agent-apprenticeship 0.1.6 → 0.1.7

package/src/agent_apprenticeship_trace/loop.py

@@ -11,6 +11,7 @@ from .codex_runner import deterministic_attempt, run_codex_attempt, run_custom_a
 from .grader import grade_attempt, apply_score_reliability
 from .verifier import verify_attempt
 from .evaluator import evaluate_attempt
+from .loop_review import reviewer_feedback_from_evaluator, write_loop_review_iteration
 from .lesson_extractor import extract_lesson
 from .training_signals import process_supervision_from_trace, reward_modeling, training_signals
 from .revision import compute_hillclimb, preference_pair
@@ -42,12 +43,25 @@ def _write_baseline_only_signals(pkg: Path, b_actual: ActualOutputs, b_grade, b_
     append_jsonl(pkg/'signals/reward_modeling.jsonl', reward_modeling(b_actual,b_grade))
 
 
+def _package_manifest(settings, **data):
+    return {
+        **data,
+        "apprenticeship_mode": getattr(settings, "apprenticeship_mode", None),
+        "mentor_mode": settings.mentor_mode,
+    }
+
+
+def _mode_label(settings) -> str:
+    return getattr(settings, "apprenticeship_mode", None) or settings.mentor_mode
+
+
 def run_task(
     raw: RawTaskRecord,
     run_root: Path,
     runner='deterministic',
     max_iterations: int | None=None,
     pre_attempt_callback: Callable[[Path], None] | None = None,
+    attempt_complete_callback: Callable[[Path, str], None] | None = None,
     revision_decision_callback: Callable[[Path], bool] | None = None,
 ) -> Path:
     settings=get_settings()
@@ -63,31 +77,89 @@ def run_task(
     baseline = _attempt(pkg, raw, spec, 'baseline', runner)
     b_actual=ActualOutputs.model_validate(read_json(pkg/'attempts/baseline/actual_outputs.json'))
     b_trace=AgentTrace.model_validate(read_json(pkg/'attempts/baseline/agent_trace.json'))
+    if attempt_complete_callback:
+        attempt_complete_callback(pkg, 'baseline')
     revised_attempt_ids=[]; selected_attempt_id=b_actual.attempt_id
     if _is_apprentice_operational_failure(b_actual):
-        write_json(pkg/'package_manifest.json', {'max_iterations':max_iter,'actual_iterations':1,'loop_stop_reason':'apprentice_agent_operational_error','baseline_attempt_id':b_actual.attempt_id,'revised_attempt_ids':revised_attempt_ids,'selected_attempt_id':selected_attempt_id})
+        write_json(pkg/'package_manifest.json', _package_manifest(settings, max_iterations=max_iter,actual_iterations=1,loop_stop_reason='apprentice_agent_operational_error',baseline_attempt_id=b_actual.attempt_id,revised_attempt_ids=revised_attempt_ids,selected_attempt_id=selected_attempt_id))
         write_artifacts_index(pkg)
         return pkg
     b_grade=grade_attempt(rubric, b_actual, 'baseline', b_trace, role_root, pkg)
     b_ver=verify_attempt(b_grade, b_actual, b_trace, role_root, pkg)
     b_grade=apply_score_reliability(b_grade, b_ver)
     write_json(pkg/'grading/baseline_grader_result.json', b_grade); write_json(pkg/'grading/baseline_verifier_result.json', b_ver)
+    write_loop_review_iteration(
+        pkg,
+        iteration=1,
+        mode=_mode_label(settings),
+        actual_outputs=b_actual,
+        trace=b_trace,
+        grader_result=b_grade,
+        verifier_result=b_ver,
+        stop_reason="awaiting_reviewer_feedback",
+        continue_loop=False,
+    )
     stop_reason='max_iterations_reached' if max_iter <= 1 else 'completed_baseline_and_revision'
     if max_iter <= 1:
         fb,rp=evaluate_attempt(b_grade, b_ver, b_actual, b_trace, f'{spec.task_id}_revised', role_root, pkg)
