Upload 655 files

tasks/offline-compute/HiveSQL/hivesql.yaml

@@ -0,0 +1,6 @@
+metastore: true
+init_cmd: spark-submit --driver-memory 2g {init_script}
+init_script_name: init_db.py
+gateway_cmd: $SPARK_HOME/sbin/start-thriftserver.sh
+gateway_stop_cmd: $SPARK_HOME/sbin/stop-thriftserver.sh
+gateway_port: 10000

tasks/offline-compute/HiveSQL/hivesql_001_en/gt/expected.csv

@@ -0,0 +1,6 @@
+business_id,business_name,cluster_set,tenant,namespaces,topic,mq_type,dw_appgroup,in_charge,description,create_time,modify_time,cluster_id,cluster_type,cluster_name,bg,category_name,is_filtered,tids,consumed_tids,unconsumed_tids,is_fully_consumed,has_unconsumed_tid,system_belong,dt
+bid001,BizName1,cluster_set_a,tenant_a,ns_a,topic_mq_1,PULSAR,appgroup1,user1,desc1,2026-01-01,2026-05-01,cid001,消息队列MQ,cluster_p1,BG1,product1,,,,,,,数据总线,20260507
+bid005,BizName5,cluster_set_b,tenant_c,ns_c,topic_mq_2,PULSAR,appgroup5,user5,desc5,2026-05-01,2026-05-05,cid005,消息队列MQ,cluster_p2,BG3,product5,,,,,,,流处理平台,20260507
+bid002,BizName2,,,,topic_tube_1,TUBEMQ,appgroup2,user2,desc2,2026-02-01,2026-05-02,cid002,TubeMQ,cluster_t1,BG2,product2,1,"tid1,tid2",tid1,,0,1,数据总线,20260507
+bid004,BizName4,,,,topic_tube_2,TUBEMQ,appgroup4,user4,desc4,2026-04-01,2026-05-04,cid004,TubeMQ,cluster_t2,BG1,product4,0,"tid3,tid4","tid3,tid4",,1,0,数据总线,20260507
+bid003,BizName3,,tenant_b,ns_b,topic_inlong_1,PULSAR,appgroup3,user3,,2026-03-01,2026-05-03,,,cluster_i1,,product3,,,,,,,流处理平台,20260507

tasks/offline-compute/HiveSQL/hivesql_001_en/gt/grade.py

@@ -0,0 +1,715 @@
+"""hivesql_001 精细评分脚本
+
+业务场景：消息队列Topic维度表前一天分区数据原样复制到今天的分区
+难度: EASY | 特征: INSERT_OVERWRITE|PARTITION|SINGLE_TABLE
+
+评分范式: 产物100分 = A(15) + B(10) + C(15) + D(40) + F(20)
+  A_executability       (15分): result.sql 能跑通且产出非空
+  B_schema              (10分): 25列(5) + 列名匹配(5)
+  C_row_alignment     (15分): 行数比例(7) + key覆盖率(8)
+  D_field_value_match   (25分): 非key字段逐列值匹配率
+  D_field_completeness  (15分): 关键字段非空/非空串比例
+  F_insert_overwrite     (5分):  INSERT OVERWRITE + PARTITION
+  F_partition_value      (5分):  dt 分区值 = 20260507
+  F_source_filter      (10分):  源表分区过滤 dt=20260506
+
+权重: EASY -> product=0.5, process=0.5
+"""
+import os
+import re
+import subprocess
+import tempfile
+import json
+import math
+
+
+def grade(workspace_path, **kwargs):
+
+    # ========== Case 配置 ==========
+    OUTPUT_TABLE = "internal_platform_db.dim_mq_topic_d_copilot_cand_query_engine_001"
+    DIFFICULTY = "EASY"
+    SOURCE_TABLES = ["dim_mq_topic_d_query_engine_001", "dim_mq_topic_d_copilot_query_engine_001"]
+    KEY_COLUMNS = ["business_id", "cluster_id", "dt"]
+    EXPECTED_COL_COUNT = 25
+    KEY_FIELDS = ["business_id", "business_name", "cluster_id", "cluster_name", "topic", "system_belong"]
+    GT_TABLE = OUTPUT_TABLE.replace("_cand_", "_")
+
+    DIFFICULTY_WEIGHTS = {
+        "EASY": (0.5, 0.5),
+        "MEDIUM": (0.6, 0.4),
+        "HARD": (0.7, 0.3),
+        "EXPERT": (0.8, 0.2),
+    }
+
+    _SPARK_SUBMIT_TIMEOUT = 300
+    _JSON_START = "__GRADE_JSON_START__"
+    _JSON_END = "__GRADE_JSON_END__"
+
+    result = {
+        "overall_score": 0.0,
+        "total_points": 0,
+        "grade": "",
+        "details": {},
+        "diagnostics": [],
+    }
+
+    # ========== 内部辅助函数 ==========
+
+    def values_match(pred_val, gt_val, abs_tol=1e-6, rel_tol=1e-4):
+        if pred_val is None and gt_val is None:
+            return True
+        if pred_val is None or gt_val is None:
+            return False
+        s_pred = str(pred_val).strip()
+        s_gt = str(gt_val).strip()
+        if s_pred == s_gt:
+            return True
+        try:
+            pv = float(s_pred)
+            gv = float(s_gt)
+            if math.isnan(pv) and math.isnan(gv):
+                return True
+            if math.isnan(pv) or math.isnan(gv):
+                return False
+            if abs(gv) < abs_tol:
+                return abs(pv - gv) <= abs_tol
