Skip to content

SCORING.md

Latest commit

 

History

History
67 lines (48 loc) · 2.05 KB

SCORING.md

File metadata and controls

67 lines (48 loc) · 2.05 KB

Scoring And Stability

Scoring is provisional. It exists to make generated runs inspectable, not to define final player-facing level scores.

Current Score Formula

ScoreState.TotalScore is currently:

ReactionScore
+ FlowDiversityScore
+ SettlementScore
+ ResilienceScore
+ RepairScore
+ AutonomyScore
- StrainPenalty
- HoardingPenalty
- DeadLoopPenalty

Only these components are active in milestone 1:

  • ReactionScore: +10 for each cell reaction.
  • FlowDiversityScore: +2 for each distinct resource seen in the bounded flow event buffer.
  • SettlementScore: +1 for each reaction event in the bounded event buffer.
  • StrainPenalty: total accumulated strain across all cells.

The other score fields exist as placeholders for later design work.

Strain Penalty

Each cell has a StrainState:

UnmetNeedTicks
+ FailedSwapCount
+ SourceBlockedTicks
+ OverCapacityPressureTicks

StrainPenalty is the sum of that total across every cell. A large negative score usually means many cells are idle with unmet needs, blocked production, full source slots, or overflow pressure.

Generated Stability Signal

Generated scenario summaries report stable signal. This is not a win condition. It is a search heuristic for finding interesting backend configurations.

A generated run currently needs all of these to count as stable:

  • at least one swap in the last summary window,
  • at least one reaction in the last summary window,
  • last-window reactions not collapsed below a quarter of the first-window reactions,
  • active cells in the last window at least max(3, cellCount / 5),
  • cells with strain no more than half of all cells.

This deliberately rejects tiny isolated loops. For example, a 100-cell map where only two cells keep cycling should not be considered stable.

Current Limitation

The score is cumulative and blunt. It is useful for backend search, but final game scoring should separate:

  • settlement,
  • breadth of participation,
  • sustained flow,
  • repair,
  • low strain,
  • efficient use of limits,
  • visual chain-reaction beauty.