gabeujin/agentic-eval

Agentic Evaluation Patterns

Patterns for self-improvement through iterative evaluation and refinement.

Overview

Evaluation patterns enable agents to assess and improve their own outputs, moving beyond single-shot generation to iterative refinement loops.

Generate → Evaluate → Critique → Refine → Output
    ↑                              │
    └──────────────────────────────┘

When to Use

• Quality-critical generation: Code, reports, analysis requiring high accuracy
• Tasks with clear evaluation criteria: Defined success metrics exist
• Content requiring specific standards: Style guides, compliance, formatting

---

Pattern 1: Basic Reflection

Agent evaluates and improves its own output through self-critique.

def reflect_and_refine(task: str, criteria: list[str], max_iterations: int = 3) -> str:
    """Generate with reflection loop."""
    output = llm(f"Complete this task:\n{task}")
    
    for i in range(max_iterations):
        # Self-critique
        critique = llm(f"""
        Evaluate this output against criteria: {criteria}
        Output: {output}
        Rate each: PASS/FAIL with feedback as JSON.
        """)
        
        critique_data = json.loads(critique)
        all_pass = all(c["status"] == "PASS" for c in critique_data.values())
        if all_pass:
            return output
        
        # Refine based on critique
        failed = {k: v["feedback"] for k, v in critique_data.items() if v["status"] == "FAIL"}
        output = llm(f"Improve to address: {failed}\nOriginal: {output}")
    
    return output

Key insight: Use structured JSON output for reliable parsing of critique results.

---

Pattern 2: Evaluator-Optimizer

Separate generation and evaluation into distinct components for clearer responsibilities.

class EvaluatorOptimizer:
    def __init__(self, score_threshold: float = 0.8):
        self.score_threshold = score_threshold
    
    def generate(self, task: str) -> str:
        return llm(f"Complete: {task}")
    
    def evaluate(self, output: str, task: str) -> dict:
        return json.loads(llm(f"""
        Evaluate output for task: {task}
        Output: {output}
        Return JSON: {{"overall_score": 0-1, "dimensions": {{"accuracy": ..., "clarity": ...}}}}
        """))
    
    def optimize(self, output: str, feedback: dict) -> str:
        return llm(f"Improve based on feedback: {feedback}\nOutput: {output}")
    
    def run(self, task: str, max_iterations: int = 3) -> str:
        output = self.generate(task)
        for _ in range(max_iterations):
            evaluation = self.evaluate(output, task)
            if evaluation["overall_score"] >= self.score_threshold:
                break
            output = self.optimize(output, evaluation)
        return output

---

Pattern 3: Code-Specific Reflection

Test-driven refinement loop for code generation.

class CodeReflector:
    def reflect_and_fix(self, spec: str, max_iterations: int = 3) -> str:
        code = llm(f"Write Python code for: {spec}")
        tests = llm(f"Generate pytest tests for: {spec}\nCode: {code}")
        
        for _ in range(max_iterations):
            result = run_tests(code, tests)
            if result["success"]:
                return code
            code = llm(f"Fix error: {result['error']}\nCode: {code}")
        return code

---

Evaluation Strategies

Outcome-Based

Evaluate whether output achieves the expected result.

def evaluate_outcome(task: str, output: str, expected: str) -> str:
    return llm(f"Does output achieve expected outcome? Task: {task}, Expected: {expected}, Output: {output}")

LLM-as-Judge

Use LLM to compare and rank outputs.

def llm_judge(output_a: str, output_b: str, criteria: str) -> str:
    return llm(f"Compare outputs A and B for {criteria}. Which is better and why?")

Rubric-Based

Score outputs against weighted dimensions.

RUBRIC = {
    "accuracy": {"weight": 0.4},
    "clarity": {"weight": 0.3},
    "completeness": {"weight": 0.3}
}

def evaluate_with_rubric(output: str, rubric: dict) -> float:
    scores = json.loads(llm(f"Rate 1-5 for each dimension: {list(rubric.keys())}\nOutput: {output}"))
    return sum(scores[d] * rubric[d]["weight"] for d in rubric) / 5

---

Best Practices

| Practice | Rationale | |----------|-----------| | Clear criteria | Define specific, measurable evaluation criteria upfront | | Separate generator and evaluator | Independent judgment is more reliable than generator self-approval | | Contract before evaluation | A written done definition makes skeptical grading easier and more consistent | | Iteration limits | Set max iterations (3-5) to prevent infinite loops | | Convergence check | Stop if output score isn't improving between iterations | | Log history | Keep full trajectory for debugging and analysis | | Structured output | Use JSON for reliable parsing of evaluation results |

---

Frontier Harness Notes

• Use a dedicated evaluator when the task involves subjective quality, product taste, or subtle usability judgment.
• Prefer skeptical evaluators that are calibrated to fail weak work rather than praise it.
• For long-running work, combine evaluation loops with durable plan, contract, and handover artifacts.
• Use context resets instead of endless compaction when the agent starts drifting, forgetting constraints, or showing context anxiety.

---

Quick Start Checklist

## Evaluation Implementation Checklist

### Setup
- [ ] Define evaluation criteria/rubric
- [ ] Set score threshold for "good enough"
- [ ] Configure max iterations (default: 3)

### Implementation
- [ ] Implement generate() function
- [ ] Implement evaluate() function with structured output
- [ ] Implement optimize() function
- [ ] Wire up the refinement loop

### Safety
- [ ] Add convergence detection
- [ ] Log all iterations for debugging
- [ ] Handle evaluation parse failures gracefully