+            return abs(pv - gv) <= abs_tol or abs(pv - gv) / max(abs(gv), 1e-12) <= rel_tol
+        except (ValueError, TypeError):
+            pass
+        return s_pred.lower() == s_gt.lower()
+
+    def _run_spark_script(script_code, timeout=_SPARK_SUBMIT_TIMEOUT):
+        with tempfile.NamedTemporaryFile(mode="w", suffix=".py", delete=False, encoding="utf-8") as f:
+            f.write(script_code)
+            script_path = f.name
+        try:
+            r = subprocess.run(
+                ["spark-submit", script_path],
+                capture_output=True, text=True, timeout=timeout,
+            )
+            stdout = r.stdout or ""
+            if _JSON_START in stdout and _JSON_END in stdout:
+                json_str = stdout.split(_JSON_START)[1].split(_JSON_END)[0].strip()
+                return json.loads(json_str), None
+            else:
+                if r.returncode == 0:
+                    for line in stdout.splitlines():
+                        if line.strip().startswith("Traceback"):
+                            return None, f"spark-submit error: {line}"
+                    return None, "spark-submit 无 JSON 输出"
+                err_msg = (r.stderr or "")[-500:]
+                return None, f"spark-submit failed: {err_msg}"
+        except subprocess.TimeoutExpired:
+            return None, f"spark-submit 超时 ({timeout}s)"
+        except Exception as e:
+            return None, f"spark-submit 异常: {e}"
+        finally:
+            try:
+                os.unlink(script_path)
+            except OSError:
+                pass
+
+    def read_table_via_spark_submit(table_name):
+        """Read table via spark-submit subprocess. Returns (cols, rows_as_lists)."""
+        if not re.match(r'^[a-zA-Z_][a-zA-Z0-9_.]*$', table_name):
+            raise ValueError(f"非法表名: {table_name}")
+        read_script = f'''
+import json
+from pyspark.sql import SparkSession
+spark = SparkSession.builder.appName("grade_read").enableHiveSupport() \\
+    .config("spark.sql.warehouse.dir", "/tmp/hive_warehouse").getOrCreate()
+try:
+    df = spark.sql("SELECT * FROM {table_name}")
+    cols = [c.lower() for c in df.columns]
+    rows = [[str(v) if v is not None else "" for v in row] for row in df.collect()]
+    print("{_JSON_START}")
+    print(json.dumps({{"cols": cols, "rows": rows}}, ensure_ascii=False))
+    print("{_JSON_END}")
+except Exception as e:
+    print("{_JSON_START}")
+    print(json.dumps({{"error": str(e)}}))
+    print("{_JSON_END}")
+finally:
+    spark.stop()
+'''
+        data, err = _run_spark_script(read_script)
+        if err:
+            raise RuntimeError(f"read_table failed: {err}")
+        if "error" in data:
+            raise RuntimeError(f"query failed: {data['error']}")
+        return data["cols"], data["rows"]
+
+    # SQL executor template (self-contained, no external dependency)
+    _SQL_EXEC_TEMPLATE = '''
+import json, re
+from pyspark.sql import SparkSession
+spark = SparkSession.builder.appName("{app_name}").enableHiveSupport() \\
+    .config("spark.sql.warehouse.dir", "/tmp/hive_warehouse").getOrCreate()
+try:
+    with open("{sql_file}", "r", encoding="utf-8") as _f:
+        _sql = _f.read()
+    _sql = re.sub(r"^\\s*set\\s+query_engine\\.\\S+\\n?", "", _sql, flags=re.IGNORECASE)
+    _stmts, _cur, _in_sq, _in_dq, _i = [], [], False, False, 0
+    while _i < len(_sql):
+        _ch = _sql[_i]
+        if _ch == "\\\\" and _i + 1 < len(_sql):
+            _cur.append(_ch); _cur.append(_sql[_i+1]); _i += 2; continue
+        if _ch == "-" and _i+1 < len(_sql) and _sql[_i+1] == "-" and not _in_sq and not _in_dq:
+            while _i < len(_sql) and _sql[_i] != "\\n": _i += 1
+            _cur.append("\\n"); continue
+        if _ch == "'" and not _in_dq: _in_sq = not _in_sq
+        elif _ch == '"' and not _in_sq: _in_dq = not _in_dq
