Сохранить accepted прогон Phase108 в autoruns

2026-06-15 23:47:11 +03:00 · 2026-06-15 23:47:11 +03:00 · fa07bb1f0e
parent 2873758579
commit fa07bb1f0e
3 changed files with 351 additions and 0 deletions
--- a/llm_normalizer/data/autorun_generators/history.json
+++ b/llm_normalizer/data/autorun_generators/history.json
@ -1,4 +1,69 @@
 [
+  {
+    "generation_id": "gen-ag06152044-9f689d",
+    "created_at": "2026-06-15T20:44:38+00:00",
+    "mode": "saved_user_sessions",
+    "title": "AGENT | Phase108 settlement pivots and debt due-date boundary",
+    "count": 8,
+    "domain": "address_phase108_settlement_pivot_debt_boundary",
+    "questions": [
+      "Покажи документы по Жуковке 51.",
+      "А по нему платежи?",
+      "А по нему договоры?",
+      "А по нему документы?",
+      "А за 2021?",
+      "По ООО Альтернатива Плюс на конец 2020 можно точно понять, какая дебиторка просрочена?",
+      "То есть просрочку доказать нельзя, коротко почему?",
+      "Ок, а пингвины умеют летать?"
+    ],
+    "generated_by": "codex_agent",
+    "saved_case_set_file": "assistant_autogen_saved_user_sessions_20260615204438_gen-ag06152044-9f689d.json",
+    "context": {
+      "llm_provider": null,
+      "model": null,
+      "assistant_prompt_version": "address_query_runtime_v1",
+      "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_20260615204438_gen-ag06152044-9f689d.json",
+      "saved_case_set_kind": "agent_semantic_scenario",
+      "agent_run": true,
+      "agent_focus": "settlement pivots + debt due-date boundary + agent validation",
+      "architecture_phase": "assistant-stage1 phase108",
+      "source_spec_file": "X:\\1C\\NDC_1C\\docs\\orchestration\\active_domain_contract.json",
+      "scenario_id": null,
+      "semantic_tags": [
+        "business_overview",
+        "checked_negative",
+        "context_boundary",
+        "context_carryover",
+        "contracts_followup",
+        "counterparty_pronoun_resolution",
+        "counterparty_root",
+        "debt_due_date_boundary",
+        "document_to_payments_pivot",
+        "documents_by_counterparty",
+        "documents_followup",
+        "fourth_pivot",
+        "honest_boundary",
+        "human_answer",
+        "integrity_guard",
+        "off_domain_living_chat",
+        "payments_followup",
+        "period_followup_reuse",
+        "pivot_seed",
+        "second_pivot",
+        "short_followup",
+        "third_pivot",
+        "year_switch_after_fourth_pivot"
+      ],
+      "validation_status": "accepted_domain_pack_loop",
+      "validated_run_dir": "artifacts\\domain_runs\\phase108_settlement_pivot_debt_boundary_20260615_p05_loop",
+      "saved_after_validated_replay": true
+    }
+  },
  {
    "generation_id": "gen-ag06151952-5d8f12",
    "created_at": "2026-06-15T19:52:40+00:00",
--- a/llm_normalizer/data/autorun_generators/saved_sessions/assistant_saved_session_20260615204438_gen-ag06152044-9f689d.json
+++ b/llm_normalizer/data/autorun_generators/saved_sessions/assistant_saved_session_20260615204438_gen-ag06152044-9f689d.json
@ -0,0 +1,237 @@
+{
+  "saved_at": "2026-06-15T20:44:38+00:00",
+  "generation_id": "gen-ag06152044-9f689d",
+  "mode": "saved_user_sessions",
+  "title": "AGENT | Phase108 settlement pivots and debt due-date boundary",
+  "agent_run": true,
+  "questions": [
+    "Покажи документы по Жуковке 51.",
+    "А по нему платежи?",
+    "А по нему договоры?",
+    "А по нему документы?",
+    "А за 2021?",
+    "По ООО Альтернатива Плюс на конец 2020 можно точно понять, какая дебиторка просрочена?",
+    "То есть просрочку доказать нельзя, коротко почему?",
+    "Ок, а пингвины умеют летать?"
