Within an area, the election-time logic is:

  1. Agents decide whether to participate.
  2. Participants pay an election fee and submit ballots.
  3. The chosen voting rule aggregates the ballots into a full collective ordering.
  4. The top-ranked option becomes the elected color ordering.
  5. Outcome quality is evaluated against the active quality target.
  6. Rewards or penalties are applied to agents.
  7. Participation learning uses the resulting signals.
  8. Mutation influenced by the elected ordering is applied at the start of the next scheduler step.

This means elections are evaluated on the current election-time state, while grid mutation is lagged to the following step.

Quality, Fees, And Rewards

Participants pay an election fee proportional to their current assets:

  • fee = election_cost_rate * assets_pre

Outcome quality is represented by quality_distance. Its source depends on the active quality mode:

  • dist_to_reality if quality_target_mode="reality"
  • puzzle_distance if quality_target_mode="puzzle"

An election counts as good if:

  • quality_distance <= break_even_distance_common

Reward direction is then binary:

  • good decision: positive sign
  • bad decision: negative sign

Reward magnitude depends on how far a preference group is from the elected ordering:

  • group_dst_to_outcome = dist(preference_group, voted_ordering)

With reward_rate_personal as the common scale:

  • if the decision is good, reward factor = 1 - group_dst_to_outcome
  • if the decision is bad, reward factor = group_dst_to_outcome

The resulting reward amount is:

  • reward_amount = sign * reward_rate_personal * factor * assets_pre

After combining reward and fee, the per-election asset change is:

  • raw_delta_abs = reward_amount - fee
  • assets_post = max(0, assets_pre + raw_delta_abs)
  • delta_abs = assets_post - assets_pre
  • delta_rel = delta_abs / assets_pre if assets_pre > 0, else 0

This delta_rel is one of the main inputs into participation learning.

No-Participation Case

If no one participates in an area election, the model does not compute a new aggregate ranking. Instead, the previous elected ordering remains in force. If there is no previous elected ordering yet, the area falls back to the ordering implied by the current realized area distribution.

Even in that case, the model still updates quality-distance monitoring and group-distance tracking. What is missing is the new collective choice, the fee and reward cycle tied to new participation, and the usual participation update that depends on those election outcomes.

Main Knobs

The main parameters for this part of the model are:

  • distance_idx
  • election_cost_rate
  • reward_rate_personal
  • break_even_distance_common
  • quality_target_mode
  • puzzle_local_kappa
  • puzzle_shock_prob

Implementation Reference

The election pipeline around tallying, rewards, and quality tracking is implemented primarily in Area and VoteAgent. Supporting distance and aggregation helpers are documented in Utility Functions.