+        if _ch == ";" and not _in_sq and not _in_dq:
+            _s = "".join(_cur).strip()
+            if _s: _stmts.append(_s)
+            _cur = []
+        else:
+            _cur.append(_ch)
+        _i += 1
+    _last = "".join(_cur).strip()
+    if _last: _stmts.append(_last)
+    for _stmt in _stmts:
+        spark.sql(_stmt)
+    print("{_JSON_START}")
+    print(json.dumps({{"ok": True}}))
+    print("{_JSON_END}")
+except Exception as e:
+    print("{_JSON_START}")
+    print(json.dumps({{"ok": False, "error": str(e)}}))
+    print("{_JSON_END}")
+finally:
+    spark.stop()
+'''
+
+    def execute_result_sql():
+        result_sql = os.path.join(workspace_path, "result.sql")
+        if not os.path.exists(result_sql):
+            return False, "no_result_file"
+        with open(result_sql, 'r', encoding='utf-8') as _rf:
+            _sql_text = _rf.read()
+        _bizdate = '20260507'
+        _sql_text = re.sub(r'\${bdp\.system\.bizdate(?:[+-]\d+)?}', _bizdate, _sql_text)
+        _yesterday = '20260506'
+        _bizdate_iso = '2026-05-07'
+        # 替换 数据平台WD变量（沙箱 spark-sql 不支持 ${...} 语法）
+        _sql_text = re.sub(r'\${bdp\.system\.bizdate}', _bizdate, _sql_text)
+        # yyyymmdd-1 必须在 yyyymmdd 之前匹配，否则会被通用正则吃掉
+        _sql_text = re.sub(r'\${yyyymmdd-1}', _yesterday, _sql_text)
+        _sql_text = re.sub(r'\${yyyymmdd(?:[+-]\d+)?}', _bizdate, _sql_text)
+        _sql_text = re.sub(r'\${bizdate-1}', _yesterday, _sql_text)
+        _sql_text = re.sub(r'\${bizdate(?:[+-]\d+)?}', _bizdate, _sql_text)
+        _sql_text = re.sub(r'\${[^}]*date[^}]*}', _bizdate, _sql_text)
+        # 替换所有未被处理的 ${...} 变量（兜底，防止 agent 使用任意变量名）
+        _sql_text = re.sub(r'\$\{[^}]+\}', _bizdate, _sql_text)
+        # Mock Spark SQL 内置时间函数，使其返回测试场景日期而非真实运行日期
+        _sql_text = re.sub(r'\bcurrent_date\s*\(\s*\)', f"date('{_bizdate_iso}')", _sql_text, flags=re.IGNORECASE)
+        _sql_text = re.sub(r'\bnow\s*\(\s*\)', f"timestamp('{_bizdate_iso} 00:00:00')", _sql_text, flags=re.IGNORECASE)
+        _sql_text = re.sub(r'\bcurrent_timestamp\s*\(\s*\)', f"timestamp('{_bizdate_iso} 00:00:00')", _sql_text, flags=re.IGNORECASE)
+        with open(result_sql, 'w', encoding='utf-8') as _wf:
+            _wf.write(_sql_text)
+        script = _SQL_EXEC_TEMPLATE.format(app_name="grade_exec", sql_file=result_sql,
+                                            _JSON_START=_JSON_START, _JSON_END=_JSON_END)
+        data, err = _run_spark_script(script)
+        if err:
+            return False, f"execution_error: {err}"
+        if data and data.get("ok"):
+            return True, None
+        return False, f"execution_error: {data.get('error', 'unknown') if data else 'no output'}"
+
+    def execute_ground_truth_sql():
+        gt_sql = os.path.join(workspace_path, "gt", "ground_truth.sql")
+        if not os.path.exists(gt_sql):
+            return False, "ground_truth.sql not found"
+        script = _SQL_EXEC_TEMPLATE.format(app_name="grade_gt", sql_file=gt_sql,
+                                            _JSON_START=_JSON_START, _JSON_END=_JSON_END)
+        data, err = _run_spark_script(script)
+        if err:
+            return False, f"gt_execution_error: {err}"
+        if data and data.get("ok"):
+            return True, None
+        return False, f"gt_execution_error: {data.get('error', 'unknown') if data else 'no output'}"
+
+    def truncate_table(table_name):
+        script = f'''
+from pyspark.sql import SparkSession
+spark = SparkSession.builder.appName("truncate").enableHiveSupport() \\
+    .config("spark.sql.warehouse.dir", "/tmp/hive_warehouse").getOrCreate()
+spark.sql("TRUNCATE TABLE {table_name}")
+spark.stop()
+'''
+        try:
+            _run_spark_script(script, timeout=120)
+        except Exception:
+            pass
+
+    def restore_hive_site():
+        """Restore hive-site.xml to canonical state (agent may have modified it)."""
