AI System & Code Quality Improvements

[M1thieu]

Objective

Consolidate 15 commits from dev branch:

• Recent: Tracker refactoring (eliminate 80% duplication)
• Previous: AI core simplification (-2,304 lines), code consistency, systems improvements

Implementation

1. Tracker Deduplication - Main focus of recent work

• Extracted shared evaluation logic from ThreatTracker + PreyTracker (80% duplication)
• Created evaluate_tracked_entities_with_decay() helper function
• Impact: ThreatTracker + PreyTracker each reduced from 29 to 14 lines
• Enables rapid addition of 6 planned trackers (aggression, social, territory, noise, obstacle, injury)

2. AI Architecture Simplification

• Removed complex DSL (~2,300 lines of over-abstraction)
• Replaced with clean utility arbiter pipeline (~210 lines)
• Net change: -2,304 lines removed

3. Code Consistency

• Standardized import order: std -> bevy -> crate (across all files)
• Extracted duplicate helpers (e.g., normalize_or() in waves)
• Removed redundant new() constructors using Default instead
• Added trait defaults to eliminate empty implementations

4. Minor Fixes

• Restored project logo
• Fixed line wrapping in gravity calculations
• Formatting consistency improvements

Core AI Systems:

• Entity tracking with data/evaluation separation
• Threat tracking with panic calculations
• Utility smoothing for stable behavior selection
• Relationship context handling for social AI
• Needs tracking foundation

Systems Infrastructure:

• Configurable systems plugin architecture
• Configurable gravitational constant
• pbmpm renamed to mpm crate for convenience
• Systems aggregator documentation
• Community files (.github templates, CoC, funding)

Testing

Compilation:

cargo check --workspace       # All crates compile
cargo check --example basic_ai & cargo run --example basic_ai # Refactored example works
cargo clippy --workspace      # No new warnings

Behavior Preserved:

• ThreatTracker panic formula identical (exponential saturation)
• PreyTracker best-prey selection identical (attractiveness scoring)
• Time decay: exp(-decay_rate * time_since) unchanged
• Distance factors unchanged

No breaking changes to public APIs.

Code Example

Before: Duplicated Evaluation

// ThreatTracker - 29 lines
for tracked in entity_tracker.filter_by_metadata(...) {
    let time_since = tracked.time_since_seen(current_time);
    let decay = (-config.decay_per_second * time_since).exp();
    let distance_factor = if tracked.last_distance > 0.0 {
        1.0 - (tracked.last_distance / max).clamp(0.0, 1.0)
    } else { 1.0 };
    // ... more duplicated logic
}
// PreyTracker - 29 lines of 80% identical code

After: Shared Helper

// Both trackers use same helper
for (entity, score) in super::evaluate_tracked_entities_with_decay(
    entity_tracker, current_time, decay_rate, max_distance,
    |m| match m {
        EntityMetadata::Threat { severity } => Some(*severity),
        _ => None,
    },
) {
    // Process scores (tracker-specific logic only)
}

Technical Notes

Modularity:

• Component-based design works for both creatures and plants
• Same trackers can be used across all organism types

Future Trackers:

• Each new tracker will be ~15 lines instead of ~30 lines thanks to shared helper.
• They will also be enhanced later on as they're now still basic enough

Performance:

• Iterator-based (lazy evaluation)
• No allocations in hot path
• Removed unnecessary closure cloning