Files
beaver_project/app-instance/backend/beaver/skills/learning/eval.py

326 lines
13 KiB
Python

"""Lightweight replay/eval reports for skill drafts."""
from __future__ import annotations
from uuid import uuid4
from beaver.engine.context import SkillContext
from beaver.engine.providers import ProviderBundle
from beaver.memory.runs import RunMemoryStore
from beaver.memory.skills import SkillDraftEvalReport, SkillLearningCandidate
from beaver.skills.learning.case_selection import select_replay_cases
from beaver.skills.learning.preservation import check_preservation
from beaver.skills.learning.replay import ReplayArmRequest, ReplayRunner
from beaver.skills.learning.surrogate import SurrogateToolEvaluator
from beaver.skills.specs import SkillDraft
class SkillDraftEvaluator:
"""Builds a bounded eval report without writing user-visible sessions."""
def __init__(
self,
run_store: RunMemoryStore,
*,
surrogate_evaluator: SurrogateToolEvaluator | None = None,
) -> None:
self.run_store = run_store
self.surrogate_evaluator = surrogate_evaluator or SurrogateToolEvaluator()
async def evaluate(
self,
*,
candidate: SkillLearningCandidate,
draft: SkillDraft,
provider_bundle: ProviderBundle | None,
replay_runner: ReplayRunner | None = None,
) -> SkillDraftEvalReport:
if provider_bundle is None or provider_bundle.main_provider is None:
return self._skipped(candidate, draft)
runs = self.run_store.list_runs()
replay_cases = select_replay_cases(candidate, runs)
if replay_runner is not None and replay_cases:
return await self._evaluate_replay(
candidate=candidate,
draft=draft,
replay_cases=replay_cases,
provider_bundle=provider_bundle,
replay_runner=replay_runner,
)
return self._evaluate_heuristic(candidate, draft, runs)
def _evaluate_heuristic(
self,
candidate: SkillLearningCandidate,
draft: SkillDraft,
runs: list,
) -> SkillDraftEvalReport:
runs_by_id = {record.run_id: record for record in runs}
cases: list[dict] = []
for run_id in candidate.source_run_ids[:8]:
record = runs_by_id.get(run_id)
if record is None:
continue
baseline = _score_from_validation(record.validation_result, record.success)
candidate_score = _candidate_score(baseline, draft)
cases.append(
{
"run_id": run_id,
"session_id": record.session_id,
"baseline_score": baseline,
"candidate_score": candidate_score,
"delta": round(candidate_score - baseline, 4),
}
)
if not cases:
cases.append(
{
"run_id": "",
"session_id": "",
"baseline_score": 0.75,
"candidate_score": _candidate_score(0.75, draft),
"delta": round(_candidate_score(0.75, draft) - 0.75, 4),
}
)
baseline_avg = sum(item["baseline_score"] for item in cases) / len(cases)
candidate_avg = sum(item["candidate_score"] for item in cases) / len(cases)
regressions = [item for item in cases if item["candidate_score"] < item["baseline_score"]]
improved = [item for item in cases if item["candidate_score"] > item["baseline_score"]]
unchanged = len(cases) - len(regressions) - len(improved)
score_delta = candidate_avg - baseline_avg
passed = not (len(regressions) > 0 and score_delta <= 0) and candidate_avg >= 0.75
return SkillDraftEvalReport(
report_id=uuid4().hex,
skill_name=draft.skill_name,
draft_id=draft.draft_id,
candidate_id=candidate.candidate_id,
passed=passed,
baseline_score_avg=round(baseline_avg, 4),
candidate_score_avg=round(candidate_avg, 4),
score_delta=round(score_delta, 4),
regression_count=len(regressions),
improved_count=len(improved),
unchanged_count=unchanged,
cases=cases,
status="completed",
created_at=_utc_now(),
)
async def _evaluate_replay(
self,
*,
candidate: SkillLearningCandidate,
draft: SkillDraft,
replay_cases: list[dict],
provider_bundle: ProviderBundle,
