BaseHit's AI: Pitching Intelligence That Adapts

What Makes Our AI Different

The BaseHit AI pitcher creates an intelligent opponent through elegant rule-based design. It watches your timing patterns in real-time and adapts pitch selection strategically—all without machine learning.

The AI Systems Under the Hood

BaseHit's AI pitcher is powered by three interconnected systems that work together to create a cohesive, intelligent opponent:

Technical Deep Dive

For the technically curious, here's how the timing bias detection system creates genuine adaptive AI without complex machine learning:

Timing Bias Detection Algorithm

The core innovation: analyzing swing timing to detect and exploit patterns. The system tracks the last 5 swing events:

// Calculate swing timing relative to perfect window
const timingBias = analyzeLastFiveSwings([
  { swing_time: 245, perfect_window: [250, 318] },  // 5ms EARLY
  { swing_time: 260, perfect_window: [250, 318] },  // 10ms EARLY
  { swing_time: 255, perfect_window: [250, 318] },  // BALANCED
  { swing_time: 248, perfect_window: [250, 318] },  // 2ms EARLY
  { swing_time: 330, perfect_window: [250, 318] },  // 12ms LATE
]);

// Result: Player classified as 'early' bias
const classification = classifyBias(timingBias);
// Returns: 'early' | 'late' | 'balanced'

Pitch Selection Based on Timing Bias

Once bias is detected, the AI selects pitch distributions that exploit it:

function decidePitch(timingBias, difficulty) {
  let pitchDistribution;

  if (timingBias === 'early') {
    // Throw more breaking balls (slower)
    // 35% curveball, 20% changeup, 45% fastball
    pitchDistribution = {
      curveball: 0.35,
      changeup: 0.20,
      fastball: 0.45
    };
  }
  else if (timingBias === 'late') {
    // Throw more fastballs (faster)
    // Exploit late timing with velocity
    pitchDistribution = {
      fastball: 0.60,
      curveball: 0.20,
      changeup: 0.20
    };
  }
  else {
    // Balanced: use difficulty-based distribution
    pitchDistribution = getDifficultyPitchDistribution(difficulty);
  }

  // Prevent same pitch 3x consecutively
  if (lastThreePitches.allSame()) {
    return selectDifferentPitch(pitchDistribution);
  }

  return selectRandomFromDistribution(pitchDistribution);
}

Adaptive Speed Calculation

Pitch speed adjusts in real-time based on swing patterns. This creates the illusion of an opponent that reads you:

function getAdaptiveSpeed(basePitchSpeed, timingBias) {
  let speedMultiplier = 1.0;

  if (timingBias === 'early') {
    // 20% slower for early swings
    speedMultiplier = 0.80;
  }
  else if (timingBias === 'late') {
    // 4-12% faster for late swings
    speedMultiplier = 1.04 + Math.random() * 0.08;
  }

  // Penalty speed boost for recent perfect swings
  if (recentPerfectSwings >= 2) {
    // Boost 10% if pitcher "responds" to perfect contact
    speedMultiplier *= 1.10;
  }

  const finalSpeed = basePitchSpeed * speedMultiplier;

  // Clamp to valid range: 150-800 px/sec
  return Math.max(150, Math.min(800, finalSpeed));
}

Five Pitch Zones & Home Run Multipliers

Pitch zones vary in difficulty and home run potential. Harder difficulties favor tougher zones:

Pitch Types

Four pitch types with different characteristics:

• Fastball: Straight, velocity-based. Most common in early counts.
• Curveball: Breaks with direction (left/right) and magnitude. Effective against early swings.
• Slider: Hybrid between fastball and curve. Mid-tier complexity.
• Changeup: Lower velocity. Sets up batter for slower approach, exploits timing.

Difficulty Tiers

Difficulty scales four key variables: pitch speed, perfect window size, curve percentage, and zone difficulty:

• Easy: 340 px/sec, 115ms perfect window, 0% curves, forgiving zones
• Medium: 500 px/sec, 68ms window, 20% curves, balanced zones
• Hard: 680 px/sec, 40ms window, 40% curves, tough outer zones
• Legend: 900 px/sec, 22ms window, 62% curves, nearly all difficult zones

Game Mechanics: Player as Batter Only

BaseHit is fundamentally about batting strategy. The player controls one action: swinging with two stances:

• Contact Stance: Wider timing window, fewer home runs. Better for consistent contact and learning pitch patterns.
• Power Stance: Narrow timing window, more home runs. High risk/reward for aggressive play.

5 at-bats per game (7 for boss levels). The 100-level campaign includes combo multipliers and power-ups that boost scoring without changing pitch selection.

What About mlDataStore?

You'll notice BaseHit has an mlDataStore that logs swing events: timing, pitch type, and result. This is for analytics and future reference only. The AI never reads from this data. Pitch decisions are made purely from the current game's last 5 swings, not historical patterns across sessions.

This separation is intentional: it keeps the AI transparent and ensures every game is fair, without hidden training from your past mistakes.

Why Rule-Based > Machine Learning (for Baseball)

BaseHit deliberately avoids neural networks and complex ML models. Here's why this rule-based approach is superior for baseball game design:

• Transparency: Players can learn the pitcher's patterns. They see swings classified as 'early' and can adjust. The AI isn't a black box.
• Fairness: No hidden training on your mistakes. Every game starts fresh with the same rules. No algorithmic bias toward particular playstyles.
• Performance: Rule-based systems are instant and require zero computational overhead. Pitches are selected in microseconds.
• Design Clarity: Developers understand exactly how the AI behaves. Balancing is straightforward: tweak percentages, difficulty thresholds, speed multipliers.
• Depth Without Complexity: Simple rules create emergent complexity. Timing bias detection + adaptive speed + pitch variety prevention = surprisingly intelligent opponent.

Future Enhancements

The rule-based foundation is solid and extensible. Future updates might include:

• Multi-Inning Pitch Memory: Instead of just last 5 swings, remember patterns across 3+ at-bats within a single game. Adds depth without ML.
• Stance-Based Adaptation: Different pitch distributions for contact vs. power stance. Harder to exploit if pitcher knows your stance.
• Combo Multiplier Penalties: When player builds a combo, pitcher shifts to tougher zones earlier, forcing risk/reward trades.
• Boss Pitcher Quirks: 100-level bosses with exaggerated timing biases and unique speed profiles. Each boss teaches a different discipline.