         write_json(pkg/'feedback/baseline_evaluator_feedback.json', fb); write_json(pkg/'feedback/revision_plan.json', rp)
+        write_loop_review_iteration(
+            pkg,
+            iteration=1,
+            mode=_mode_label(settings),
+            actual_outputs=b_actual,
+            trace=b_trace,
+            grader_result=b_grade,
+            verifier_result=b_ver,
+            evaluator_feedback=fb,
+            revision_plan=rp,
+            reviewer_feedback=reviewer_feedback_from_evaluator(fb, b_grade, b_ver),
+            stop_reason=stop_reason,
+            continue_loop=False,
+        )
         if revision_decision_callback:
             revision_decision_callback(pkg)
         _write_baseline_only_signals(pkg, b_actual, b_grade, b_ver, b_trace, fb)
-        write_json(pkg/'package_manifest.json', {'max_iterations':max_iter,'actual_iterations':actual_iterations,'loop_stop_reason':stop_reason,'baseline_attempt_id':b_actual.attempt_id,'revised_attempt_ids':revised_attempt_ids,'selected_attempt_id':selected_attempt_id})
+        write_json(pkg/'package_manifest.json', _package_manifest(settings, max_iterations=max_iter,actual_iterations=actual_iterations,loop_stop_reason=stop_reason,baseline_attempt_id=b_actual.attempt_id,revised_attempt_ids=revised_attempt_ids,selected_attempt_id=selected_attempt_id))
         write_artifacts_index(pkg)
         return pkg
     fb,rp=evaluate_attempt(b_grade, b_ver, b_actual, b_trace, f'{spec.task_id}_revised', role_root, pkg)
     write_json(pkg/'feedback/baseline_evaluator_feedback.json', fb); write_json(pkg/'feedback/revision_plan.json', rp)
+    baseline_reviewer_feedback=reviewer_feedback_from_evaluator(fb, b_grade, b_ver)
+    write_loop_review_iteration(
+        pkg,
+        iteration=1,
+        mode=_mode_label(settings),
+        actual_outputs=b_actual,
+        trace=b_trace,
+        grader_result=b_grade,
+        verifier_result=b_ver,
+        evaluator_feedback=fb,
+        revision_plan=rp,
+        reviewer_feedback=baseline_reviewer_feedback,
+        stop_reason="revision_requested",
+        continue_loop=True,
+    )
     if revision_decision_callback and not revision_decision_callback(pkg):
         stop_reason='mentor_decision_finish'
+        existing_reviewer = read_json(pkg/'loops/iterations/001/reviewer_feedback.json') if (pkg/'loops/iterations/001/reviewer_feedback.json').exists() else {}
+        if existing_reviewer.get('reviewer_type') not in {'human_expert','simulated_for_test','org_custom_reviewer'}:
+            write_loop_review_iteration(
+                pkg,
+                iteration=1,
+                mode=_mode_label(settings),
+                actual_outputs=b_actual,
+                trace=b_trace,
+                grader_result=b_grade,
+                verifier_result=b_ver,
+                evaluator_feedback=fb,
+                revision_plan=rp,
+                reviewer_feedback=baseline_reviewer_feedback,
+                stop_reason=stop_reason,
+                continue_loop=False,
+            )
         _write_baseline_only_signals(pkg, b_actual, b_grade, b_ver, b_trace, fb)
-        write_json(pkg/'package_manifest.json', {'max_iterations':max_iter,'actual_iterations':actual_iterations,'loop_stop_reason':stop_reason,'baseline_attempt_id':b_actual.attempt_id,'revised_attempt_ids':revised_attempt_ids,'selected_attempt_id':selected_attempt_id})
