diff --git a/llm_normalizer/data/autorun_generators/history.json b/llm_normalizer/data/autorun_generators/history.json index f1d01ce..8827b27 100644 --- a/llm_normalizer/data/autorun_generators/history.json +++ b/llm_normalizer/data/autorun_generators/history.json @@ -1,4 +1,58 @@ [ + { + "generation_id": "gen-ag06011720-49fcf4", + "created_at": "2026-06-01T17:20:01+00:00", + "mode": "saved_user_sessions", + "title": "AGENT | Hm business profile dependency mini pack", + "count": 6, + "domain": "assistant_stage1_business_overview", + "questions": [ + "какой у ООО \"Альтернатива Плюс\" самый доходный год", + "Как ты оценишь деятельность компании?", + "то есть бизнес прибыльный или это только обороты?", + "от кого мы больше всего зависим по входящим деньгам?", + "а по поставщикам и исходящим платежам от кого зависимость?", + "что дальше проверять в первую очередь?" + ], + "generated_by": "codex_agent", + "saved_case_set_file": "assistant_autogen_saved_user_sessions_20260601172001_gen-ag06011720-49fcf4.json", + "context": { + "llm_provider": null, + "model": null, + "assistant_prompt_version": null, + "decomposition_prompt_version": null, + "prompt_fingerprint": null, + "autogen_personality_id": null, + "autogen_personality_prompt": null, + "source_session_id": null, + "saved_session_file": "assistant_saved_session_20260601172001_gen-ag06011720-49fcf4.json", + "saved_case_set_kind": "agent_semantic_scenario", + "agent_run": true, + "agent_focus": "Mini semantic replay for broad HM business profile: company evaluation, profit boundary, customer/supplier dependency signals, and next checks must stay business-useful, evidence-bounded, and free of internal MCP mechanics.", + "architecture_phase": "turnaround_11", + "source_spec_file": "X:\\1C\\NDC_1C\\docs\\orchestration\\hm_business_profile_dependency_mini_pack_20260526.json", + "scenario_id": "hm_business_profile_dependency_mini_pack_20260526", + "semantic_tags": [ + "analyst_action_plan", + "broad_business_evaluation", + "business_overview", + "cashflow_not_profit", + "company_context", + "coverage_honesty", + "customer_dependency", + "incoming_counterparty_concentration", + "next_checks", + "outgoing_counterparty_concentration", + "profit_boundary", + "profitability_boundary", + "supplier_dependency", + "yearly_ranking" + ], + "validation_status": "accepted_live_replay", + "validated_run_dir": "artifacts\\domain_runs\\hm_business_profile_dependency_mini_pack_p5", + "saved_after_validated_replay": true + } + }, { "generation_id": "gen-ag05242306-7b49ab", "created_at": "2026-05-24T23:06:41+00:00", @@ -1499,4 +1553,4 @@ "latest_acceptance": null } } -] \ No newline at end of file +] diff --git a/llm_normalizer/data/autorun_generators/saved_sessions/assistant_saved_session_20260601172001_gen-ag06011720-49fcf4.json b/llm_normalizer/data/autorun_generators/saved_sessions/assistant_saved_session_20260601172001_gen-ag06011720-49fcf4.json new file mode 100644 index 0000000..d4ad057 --- /dev/null +++ b/llm_normalizer/data/autorun_generators/saved_sessions/assistant_saved_session_20260601172001_gen-ag06011720-49fcf4.json @@ -0,0 +1,193 @@ +{ + "saved_at": "2026-06-01T17:20:01+00:00", + "generation_id": "gen-ag06011720-49fcf4", + "mode": "saved_user_sessions", + "title": "AGENT | Hm business profile dependency mini pack", + "agent_run": true, + "questions": [ + "какой у ООО \"Альтернатива Плюс\" самый доходный год", + "Как ты оценишь деятельность компании?", + "то есть бизнес прибыльный или это только обороты?", + "от кого мы больше всего зависим по входящим деньгам?", + "а по поставщикам и исходящим платежам от кого зависимость?", + "что дальше проверять в первую очередь?" + ], + "metadata": { + "assistant_prompt_version": null, + "decomposition_prompt_version": null, + "prompt_fingerprint": null, + "agent_focus": "Mini