+        canonical_hive_site = '''<?xml version="1.0"?>
+<?xml-stylesheet type="text/xsl" href="configuration.xsl"?>
+<configuration>
+  <property>
+    <name>hive.metastore.uris</name>
+    <value>thrift://localhost:9083</value>
+  </property>
+  <property>
+    <name>hive.metastore.warehouse.dir</name>
+    <value>/tmp/hive_warehouse</value>
+  </property>
+  <property>
+    <name>javax.jdo.option.ConnectionURL</name>
+    <value>jdbc:derby:;databaseName=/tmp/hive_metastore_db;create=true</value>
+  </property>
+  <property>
+    <name>javax.jdo.option.ConnectionDriverName</name>
+    <value>org.apache.derby.jdbc.EmbeddedDriver</value>
+  </property>
+  <property>
+    <name>datanucleus.schema.autoCreateAll</name>
+    <value>true</value>
+  </property>
+  <property>
+    <name>hive.metastore.schema.verification</name>
+    <value>false</value>
+  </property>
+</configuration>
+'''
+        hive_site_path = os.path.join(os.environ.get('SPARK_HOME', '/opt/spark'), 'conf', 'hive-site.xml')
+        try:
+            with open(hive_site_path, 'w') as f:
+                f.write(canonical_hive_site)
+        except Exception:
+            pass
+
+    def finalize(result):
+        product_weight, process_weight = DIFFICULTY_WEIGHTS.get(DIFFICULTY, (0.7, 0.3))
+        product_dims = ["A_executability"] + ['B_schema', 'C_row_alignment', 'D_field_completeness', 'D_field_value_match', 'F_insert_overwrite', 'F_partition_value', 'F_source_filter']
+        product_raw = sum(result["details"].get(d, {}).get("score", 0) for d in product_dims)
+        product_ratio = product_raw / 100.0
+        for dim in ["H_efficiency"]:
+            if dim in result["details"]:
+                raw = result["details"][dim].get("score", 0)
+                result["details"][dim]["score_before_scaling"] = raw
+                result["details"][dim]["score"] = round(raw * product_ratio, 2)
+                result["details"][dim]["product_ratio"] = round(product_ratio, 4)
+        process_dims = ["G_exploration", "H_efficiency", "I_self_verification"]
+        process_raw = sum(result["details"].get(d, {}).get("score", 0) for d in process_dims)
+        product_score = round(product_raw * product_weight, 2)
+        process_score = round(process_raw * process_weight, 2)
+        total = round(product_score + process_score, 2)
+        result["total_points"] = total
+        result["product_points"] = product_score
+        result["process_points"] = round(process_score, 2)
+        result["weights"] = {"product": product_weight, "process": process_weight}
+        result["overall_score"] = round(total / 100.0, 4)
+        if total >= 90:
+            result["grade"] = "优秀"
+        elif total >= 75:
+            result["grade"] = "良好"
+        elif total >= 60:
+            result["grade"] = "合格"
+        elif total >= 40:
+            result["grade"] = "偏弱"
+        else:
+            result["grade"] = "不合格"
+        return result
+
+    # Restore hive-site.xml (agent may have modified it)
+    restore_hive_site()
+
+    # ========== A. 可执行性 (15分) ==========
+    a_items = {"A1_exec_ok": 0, "A2_has_data": 0}
+    exec_ok, exec_err = execute_result_sql()
+    if not exec_ok:
+        detail = "未产出 result.sql" if exec_err == "no_result_file" else str(exec_err)[:200]
+        result["details"]["A_executability"] = {"score": 0, "max": 15, "detail": detail}
+        result["error"] = detail
+        for dim in ['B_schema', 'C_row_alignment', 'D_field_completeness', 'D_field_value_match', 'F_insert_overwrite', 'F_partition_value', 'F_source_filter']:
+            result["details"][dim] = {"score": 0, "max": 0, "items": {}}
+        return finalize(result)
+
+    a_items["A1_exec_ok"] = 8
+
+    pred_headers, pred_rows = [], []
+    try:
+        pred_headers, pred_rows = read_table_via_spark_submit(OUTPUT_TABLE)
+    except Exception as e:
+        result["diagnostics"].append(f"read_pred_failed: {e}")
+
+    if not pred_rows:
+        result["details"]["A_executability"] = {"score": 8, "max": 15, "items": a_items}
+        for dim in ['B_schema', 'C_row_alignment', 'D_field_completeness', 'D_field_value_match', 'F_insert_overwrite', 'F_partition_value', 'F_source_filter']:
+            result["details"][dim] = {"score": 0, "max": 0, "items": {}}
+        return finalize(result)
+
+    a_items["A2_has_data"] = 7
+    result["details"]["A_executability"] = {"score": 15, "max": 15, "items": a_items}
+
+    # ========== Execute GT + Read GT ==========