+        write_json(pkg/'package_manifest.json', _package_manifest(settings, max_iterations=max_iter,actual_iterations=actual_iterations,loop_stop_reason=stop_reason,baseline_attempt_id=b_actual.attempt_id,revised_attempt_ids=revised_attempt_ids,selected_attempt_id=selected_attempt_id))
         write_artifacts_index(pkg)
         return pkg
     revised = _attempt(pkg, raw, spec, 'revised', runner, fb.feedback_summary)
@@ -98,14 +170,41 @@ def run_task(
     r_ver=verify_attempt(r_grade, r_actual, r_trace, role_root, pkg)
     r_grade=apply_score_reliability(r_grade, r_ver)
     write_json(pkg/'grading/revised_grader_result.json', r_grade); write_json(pkg/'grading/revised_verifier_result.json', r_ver)
+    write_loop_review_iteration(
+        pkg,
+        iteration=2,
+        mode=_mode_label(settings),
+        actual_outputs=r_actual,
+        trace=r_trace,
+        grader_result=r_grade,
+        verifier_result=r_ver,
+        stop_reason="awaiting_reviewer_feedback",
+        continue_loop=False,
+    )
+    r_fb,r_rp=evaluate_attempt(r_grade, r_ver, r_actual, r_trace, f'{spec.task_id}_final', role_root, pkg)
+    write_json(pkg/'feedback/revised_evaluator_feedback.json', r_fb); write_json(pkg/'feedback/revised_revision_plan.json', r_rp)
     revised_attempt_ids.append(r_actual.attempt_id)
     selected_attempt_id=r_actual.attempt_id if (r_grade.final_score or r_grade.score) >= (b_grade.final_score or b_grade.score) else b_actual.attempt_id
+    write_loop_review_iteration(
+        pkg,
+        iteration=2,
+        mode=_mode_label(settings),
+        actual_outputs=r_actual,
+        trace=r_trace,
+        grader_result=r_grade,
+        verifier_result=r_ver,
+        evaluator_feedback=r_fb,
+        revision_plan=r_rp,
+        reviewer_feedback=reviewer_feedback_from_evaluator(r_fb, r_grade, r_ver),
+        stop_reason=stop_reason,
+        continue_loop=False,
+    )
     hill=compute_hillclimb(b_grade, r_grade); write_json(pkg/'signals/hillclimb_result.json', hill)
     lesson=extract_lesson(hill); write_json(pkg/'signals/lesson_pack.json', lesson)
     for sig in training_signals(hill, ['attempts/baseline/agent_trace.json','attempts/revised/agent_trace.json'], ['grading/baseline_grader_result.json','grading/revised_grader_result.json'], ['grading/baseline_verifier_result.json','grading/revised_verifier_result.json']): append_jsonl(pkg/'signals/training_signals.jsonl', sig)
     for ex in process_supervision_from_trace(b_trace, b_grade, b_ver, fb)+process_supervision_from_trace(r_trace, r_grade, r_ver, fb): append_jsonl(pkg/'signals/process_supervision.jsonl', ex)
     append_jsonl(pkg/'signals/reward_modeling.jsonl', reward_modeling(b_actual,b_grade)); append_jsonl(pkg/'signals/reward_modeling.jsonl', reward_modeling(r_actual,r_grade))
     append_jsonl(pkg/'signals/revision_preference_pairs.jsonl', preference_pair(hill, 'rubric/rubric.json'))
-    write_json(pkg/'package_manifest.json', {'max_iterations':max_iter,'actual_iterations':actual_iterations,'loop_stop_reason':stop_reason,'baseline_attempt_id':b_actual.attempt_id,'revised_attempt_ids':revised_attempt_ids,'selected_attempt_id':selected_attempt_id})
+    write_json(pkg/'package_manifest.json', _package_manifest(settings, max_iterations=max_iter,actual_iterations=actual_iterations,loop_stop_reason=stop_reason,baseline_attempt_id=b_actual.attempt_id,revised_attempt_ids=revised_attempt_ids,selected_attempt_id=selected_attempt_id))
     write_artifacts_index(pkg)
     return pkg