+  ],
+  "metadata": {
+    "assistant_prompt_version": "address_query_runtime_v1",
+    "decomposition_prompt_version": null,
+    "prompt_fingerprint": null,
+    "agent_focus": "settlement pivots + debt due-date boundary + agent validation",
+    "architecture_phase": "assistant-stage1 phase108",
+    "source_spec_file": "X:\\1C\\NDC_1C\\docs\\orchestration\\active_domain_contract.json",
+    "scenario_id": null,
+    "semantic_tags": [
+      "business_overview",
+      "checked_negative",
+      "context_boundary",
+      "context_carryover",
+      "contracts_followup",
+      "counterparty_pronoun_resolution",
+      "counterparty_root",
+      "debt_due_date_boundary",
+      "document_to_payments_pivot",
+      "documents_by_counterparty",
+      "documents_followup",
+      "fourth_pivot",
+      "honest_boundary",
+      "human_answer",
+      "integrity_guard",
+      "off_domain_living_chat",
+      "payments_followup",
+      "period_followup_reuse",
+      "pivot_seed",
+      "second_pivot",
+      "short_followup",
+      "third_pivot",
+      "year_switch_after_fourth_pivot"
+    ],
+    "validation_status": "accepted_domain_pack_loop",
+    "validated_run_dir": "artifacts\\domain_runs\\phase108_settlement_pivot_debt_boundary_20260615_p05_loop",
+    "saved_after_validated_replay": true,
+    "save_gate": {
+      "schema_version": "agent_semantic_save_gate_v1",
+      "validation_status": "accepted_domain_pack_loop",
+      "validated_run_dir": "artifacts\\domain_runs\\phase108_settlement_pivot_debt_boundary_20260615_p05_loop",
+      "final_status": "accepted",
+      "loop_id": "phase108_settlement_pivot_debt_boundary_20260615_p05_loop",
+      "target_score": 88,
+      "iterations_ran": 1,
+      "quality_score": 92,
+      "repair_target_count": 0,
+      "repair_target_severity_counts": {
+        "P0": 0,
+        "P1": 0,
+        "P2": 0
+      },
+      "accepted_gate": true,
+      "effective_runtime": {
+        "manifest_path": "artifacts\\domain_runs\\phase108_settlement_pivot_debt_boundary_20260615_p05_loop\\effective_runtime.json",
+        "runner": "domain_case_loop.run-pack-loop",
+        "git_sha": "e9f0ef6b4129e1e30d010c8ef9272afd742629fe",
+        "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": "Покажи документы по Жуковке 51.",
+        "created_at": "2026-06-15T20:44:38+00:00",
+        "reply_type": null,
+        "trace_id": null,
+        "debug": null
+      },
+      {
+        "message_id": "agent-user-002",
+        "role": "user",
+        "text": "А по нему платежи?",
+        "created_at": "2026-06-15T20:44:38+00:00",
+        "reply_type": null,
+        "trace_id": null,
+        "debug": null
+      },
+      {
+        "message_id": "agent-user-003",
+        "role": "user",
+        "text": "А по нему договоры?",
+        "created_at": "2026-06-15T20:44:38+00:00",
+        "reply_type": null,
+        "trace_id": null,
+        "debug": null
+      },
+      {
+        "message_id": "agent-user-004",
+        "role": "user",
+        "text": "А по нему документы?",
+        "created_at": "2026-06-15T20:44:38+00:00",
+        "reply_type": null,
+        "trace_id": null,
+        "debug": null
+      },
+      {
+        "message_id": "agent-user-005",
+        "role": "user",
+        "text": "А за 2021?",
+        "created_at": "2026-06-15T20:44:38+00:00",
+        "reply_type": null,
+        "trace_id": null,
+        "debug": null
+      },
+      {
+        "message_id": "agent-user-006",
+        "role": "user",
+        "text": "По ООО Альтернатива Плюс на конец 2020 можно точно понять, какая дебиторка просрочена?",
+        "created_at": "2026-06-15T20:44:38+00:00",
+        "reply_type": null,
+        "trace_id": null,
+        "debug": null
+      },
+      {
+        "message_id": "agent-user-007",
+        "role": "user",
+        "text": "То есть просрочку доказать нельзя, коротко почему?",
+        "created_at": "2026-06-15T20:44:38+00:00",
+        "reply_type": null,
+        "trace_id": null,
+        "debug": null
+      },
+      {
+        "message_id": "agent-user-008",
+        "role": "user",
+        "text": "Ок, а пингвины умеют летать?",
+        "created_at": "2026-06-15T20:44:38+00:00",