+    truncate_table(GT_TABLE)
+    gt_ok, gt_err = execute_ground_truth_sql()
+    if not gt_ok:
+        result["error"] = f"ground_truth failed: {gt_err}"
+        for dim in ['B_schema', 'C_row_alignment', 'D_field_completeness', 'D_field_value_match', 'F_insert_overwrite', 'F_partition_value', 'F_source_filter']:
+            result["details"][dim] = {"score": 0, "max": 0, "items": {}}
+        return finalize(result)
+
+    gt_headers, gt_rows = [], []
+    try:
+        gt_headers, gt_rows = read_table_via_spark_submit(GT_TABLE)
+    except Exception as e:
+        result["error"] = f"read_gt_failed: {e}"
+        for dim in ['B_schema', 'C_row_alignment', 'D_field_completeness', 'D_field_value_match', 'F_insert_overwrite', 'F_partition_value', 'F_source_filter']:
+            result["details"][dim] = {"score": 0, "max": 0, "items": {}}
+        return finalize(result)
+
+    # ========== B/C/D/F 维度评分 ==========
+    pred_col_map = {h: i for i, h in enumerate(pred_headers)}
+    gt_col_map = {h: i for i, h in enumerate(gt_headers)}
+
+    # ========== B. Schema正确性 (10分) ==========
+    b_items = {}
+    # B1: 列数 (5分)
+    if len(pred_headers) == EXPECTED_COL_COUNT:
+        b_items["B1_col_count"] = 5
+    elif abs(len(pred_headers) - EXPECTED_COL_COUNT) <= 2:
+        b_items["B1_col_count"] = 3
+    else:
+        b_items["B1_col_count"] = 0
+
+    # B2: 列名匹配 (5分)
+    gt_col_set = set(gt_headers)
+    pred_col_set = set(pred_headers)
+    name_match_rate = len(gt_col_set & pred_col_set) / max(len(gt_col_set), 1)
+    if name_match_rate >= 0.95:
+        b_items["B2_col_names"] = 5
+    elif name_match_rate >= 0.8:
+        b_items["B2_col_names"] = 3
+    else:
+        b_items["B2_col_names"] = 0
+
+    b_score = sum(b_items.values())
+    result["details"]["B_schema"] = {
+        "score": b_score, "max": 10,
+        "detail": {"col_count": len(pred_headers), "name_match_rate": round(name_match_rate, 4), "items": b_items},
+    }
+
+    # ========== C. 行一致性 (15分) ==========
+    c_items = {}
+    gt_row_count = len(gt_rows)
+    pred_row_count = len(pred_rows)
+
+    # C1: 行数比例 (7分)
+    if gt_row_count > 0:
+        ratio = pred_row_count / gt_row_count
+        if 0.95 <= ratio <= 1.05:
+            c_items["C1_row_ratio"] = 7
+        elif 0.7 <= ratio <= 1.3:
+            c_items["C1_row_ratio"] = 4
+        else:
+            c_items["C1_row_ratio"] = 0
+    else:
+        c_items["C1_row_ratio"] = 7 if pred_row_count == 0 else 0
+
+    # C2: key覆盖率 (8分)
+    key_cols_avail = [k for k in KEY_COLUMNS if k in gt_col_map and k in pred_col_map]
+    if key_cols_avail and gt_row_count > 0:
+        gt_keys = set()
+        for row in gt_rows:
+            key = tuple(row[gt_col_map[k]] if gt_col_map[k] < len(row) else "" for k in key_cols_avail)
+            gt_keys.add(key)
+        pred_keys = set()
+        for row in pred_rows:
+            key = tuple(row[pred_col_map[k]] if pred_col_map[k] < len(row) else "" for k in key_cols_avail)
+            pred_keys.add(key)
+        coverage = len(gt_keys & pred_keys) / max(len(gt_keys), 1)
+        if coverage >= 0.995:
+            c_items["C2_key_coverage"] = 8
+        elif coverage >= 0.9:
+            c_items["C2_key_coverage"] = 6
+        elif coverage >= 0.7:
+            c_items["C2_key_coverage"] = 3
+        else:
+            c_items["C2_key_coverage"] = round(8 * coverage, 2)
+    else:
+        c_items["C2_key_coverage"] = 0
+
+    c_score = sum(v for v in c_items.values())
+    result["details"]["C_row_alignment"] = {"score": c_score, "max": 15, "detail": c_items}
+
+    # 构建索引
+    if key_cols_avail:
+        pred_index = {}
+        for row in pred_rows:
+            key = tuple(row[pred_col_map[k]] if pred_col_map[k] < len(row) else "" for k in key_cols_avail)
+            pred_index[key] = row
+        gt_index = {}
+        for row in gt_rows:
+            key = tuple(row[gt_col_map[k]] if gt_col_map[k] < len(row) else "" for k in key_cols_avail)
+            gt_index[key] = row
+    else:
+        pred_index = {}
+        gt_index = {}
+
+    # ========== D. 字段值匹配 (25分) ==========
+    value_cols = [c for c in gt_headers if c in pred_col_map and c not in KEY_COLUMNS]
+    if not value_cols:
+        value_cols = [c for c in gt_headers if c in pred_col_map]
+
+    d_val_items = {}
+    per_col_weight = 25.0 / max(len(value_cols), 1)
+    d_val_score = 0
+    for col in value_cols:
+        gt_ci = gt_col_map.get(col)