semantic replay for broad HM business profile: company evaluation, profit boundary, customer/supplier dependency signals, and next checks must stay business-useful, evidence-bounded, and free of internal MCP mechanics.", + "architecture_phase": "turnaround_11", + "source_spec_file": "X:\\1C\\NDC_1C\\docs\\orchestration\\hm_business_profile_dependency_mini_pack_20260526.json", + "scenario_id": "hm_business_profile_dependency_mini_pack_20260526", + "semantic_tags": [ + "analyst_action_plan", + "broad_business_evaluation", + "business_overview", + "cashflow_not_profit", + "company_context", + "coverage_honesty", + "customer_dependency", + "incoming_counterparty_concentration", + "next_checks", + "outgoing_counterparty_concentration", + "profit_boundary", + "profitability_boundary", + "supplier_dependency", + "yearly_ranking" + ], + "validation_status": "accepted_live_replay", + "validated_run_dir": "artifacts\\domain_runs\\hm_business_profile_dependency_mini_pack_p5", + "saved_after_validated_replay": true, + "save_gate": { + "schema_version": "agent_semantic_save_gate_v1", + "validation_status": "accepted_live_replay", + "validated_run_dir": "artifacts\\domain_runs\\hm_business_profile_dependency_mini_pack_p5", + "final_status": "accepted", + "review_overall_status": "pass", + "business_overall_status": "pass", + "steps_total": 6, + "steps_passed": 6, + "steps_failed": 0, + "steps_with_business_failures": 0, + "steps_with_business_warnings": 0, + "acceptance_gate_passed": true, + "effective_runtime": { + "manifest_path": "artifacts\\domain_runs\\hm_business_profile_dependency_mini_pack_p5\\effective_runtime.json", + "runner": "domain_truth_harness.run-live", + "git_sha": "92473c894f7106f70d01de9d75fa83ad148986dd", + "backend_url": "http://127.0.0.1:8787", + "mcp_proxy_url": "http://127.0.0.1:6003", + "llm_provider": "local", + "llm_model": "unsloth/qwen3-30b-a3b-instruct-2507", + "temperature": 0.8, + "max_output_tokens": 900, + "prompt_version": "normalizer_v2_0_2", + "prompt_source": "file", + "prompt_hash": "f36e36a5e491cd24511b380e0c0059cb01aea5e2af738fb9f01671cd291735c1", + "prompt_registry_status": "pass" + }, + "saved_after_validated_replay": true + } + }, + "source_session_id": null, + "session": { + "session_id": null, + "mode": "agent_semantic_run", + "items": [ + { + "message_id": "agent-user-001", + "role": "user", + "text": "какой у ООО \"Альтернатива Плюс\" самый доходный год", + "created_at": "2026-06-01T17:20:01+00:00", + "reply_type": null, + "trace_id": null, + "debug": null + }, + { + "message_id": "agent-user-002", + "role": "user", + "text": "Как ты оценишь деятельность компании?", + "created_at": "2026-06-01T17:20:01+00:00", + "reply_type": null, + "trace_id": null, + "debug": null + }, + { + "message_id": "agent-user-003", + "role": "user", + "text": "то есть бизнес прибыльный или это только обороты?", + "created_at": "2026-06-01T17:20:01+00:00", + "reply_type": null, + "trace_id": null, + "debug": null + }, + { + "message_id": "agent-user-004", + "role": "user", + "text": "от кого мы больше всего зависим по входящим деньгам?", + "created_at": "2026-06-01T17:20:01+00:00", + "reply_type": null, + "trace_id": null, + "debug": null + }, + { + "message_id": "agent-user-005", + "role": "user", + "text": "а по поставщикам и исходящим платежам от кого зависимость?", + "created_at": "2026-06-01T17:20:01+00:00", + "reply_type": null, + "trace_id": null, + "debug": null + }, + { + "message_id": "agent-user-006", + "role": "user", + "text": "что дальше проверять в первую очередь?", + "created_at": "2026-06-01T17:20:01+00:00", + "reply_type": null, + "trace_id": null, + "debug": null + } + ], + "agent_run": true, + "metadata": { + "assistant_prompt_version": null, + "decomposition_prompt_version": null, + "prompt_fingerprint": null, + "agent_focus": "Mini