replay_runner: ReplayRunner,
) -> SkillDraftEvalReport:
case_reports: list[dict] = []
legacy_cases: list[dict] = []
for case in replay_cases:
baseline = await replay_runner.run_arm(
ReplayArmRequest(
case_id=f"{case['run_id']}:baseline",
arm="baseline",
task_text=str(case["task_text"]),
pinned_skill_names=list(case.get("baseline_skill_names") or []),
pinned_skill_contexts=[],
provider_bundle=provider_bundle,
model_settings={"max_tool_iterations": 4, "temperature": 0.0},
)
)
candidate_arm = await replay_runner.run_arm(
ReplayArmRequest(
case_id=f"{case['run_id']}:candidate",
arm="candidate",
task_text=str(case["task_text"]),
pinned_skill_names=[],
pinned_skill_contexts=[_draft_skill_context(draft)],
provider_bundle=provider_bundle,
model_settings={"max_tool_iterations": 4, "temperature": 0.0},
)
)
surrogate = await self.surrogate_evaluator.evaluate(
task_text=str(case["task_text"]),
baseline=baseline,
candidate=candidate_arm,
)
baseline_score = surrogate["baseline_score"]
candidate_score = surrogate["candidate_score"]
case_report = {
"run_id": case["run_id"],
"task_id": case.get("task_id"),
"session_id": case.get("session_id"),
"baseline": baseline,
"candidate": candidate_arm,
"baseline_score": baseline_score,
"candidate_score": candidate_score,
"delta": round(candidate_score - baseline_score, 4),
"execution_coverage": _arm_mode_coverage(baseline, candidate_arm, "executed"),
"surrogate_coverage": _arm_mode_coverage(baseline, candidate_arm, "surrogate"),
"blocked_tool_count": _arm_mode_count(baseline, candidate_arm, "blocked"),
"confidence": surrogate["confidence"],
"tool_calls": [*baseline.get("tool_calls", []), *candidate_arm.get("tool_calls", [])],
"artifacts": [*baseline.get("artifacts", []), *candidate_arm.get("artifacts", [])],
"side_effects": [*baseline.get("side_effects", []), *candidate_arm.get("side_effects", [])],
"validator_notes": list(surrogate.get("notes") or []),
}
case_reports.append(case_report)
legacy_cases.append(
{
"run_id": case["run_id"],
"session_id": case.get("session_id") or "",
"baseline_score": baseline_score,
"candidate_score": candidate_score,
"delta": round(candidate_score - baseline_score, 4),
}
)
preservation_report = _preservation_report(candidate, draft)
return _report_from_case_reports(candidate, draft, case_reports, legacy_cases, preservation_report)
def _skipped(self, candidate: SkillLearningCandidate, draft: SkillDraft) -> SkillDraftEvalReport:
return SkillDraftEvalReport(
report_id=uuid4().hex,
skill_name=draft.skill_name,
draft_id=draft.draft_id,
candidate_id=candidate.candidate_id,
passed=True,
baseline_score_avg=0.0,
candidate_score_avg=0.0,
score_delta=0.0,
regression_count=0,
improved_count=0,
unchanged_count=0,
cases=[],
status="skipped_provider_unavailable",
created_at=_utc_now(),
)
def _score_from_validation(validation: dict | None, success: bool) -> float:
if isinstance(validation, dict) and "score" in validation:
try:
return max(0.0, min(1.0, float(validation.get("score") or 0.0)))
except (TypeError, ValueError):
pass
return 0.8 if success else 0.4
def _candidate_score(baseline: float, draft: SkillDraft) -> float:
content = draft.proposed_content.strip()
if not content and draft.proposal_kind != "retire_skill":
return 0.0
if "regression" in content.lower():
return max(0.0, baseline - 0.2)
return min(1.0, max(0.75, baseline + 0.05))
def _draft_skill_context(draft: SkillDraft) -> SkillContext:
tool_hints = draft.proposed_frontmatter.get("tools")
return SkillContext(
name=f"draft:{draft.skill_name}",
content=draft.proposed_content,
version=draft.draft_id,
content_hash="draft",
activation_reason="skill_replay_eval_candidate",