+        pred_ci = pred_col_map.get(col)
+        if gt_ci is None or pred_ci is None:
+            d_val_items[col] = {"pass_rate": 0.0, "score": 0, "reason": "column_missing"}
+            continue
+        matches = 0
+        total = 0
+        for key, gt_row in gt_index.items():
+            pred_row = pred_index.get(key)
+            if pred_row is None:
+                total += 1
+                continue
+            gt_val = gt_row[gt_ci].strip() if gt_ci < len(gt_row) else ""
+            pred_val = pred_row[pred_ci].strip() if pred_ci < len(pred_row) else ""
+            if values_match(pred_val, gt_val):
+                matches += 1
+            total += 1
+        rate = matches / max(total, 1)
+        col_score = rate * per_col_weight
+        d_val_score += col_score
+        d_val_items[col] = {"pass_rate": round(rate, 4), "score": round(col_score, 2)}
+
+    result["details"]["D_field_value_match"] = {
+        "score": round(d_val_score, 2), "max": 25, "detail": d_val_items,
+    }
+
+    # ========== D. 字段完整性 (15分) ==========
+    d_comp_items = {}
+    d_comp_score = 0
+    per_field_weight = 15.0 / max(len(KEY_FIELDS), 1)
+    for field in KEY_FIELDS:
+        gt_ci = gt_col_map.get(field)
+        pred_ci = pred_col_map.get(field)
+        if gt_ci is None or pred_ci is None:
+            d_comp_items[field] = {"score": 0, "reason": "column_missing"}
+            continue
+        # GT 中该字段非空行数
+        gt_nonempty = 0
+        pred_nonempty = 0
+        total = 0
+        for key, gt_row in gt_index.items():
+            gt_val = gt_row[gt_ci].strip() if gt_ci < len(gt_row) else ""
+            if gt_val:
+                total += 1
+                pred_row = pred_index.get(key)
+                if pred_row is not None:
+                    pred_val = pred_row[pred_ci].strip() if pred_ci < len(pred_row) else ""
+                    if pred_val:
+                        pred_nonempty += 1
+        gt_nonempty = total
+        rate = pred_nonempty / max(gt_nonempty, 1)
+        field_score = rate * per_field_weight
+        d_comp_score += field_score
+        d_comp_items[field] = {"nonempty_rate": round(rate, 4), "score": round(field_score, 2)}
+
+    result["details"]["D_field_completeness"] = {
+        "score": round(d_comp_score, 2), "max": 15, "detail": d_comp_items,
+    }
+
+    # ========== F. INSERT OVERWRITE (5分) ==========
+    result_sql_path = os.path.join(workspace_path, "result.sql")
+    sql_text = ""
+    try:
+        with open(result_sql_path, "r", encoding="utf-8") as f:
+            sql_text = f.read().lower()
+    except Exception:
+        sql_text = ""
+
+    f_insert_items = {}
+    has_overwrite = "insert overwrite" in sql_text
+    has_partition = "partition" in sql_text
+    if has_overwrite and has_partition:
+        f_insert_items["insert_overwrite_partition"] = 5
+    elif has_overwrite:
+        f_insert_items["insert_overwrite_partition"] = 3
+    else:
+        f_insert_items["insert_overwrite_partition"] = 0
+    result["details"]["F_insert_overwrite"] = {
+        "score": f_insert_items["insert_overwrite_partition"], "max": 5, "detail": f_insert_items,
+    }
+
+    # ========== F. 分区值 (5分) ==========
+    f_part_items = {}
+    if "dt" in sql_text and "20260507" in sql_text:
+        f_part_items["partition_value"] = 5
+    else:
+        f_part_items["partition_value"] = 0
+    result["details"]["F_partition_value"] = {
+        "score": f_part_items["partition_value"], "max": 5, "detail": f_part_items,
+    }
+
+    # ========== F. 源表分区过滤 (10分) ==========
+    f_filter_items = {}
+    source_table_lower = SOURCE_TABLES[0].lower()
+    has_source_table = source_table_lower in sql_text
+    has_dt_filter = "dt" in sql_text and "20260506" in sql_text
+    if has_source_table and has_dt_filter:
+        f_filter_items["source_partition_filter"] = 10
+    elif has_dt_filter:
+        f_filter_items["source_partition_filter"] = 7
+    elif has_source_table:
+        f_filter_items["source_partition_filter"] = 3
+    else:
+        f_filter_items["source_partition_filter"] = 0
+    result["details"]["F_source_filter"] = {
+        "score": f_filter_items["source_partition_filter"], "max": 10, "detail": f_filter_items,
+    }
+
+    # 汇总
+    total = (15 + b_score + c_score + d_val_score + d_comp_score
+             + f_insert_items["insert_overwrite_partition"]
+             + f_part_items["partition_value"]
+             + f_filter_items["source_partition_filter"])
+
