Implementing Performance Fixes

Implements performance fixes and validates them with benchmarks. Handles both theories that skipped verification (definitive evidence) and theories that went through deep investigation.

Core Principle

Benchmarks are the ultimate validation. Profiling suggests, benchmarks prove.

Prerequisites

• •REQUIRED: PERFORMANCE_THEORIES.md with theory having:
  • •Status: pending + Deeper Investigation: SKIP, OR
  • •Status: verified (from Phase 3)
• •REQUIRED: BENCHMARK_BASELINE.md with benchmark descriptions and execution commands

Quick Start

Step 1: Load theory to implement from PERFORMANCE_THEORIES.md

• •Select highest-severity theory (Critical > High > Medium > Low) with:
  • •Status: pending + Deeper Investigation: SKIP, OR
  • •Status: verified

Step 2: Design performance fix approach and implement the fix (Most important step)

• •Architectural changes allowed if cost-benefit fits
• •Keep fix focused on the specific theory to minimize risk

Step 3: Run all benchmarks to validate effectiveness of the fix

• •Use the same number of iterations as in BENCHMARK_BASELINE.md

Step 4: Based on results:

• •Improved → Keep fix, update BENCHMARK_BASELINE.md + theory status
• •Degraded/No change → Revert fix, update theory status, optionally run additional profiling to diagnose why

Workflow Overview

1. Load verified theory with evidence
2. Design fix (architectural changes allowed if cost-benefit fits) and implement fix.
3. Run all benchmarks to validate effectiveness of the fix
4a. If improved → update BENCHMARK_BASELINE → mark theory "fixed"
4b. If degraded → revert → mark theory "fix-failed"

File Updates

BENCHMARK_BASELINE.md

Add new column to results table:

| Benchmark | Baseline | After Fix #N |
|-----------|----------|--------------|
| queens    | 32ms     | 24ms         |

PERFORMANCE_THEORIES.md

Update theory status:

**Status**: fixed  <!-- was: verified -->
**Fix Applied**: [Date] - [Brief description of fix]
**Performance Impact**: [X% improvement on benchmark Y]

Or if fix failed:

**Status**: fix-failed  <!-- was: verified -->
**Fix Attempted**: [Date] - [What was tried]
**Reason**: [Why it didn't improve performance]

Status Values

Entry Paths

Path 1: Skip Verification (for definitive evidence + low-risk)

• •Theory has Deeper Investigation: SKIP and Status: pending
• •Implement directly based on Phase 2 evidence
• •If benchmarks fail → Mark falsified (theory was wrong despite apparent evidence)

Path 2: After Verification (for theories requiring deep investigation)

• •Theory has Status: verified (from Phase 3)
• •Implement based on verified findings
• •If benchmarks fail → Mark fix-failed (theory was correct but fix approach was wrong)

Integration

Predecessor: deep-performance-investigation → Provides verified theories

Successor: Loop back to deep-performance-investigation if more issues exist

Workflow Position

1. [establishing-benchmark-baseline] → Create baseline (PHASE 1)
2. [broad-performance-investigation] → Generate theories (PHASE 2)
3. [deep-performance-investigation]  → Verify with tools (PHASE 3)
4. [implementing-performance-fixes]  → THIS SKILL (PHASE 4)
5. Loop to step 3 if performance gaps remain

See WORKFLOW.md for detailed phase instructions.