tool_hints=[str(item) for item in tool_hints if str(item).strip()] if isinstance(tool_hints, list) else [],
)
def _preservation_report(candidate: SkillLearningCandidate, draft: SkillDraft) -> dict | None:
if candidate.kind not in {"revise_skill", "merge_skills"}:
return None
base_content = str(candidate.evidence.get("base_content") or "") if isinstance(candidate.evidence, dict) else ""
if not base_content.strip():
return None
return check_preservation(base_content=base_content, draft_content=draft.proposed_content)
def _report_from_case_reports(
candidate: SkillLearningCandidate,
draft: SkillDraft,
case_reports: list[dict],
legacy_cases: list[dict],
preservation_report: dict | None,
) -> SkillDraftEvalReport:
baseline_avg = sum(item["baseline_score"] for item in legacy_cases) / len(legacy_cases)
candidate_avg = sum(item["candidate_score"] for item in legacy_cases) / len(legacy_cases)
regressions = [item for item in legacy_cases if item["candidate_score"] < item["baseline_score"]]
improved = [item for item in legacy_cases if item["candidate_score"] > item["baseline_score"]]
unchanged = len(legacy_cases) - len(regressions) - len(improved)
execution, surrogate, blocked = _coverage(case_reports)
confidence = _confidence(execution, surrogate, blocked, [item.get("confidence") for item in case_reports])
score_delta = candidate_avg - baseline_avg
passed = candidate_avg >= 0.75 and not (regressions and score_delta <= 0) and blocked < 1.0
return SkillDraftEvalReport(
report_id=uuid4().hex,
skill_name=draft.skill_name,
draft_id=draft.draft_id,
candidate_id=candidate.candidate_id,
passed=passed,
baseline_score_avg=round(baseline_avg, 4),
candidate_score_avg=round(candidate_avg, 4),
score_delta=round(score_delta, 4),
regression_count=len(regressions),
improved_count=len(improved),
unchanged_count=unchanged,
cases=legacy_cases,
status="completed",
created_at=_utc_now(),
eval_version="replay-v1",
mode="replay",
execution_coverage=execution,
surrogate_coverage=surrogate,
blocked_coverage=blocked,
confidence=confidence,
case_reports=case_reports,
tool_mode_summary={"executed": execution, "surrogate": surrogate, "blocked": blocked},
preservation_report=preservation_report,
)
def _coverage(case_reports: list[dict]) -> tuple[float, float, float]:
counts = {"executed": 0, "surrogate": 0, "blocked": 0}
for report in case_reports:
for call in report.get("tool_calls") or []:
if isinstance(call, dict) and call.get("mode") in counts:
counts[str(call["mode"])] += 1
total = sum(counts.values())
if total == 0:
return 1.0, 0.0, 0.0
return (
round(counts["executed"] / total, 4),
round(counts["surrogate"] / total, 4),
round(counts["blocked"] / total, 4),
)
def _confidence(execution: float, surrogate: float, blocked: float, case_confidences: list[object]) -> str:
if blocked > 0.0:
return "low"
if execution >= 0.75 and surrogate <= 0.25:
return "high"
if execution >= 0.25 or "medium" in case_confidences:
return "medium"
return "low"
def _arm_mode_coverage(baseline: dict, candidate: dict, mode: str) -> float:
calls = [*baseline.get("tool_calls", []), *candidate.get("tool_calls", [])]
if not calls:
return 1.0 if mode == "executed" else 0.0
return round(sum(1 for call in calls if isinstance(call, dict) and call.get("mode") == mode) / len(calls), 4)
def _arm_mode_count(baseline: dict, candidate: dict, mode: str) -> int:
calls = [*baseline.get("tool_calls", []), *candidate.get("tool_calls", [])]
return sum(1 for call in calls if isinstance(call, dict) and call.get("mode") == mode)
def _utc_now() -> str:
from datetime import datetime, timezone
return datetime.now(timezone.utc).isoformat()