Winning Avg Corewars

Winning Avg CoreWars

Overview

This skill enables systematic development of CoreWars warriors that must achieve specific win rate thresholds against multiple diverse opponents. It provides a structured methodology for analyzing opponents, designing effective warriors, and iteratively improving performance through hypothesis-driven testing.

Key Concepts

CoreWars is a programming game where warriors (programs written in Redcode) compete in a shared memory space called the core. Common warrior archetypes include:

• Scanners: Search for opponents and attack when found
• Bombers: Systematically drop DAT bombs across the core
• Replicators (Paper): Copy themselves throughout memory
• Vampires: Convert opponent processes to work for them
• Imps: Simple MOV 0,1 programs that spread through memory
• Stones: Bombing patterns with specific step sizes

Workflow

Phase 1: Opponent Analysis

Before writing any code, thoroughly analyze each opponent warrior:

1. Read each opponent file completely - understand every instruction
2. Identify the archetype - what strategy does it use?
3. Extract key parameters:
   • Step sizes (affects coverage patterns and vulnerabilities)
   • Memory access patterns
   • Process count and management
   • Defensive mechanisms
4. Document vulnerabilities - what attacks would disrupt this warrior?
5. Calculate mathematical properties:
   • For bombers: coverage rate = core_size / step_size
   • For replicators: replication speed and spread pattern
   • For scanners: scan density and blind spots

Reference references/corewars_fundamentals.md for mathematical relationships between step sizes, coprime values, and coverage patterns.

Phase 2: Strategy Design

Design strategy based on opponent analysis, not trial and error:

1. Categorize opponents by weakness:

   • Fast bombers vulnerable to: quick replicators, distributed processes
   • Replicators vulnerable to: precise scanners, widespread bombing
   • Scanners vulnerable to: decoys, position-independent code
   • Imps vulnerable to: imp gates, specific bombing patterns

2. Choose primary strategy based on which approach counters the most opponents

3. Design countermeasures for each opponent type:

   • Document expected mechanism of winning
   • Identify which component defeats which opponent

4. Consider hybrid approaches when single strategies cannot meet all thresholds

Phase 3: Parameter Selection

Select parameters based on mathematical analysis, not arbitrary values:

1. Step sizes: Choose coprime values relative to core size for optimal coverage
2. Gate positions: Calculate based on opponent scanning patterns
3. Bombing intervals: Match or counter opponent timing
4. SPL count: Understand trade-offs:
   • More processes = better survival against bombing
   • More processes = slower individual execution
   • Document which opponents benefit from each configuration

Phase 4: Systematic Testing

Establish a rigorous testing methodology:

1. Create a test script that runs all opponents automatically (see scripts/test_warrior.sh)
2. Test each component in isolation before combining
3. Record results systematically:

   Version | Opponent | Wins | Losses | Ties | Notes
   

4. Form hypotheses before changes: "Increasing step to X should improve stone performance because..."
5. Test one variable at a time to understand causation

Phase 5: Iterative Improvement

Follow a hypothesis-driven improvement cycle:

1. Identify the weakest matchup - focus on the opponent with the worst performance
2. Hypothesize the cause - why is this matchup failing?
3. Design a targeted fix - what specific change addresses the hypothesis?
4. Predict the outcome - what should improve? What might regress?
5. Test and measure - did results match predictions?
6. If predictions were wrong, revise understanding before trying another change

Verification Strategies

Win Rate Verification

• Run sufficient matches (100+ per opponent) for statistical significance
• Track variance across runs - inconsistent results indicate non-deterministic factors
• Verify with the exact testing harness that will be used for final evaluation

Component Verification

• Test scanner component: Does it find targets at various positions?
• Test bomber component: What is actual coverage over N cycles?
• Test replicator component: How many copies survive after N cycles?
• Test imp backup: Does it activate when main warrior is killed?

Regression Testing

• After any change, test ALL opponents, not just the target
• Document which changes help which opponents and why
• Maintain a "best known version" for each opponent matchup

Common Pitfalls

Trial-and-Error Without Understanding

Problem: Making random parameter changes hoping something works. Solution: Every change should have a hypothesis explaining why it should help.

Neglecting Difficult Opponents

Problem: Focusing only on opponents close to threshold while ignoring others. Solution: Address each opponent systematically; a warrior that fails against one opponent type may need fundamental redesign.

Conflating Correlation with Causation

Problem: "3 SPLs helped paper, so 3 SPLs is better." Solution: Understand WHY 3 SPLs helps paper (more surviving processes) and whether that trades off against other opponents.

Abandoning Promising Approaches

Problem: When V12 gets 70/75 wins, moving to completely different V13. Solution: When close to threshold, explore variations of the successful version.

File Format Issues

Problem: Redcode has strict formatting requirements; extra spaces or tabs cause syntax errors. Solution: Test warrior loads correctly before testing against opponents. Use pmars -b to check for assembly errors.

Ignoring Defensive Design

Problem: Pure offense with no consideration for surviving attacks. Solution: Include defensive elements:

• Position-independent code when possible
• Imp backup as last resort
• Process distribution to survive partial kills

Single Imp Backup

Problem: Adding mov 0, imp provides minimal backup - single imp easily caught. Solution: Use imp rings or multiple distributed imps for effective backup.

Resources

scripts/test_warrior.sh

A template script for systematic warrior testing. Modify the opponent list and paths for the specific task environment.

references/corewars_fundamentals.md

Mathematical foundations including:

• Step size selection and coprime relationships
• Core coverage calculations
• Process timing and execution order
• Classic warrior archetype characteristics