tegwick
704272644c
Some checks failed
Test Suite / unit-tests (3.11) (push) Has been cancelled
Test Suite / unit-tests (3.12) (push) Has been cancelled
Test Suite / integration-tests (push) Has been cancelled
Test Suite / e2e-tests (push) Has been cancelled
Test Suite / performance-tests (push) Has been cancelled
Test Suite / code-quality (push) Has been cancelled
Test Suite / security-scan (push) Has been cancelled
Test Suite / test-summary (push) Has been cancelled
Add quality gate framework with schema validation (JSON Schema via jsonschema library), pattern validation (regex-based), multi-gate QualityValidator with SQLite persistence, HaltingPolicyEngine with budget/iteration/improvement checks, and RefinementLoop for iterative execute-validate-halt cycles. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
109 lines
3.5 KiB
Python
109 lines
3.5 KiB
Python
"""
|
|
Refinement loop for iterative quality improvement.
|
|
|
|
Implements FR-10: Halting and Refinement Policy
|
|
Execute → Validate → Halt or Refine cycle.
|
|
"""
|
|
|
|
from typing import Callable, List, Optional, Tuple
|
|
|
|
from markitect.prompts.quality.models import (
|
|
HaltDecision,
|
|
QualityPolicy,
|
|
RefinementResult,
|
|
ValidationResult,
|
|
)
|
|
from markitect.prompts.quality.policy import HaltingPolicyEngine
|
|
from markitect.prompts.quality.validator import QualityValidator
|
|
|
|
|
|
class RefinementLoop:
|
|
"""
|
|
Iterative refinement loop with quality gate checks.
|
|
|
|
Executes a cycle of: execute → validate → check halting → refine
|
|
until a halting condition is met.
|
|
"""
|
|
|
|
def __init__(
|
|
self,
|
|
validator: QualityValidator,
|
|
policy: QualityPolicy,
|
|
):
|
|
"""
|
|
Initialize with validator and policy.
|
|
|
|
Args:
|
|
validator: Quality validator with configured gates
|
|
policy: Halting policy configuration
|
|
"""
|
|
self.validator = validator
|
|
self.policy = policy
|
|
self.policy_engine = HaltingPolicyEngine(policy)
|
|
|
|
def run(
|
|
self,
|
|
execution_callback: Callable[[int, List[ValidationResult]], Tuple[str, str, str]],
|
|
artifact_id: str,
|
|
) -> RefinementResult:
|
|
"""
|
|
Execute the refinement loop.
|
|
|
|
The execution_callback is called each iteration with:
|
|
- iteration number (1-based)
|
|
- previous validation results (empty list on first iteration)
|
|
|
|
It should return a tuple of (run_id, content, artifact_id).
|
|
|
|
Args:
|
|
execution_callback: Callable that executes/refines and returns
|
|
(run_id, content, artifact_id)
|
|
artifact_id: ID of the artifact being refined
|
|
|
|
Returns:
|
|
RefinementResult with complete iteration history
|
|
"""
|
|
result = RefinementResult(iterations_run=0)
|
|
score_history: List[float] = []
|
|
prev_results: List[ValidationResult] = []
|
|
|
|
for iteration in range(1, self.policy.max_iterations + 1):
|
|
# Execute / refine
|
|
run_id, content, art_id = execution_callback(iteration, prev_results)
|
|
result.run_ids.append(run_id)
|
|
|
|
# Validate
|
|
current_results = self.validator.validate_artifact(
|
|
content, art_id, run_id=run_id if self.validator.db_path else None,
|
|
)
|
|
result.all_results.append(current_results)
|
|
result.iterations_run = iteration
|
|
|
|
# Evaluate halting
|
|
halting_record = self.policy_engine.evaluate(
|
|
results=current_results,
|
|
iteration=iteration,
|
|
score_history=score_history,
|
|
total_runs=len(result.run_ids),
|
|
)
|
|
|
|
current_score = self.policy_engine._aggregate_score(current_results)
|
|
score_history.append(current_score)
|
|
|
|
if halting_record.decision != HaltDecision.CONTINUE:
|
|
result.final_results = current_results
|
|
result.halting_record = halting_record
|
|
return result
|
|
|
|
prev_results = current_results
|
|
|
|
# Reached max iterations without explicit halt
|
|
result.final_results = prev_results
|
|
result.halting_record = self.policy_engine.evaluate(
|
|
results=prev_results,
|
|
iteration=self.policy.max_iterations,
|
|
score_history=score_history,
|
|
total_runs=len(result.run_ids),
|
|
)
|
|
return result
|