+        "reply_type": null,
+        "trace_id": null,
+        "debug": null
+      }
+    ],
+    "agent_run": true,
+    "metadata": {
+      "assistant_prompt_version": "address_query_runtime_v1",
+      "decomposition_prompt_version": null,
+      "prompt_fingerprint": null,
+      "agent_focus": "settlement pivots + debt due-date boundary + agent validation",
+      "architecture_phase": "assistant-stage1 phase108",
+      "source_spec_file": "X:\\1C\\NDC_1C\\docs\\orchestration\\active_domain_contract.json",
+      "scenario_id": null,
+      "semantic_tags": [
+        "business_overview",
+        "checked_negative",
+        "context_boundary",
+        "context_carryover",
+        "contracts_followup",
+        "counterparty_pronoun_resolution",
+        "counterparty_root",
+        "debt_due_date_boundary",
+        "document_to_payments_pivot",
+        "documents_by_counterparty",
+        "documents_followup",
+        "fourth_pivot",
+        "honest_boundary",
+        "human_answer",
+        "integrity_guard",
+        "off_domain_living_chat",
+        "payments_followup",
+        "period_followup_reuse",
+        "pivot_seed",
+        "second_pivot",
+        "short_followup",
+        "third_pivot",
+        "year_switch_after_fourth_pivot"
+      ],
+      "validation_status": "accepted_domain_pack_loop",
+      "validated_run_dir": "artifacts\\domain_runs\\phase108_settlement_pivot_debt_boundary_20260615_p05_loop",
+      "saved_after_validated_replay": true,
+      "save_gate": {
+        "schema_version": "agent_semantic_save_gate_v1",
+        "validation_status": "accepted_domain_pack_loop",
+        "validated_run_dir": "artifacts\\domain_runs\\phase108_settlement_pivot_debt_boundary_20260615_p05_loop",
+        "final_status": "accepted",
+        "loop_id": "phase108_settlement_pivot_debt_boundary_20260615_p05_loop",
+        "target_score": 88,
+        "iterations_ran": 1,
+        "quality_score": 92,
+        "repair_target_count": 0,
+        "repair_target_severity_counts": {
+          "P0": 0,
+          "P1": 0,
+          "P2": 0
+        },
+        "accepted_gate": true,
+        "effective_runtime": {
+          "manifest_path": "artifacts\\domain_runs\\phase108_settlement_pivot_debt_boundary_20260615_p05_loop\\effective_runtime.json",
+          "runner": "domain_case_loop.run-pack-loop",
+          "git_sha": "e9f0ef6b4129e1e30d010c8ef9272afd742629fe",
+          "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
+      }
+    }
+  }
+}
--- a/llm_normalizer/data/eval_cases/assistant_autogen_saved_user_sessions_20260615204438_gen-ag06152044-9f689d.json
+++ b/llm_normalizer/data/eval_cases/assistant_autogen_saved_user_sessions_20260615204438_gen-ag06152044-9f689d.json
@ -0,0 +1,49 @@
+{
+  "suite_id": "assistant_saved_session_gen-ag06152044-9f689d",
+  "suite_version": "0.1.0",
+  "schema_version": "assistant_saved_session_suite_v0_1",
+  "generated_at": "2026-06-15T20:44:38+00:00",
+  "generation_id": "gen-ag06152044-9f689d",
+  "mode": "saved_user_sessions",
+  "title": "AGENT | Phase108 settlement pivots and debt due-date boundary",
+  "domain": "address_phase108_settlement_pivot_debt_boundary",
+  "scenario_count": 1,
+  "case_ids": [
+    "SAVED-001"
+  ],
+  "cases": [
+    {
+      "case_id": "SAVED-001",
+      "scenario_tag": "agent_saved_user_sessions",
+      "title": "AGENT | Phase108 settlement pivots and debt due-date boundary",
+      "question_type": "followup",
+      "broadness_level": "medium",
+      "turns": [
+        {
+          "user_message": "Покажи документы по Жуковке 51."
+        },
+        {
+          "user_message": "А по нему платежи?"
+        },
+        {
+          "user_message": "А по нему договоры?"
+        },
+        {
+          "user_message": "А по нему документы?"
+        },
+        {
+          "user_message": "А за 2021?"
+        },
+        {
+          "user_message": "По ООО Альтернатива Плюс на конец 2020 можно точно понять, какая дебиторка просрочена?"
+        },
+        {
+          "user_message": "То есть просрочку доказать нельзя, коротко почему?"
+        },
+        {
+          "user_message": "Ок, а пингвины умеют летать?"
+        }
+      ]
+    }
+  ]
+}