+    # ========== G~I 过程评分 ==========
+    TRANSCRIPT_PATH = "/tmp/dataclaw_chat.jsonl"
+    OUTPUT_TABLE_SHORT = OUTPUT_TABLE.split(".")[-1]
+    INPUT_TABLE_SHORT = SOURCE_TABLES[0]
+
+    transcript_entries = []
+    has_transcript = False
+    try:
+        if os.path.exists(TRANSCRIPT_PATH):
+            with open(TRANSCRIPT_PATH, "r", encoding="utf-8", errors="ignore") as f:
+                for line in f:
+                    line = line.strip()
+                    if line:
+                        try:
+                            transcript_entries.append(json.loads(line))
+                        except json.JSONDecodeError:
+                            continue
+            if len(transcript_entries) > 2:
+                has_transcript = True
+    except Exception:
+        pass
+
+    if not has_transcript:
+        result["details"]["G_exploration"] = {"score": 0, "max": 35, "items": {"no_transcript": True}}
+        result["details"]["H_efficiency"] = {"score": 0, "max": 40, "items": {"no_transcript": True}}
+        result["details"]["I_self_verification"] = {"score": 0, "max": 25, "items": {"no_transcript": True}}
+        return finalize(result)
+
+    # Parse transcript into structured events
+    tool_uses = []
+    first_write_result_idx = None
+    last_spark_submit_success_idx = None
+    write_result_count = 0
+    logic_error_retries = 0
+
+    for idx, entry in enumerate(transcript_entries):
+        content = entry.get("content", [])
+        if isinstance(content, str):
+            content = [content]
+
+        for block_str in content:
+            if not isinstance(block_str, str):
+                continue
+            if "ToolUseBlock" in block_str:
+                name_match = re.search(r"name='([^']+)'", block_str)
+                input_match = re.search(r"input=(\{.*\})", block_str)
+                if name_match:
+                    tool_name = name_match.group(1)
+                    tool_input = input_match.group(1) if input_match else ""
+                    tool_uses.append((idx, tool_name, tool_input))
+                    if tool_name == "Write" and "result.sql" in tool_input:
+                        write_result_count += 1
+                        if first_write_result_idx is None:
+                            first_write_result_idx = idx
+            if "ToolResultBlock" in block_str:
+                if "Traceback" in block_str or "Exception" in block_str:
+                    env_errors = ["Derby", "metastore", "HiveMetaStore", "Connection refused",
+                                  "db.lck", "TTransportException", "port 10000"]
+                    is_env_error = any(e in block_str for e in env_errors)
+                    has_spark_submit = any(t[1] == "Bash" and "spark-submit" in t[2] and "result.sql" in t[2]
+                                          for t in tool_uses)
+                    if not is_env_error and has_spark_submit:
+                        logic_error_retries += 1
+                if ("spark-submit" in block_str or "spark-sql" in block_str) and "result.sql" in block_str:
+                    if "Exit Code: 0" in block_str and "Traceback" not in block_str:
+                        last_spark_submit_success_idx = idx
+
+    before_first_write = first_write_result_idx if first_write_result_idx is not None else len(transcript_entries)
+
+    # ===== G. 探索充分性 (35分) =====
+    g_items = {}
+    g1_pass = any(name == "Read" and "schema" in inp.lower()
+                  for idx, name, inp in tool_uses if idx < before_first_write)
+    g_items["G1_source_schema"] = 9 if g1_pass else 0
+
+    g2_pass = any(name == "Bash" and INPUT_TABLE_SHORT in inp
+                  and "SELECT" in inp.upper() and "LIMIT" in inp.upper()
+                  for idx, name, inp in tool_uses if idx < before_first_write)
+    g_items["G2_source_sample"] = 9 if g2_pass else 0
+
+    g3_pass = any(name == "Bash" and INPUT_TABLE_SHORT in inp
+                  and ("GROUP BY" in inp.upper() or "DISTINCT" in inp.upper() or "COUNT" in inp.upper())
+                  for idx, name, inp in tool_uses if idx < before_first_write)
+    g_items["G3_distribution"] = 9 if g3_pass else 0
+
+    g4_pass = any(name == "Bash" and ("DESCRIBE" in inp.upper() or "SHOW CREATE" in inp.upper())
+                  and OUTPUT_TABLE_SHORT in inp
+                  for idx, name, inp in tool_uses if idx < before_first_write)
+    g_items["G4_target_schema"] = 8 if g4_pass else 0
+
+    g_score = sum(g_items.values())