semantic replay for broad HM business profile: company evaluation, profit boundary, customer/supplier dependency signals, and next checks must stay business-useful, evidence-bounded, and free of internal MCP mechanics.", + "architecture_phase": "turnaround_11", + "source_spec_file": "X:\\1C\\NDC_1C\\docs\\orchestration\\hm_business_profile_dependency_mini_pack_20260526.json", + "scenario_id": "hm_business_profile_dependency_mini_pack_20260526", + "semantic_tags": [ + "analyst_action_plan", + "broad_business_evaluation", + "business_overview", + "cashflow_not_profit", + "company_context", + "coverage_honesty", + "customer_dependency", + "incoming_counterparty_concentration", + "next_checks", + "outgoing_counterparty_concentration", + "profit_boundary", + "profitability_boundary", + "supplier_dependency", + "yearly_ranking" + ], + "validation_status": "accepted_live_replay", + "validated_run_dir": "artifacts\\domain_runs\\hm_business_profile_dependency_mini_pack_p5", + "saved_after_validated_replay": true, + "save_gate": { + "schema_version": "agent_semantic_save_gate_v1", + "validation_status": "accepted_live_replay", + "validated_run_dir": "artifacts\\domain_runs\\hm_business_profile_dependency_mini_pack_p5", + "final_status": "accepted", + "review_overall_status": "pass", + "business_overall_status": "pass", + "steps_total": 6, + "steps_passed": 6, + "steps_failed": 0, + "steps_with_business_failures": 0, + "steps_with_business_warnings": 0, + "acceptance_gate_passed": true, + "effective_runtime": { + "manifest_path": "artifacts\\domain_runs\\hm_business_profile_dependency_mini_pack_p5\\effective_runtime.json", + "runner": "domain_truth_harness.run-live", + "git_sha": "92473c894f7106f70d01de9d75fa83ad148986dd", + "backend_url": "http://127.0.0.1:8787", + "mcp_proxy_url": "http://127.0.0.1:6003", + "llm_provider": "local", + "llm_model": "unsloth/qwen3-30b-a3b-instruct-2507", + "temperature": 0.8, + "max_output_tokens": 900, + "prompt_version": "normalizer_v2_0_2", + "prompt_source": "file", + "prompt_hash": "f36e36a5e491cd24511b380e0c0059cb01aea5e2af738fb9f01671cd291735c1", + "prompt_registry_status": "pass" + }, + "saved_after_validated_replay": true + } + } + } +} diff --git a/llm_normalizer/data/eval_cases/assistant_autogen_saved_user_sessions_20260601172001_gen-ag06011720-49fcf4.json b/llm_normalizer/data/eval_cases/assistant_autogen_saved_user_sessions_20260601172001_gen-ag06011720-49fcf4.json new file mode 100644 index 0000000..6406c56 --- /dev/null +++ b/llm_normalizer/data/eval_cases/assistant_autogen_saved_user_sessions_20260601172001_gen-ag06011720-49fcf4.json @@ -0,0 +1,43 @@ +{ + "suite_id": "assistant_saved_session_gen-ag06011720-49fcf4", + "suite_version": "0.1.0", + "schema_version": "assistant_saved_session_suite_v0_1", + "generated_at": "2026-06-01T17:20:01+00:00", + "generation_id": "gen-ag06011720-49fcf4", + "mode": "saved_user_sessions", + "title": "AGENT | Hm business profile dependency mini pack", + "domain": "assistant_stage1_business_overview", + "scenario_count": 1, + "case_ids": [ + "SAVED-001" + ], + "cases": [ + { + "case_id": "SAVED-001", + "scenario_tag": "agent_saved_user_sessions", + "title": "AGENT | Hm business profile dependency mini pack", + "question_type": "followup", + "broadness_level": "medium", + "turns": [ + { + "user_message": "какой у ООО \"Альтернатива Плюс\" самый доходный год" + }, + { + "user_message": "Как ты оценишь деятельность компании?" + }, + { + "user_message": "то есть бизнес прибыльный или это только обороты?" + }, + { + "user_message": "от кого мы больше всего зависим по входящим деньгам?" + }, + { + "user_message": "а по поставщикам и исходящим платежам от кого зависимость?" + }, + { + "user_message": "что дальше проверять в первую очередь?" + } + ] + } + ] +}