+    result["details"]["G_exploration"] = {"score": g_score, "max": 35, "items": g_items}
+
+    # ===== H. 执行效率 (40分) =====
+    h_items = {}
+    if write_result_count <= 2:
+        h_items["H1_few_submissions"] = 20
+    elif write_result_count <= 4:
+        h_items["H1_few_submissions"] = 13
+    elif write_result_count <= 6:
+        h_items["H1_few_submissions"] = 7
+    else:
+        h_items["H1_few_submissions"] = 0
+
+    if logic_error_retries == 0:
+        h_items["H2_no_logic_errors"] = 13
+    elif logic_error_retries <= 1:
+        h_items["H2_no_logic_errors"] = 7
+    else:
+        h_items["H2_no_logic_errors"] = 0
+
+    h_items["H3_no_redundancy"] = 7
+    h_score = sum(h_items.values())
+    result["details"]["H_efficiency"] = {"score": min(h_score, 40), "max": 40, "items": h_items}
+
+    # ===== I. 自验证行为 (25分) =====
+    i_items = {}
+    post_submit_uses = []
+    if last_spark_submit_success_idx is not None:
+        post_submit_uses = [(idx, name, inp) for idx, name, inp in tool_uses
+                            if idx > last_spark_submit_success_idx]
+
+    i1_pass = any(name == "Bash" and OUTPUT_TABLE_SHORT in inp and "SELECT" in inp.upper()
+                  for _, name, inp in post_submit_uses)
+    i_items["I1_query_output"] = 8 if i1_pass else 0
+
+    i2_pass = any(name == "Bash" and OUTPUT_TABLE_SHORT in inp
+                  and ("COUNT" in inp.upper() or "GROUP BY" in inp.upper())
+                  for _, name, inp in post_submit_uses)
+    i_items["I2_check_count"] = 9 if i2_pass else 0
+
+    i3_pass = any(name == "Bash" and OUTPUT_TABLE_SHORT in inp
+                  and ("LIMIT" in inp.upper() or "SELECT *" in inp.upper())
+                  for _, name, inp in post_submit_uses)
+    i_items["I3_check_values"] = 8 if i3_pass else 0
+
+    i_score = sum(i_items.values())
+    result["details"]["I_self_verification"] = {"score": i_score, "max": 25, "items": i_items}
+
+    return finalize(result)

tasks/offline-compute/HiveSQL/hivesql_001_en/gt/grade_spec.csv

@@ -0,0 +1,19 @@
+维度,维度全名,子维度,满分,说明
+A,A_executability,executability,15,result.sql 能跑通且产出非空
+B,B_schema,schema,10,25列(5) + 列名匹配(5)
+C,C_row_alignment,row_consistency,15,行数比例(7) + key覆盖率(8)
+D,D_field_value_match,field_value_match,25,非key字段逐列值匹配率
+D,D_field_completeness,field_completeness,15,关键字段非空/非空串比例
+F,F_insert_overwrite,insert_overwrite,5,INSERT OVERWRITE + PARTITION
+F,F_partition_value,partition_value,5,dt 分区值 = 20260507
+F,F_source_filter,source_filter,10,源表分区过滤 dt=20260506
+
+总计,,,100,
+
+# 评分模板,评分范式: 产物100分 = A(15) + B(10) + C(15) + D(40) + F(20)
+# 难度,EASY
+# 权重,"product=0.5, process=0.5"
+# 范式,new
+# Key列,"business_id, cluster_id, dt"
+# 预期列数,25
+# 输出表,internal_platform_db.dim_mq_topic_d_copilot_cand_query_engine_001

tasks/offline-compute/HiveSQL/hivesql_001_en/gt/ground_truth.sql

@@ -0,0 +1,28 @@
+INSERT overwrite TABLE internal_platform_db.dim_mq_topic_d_copilot_query_engine_001 PARTITION (dt = '20260507')
+SELECT
+  business_id
+  ,business_name
+  ,cluster_set
+  ,tenant
+  ,namespaces
+  ,topic
+  ,mq_type
+  ,dw_appgroup
+  ,in_charge
+  ,description
+  ,create_time
+  ,modify_time
+  ,cluster_id
+  ,cluster_type
+  ,cluster_name
+  ,bg
+  ,category_name
+  ,is_filtered
+  ,tids
+  ,consumed_tids
+  ,unconsumed_tids
+  ,is_fully_consumed
+  ,has_unconsumed_tid
+  ,system_belong
+FROM internal_platform_db.dim_mq_topic_d_query_engine_001
+where dt = '20260506'

tasks/offline-compute/HiveSQL/hivesql_001_en/init/init_db.py

@@ -0,0 +1,53 @@
+#!/usr/bin/env python3
+import re
+from pyspark.sql import SparkSession
+
+spark = SparkSession.builder \
+    .appName('hivesql_bench_init') \
+    .enableHiveSupport() \
+    .config('spark.sql.warehouse.dir', '/tmp/hive_warehouse') \
+    .getOrCreate()
+
+spark.sql('CREATE DATABASE IF NOT EXISTS internal_platform_db')
+
+
+def _execute_sql_file(spark, sql_path):
+    """Read SQL file, remove SuperSQL SET headers, split by semicolons, execute."""
+    with open(sql_path, 'r', encoding='utf-8') as f:
+        content = f.read()
+    content = re.sub(r'^\s*set\s+query_engine\.\S+\n?', '', content, flags=re.IGNORECASE)
+    stmts, cur, in_sq, in_dq, i = [], [], False, False, 0
+    while i < len(content):
+        ch = content[i]
+        if ch == '\\' and i + 1 < len(content):
+            cur.append(ch); cur.append(content[i + 1]); i += 2; continue
+        if ch == '-' and i + 1 < len(content) and content[i + 1] == '-' and not in_sq and not in_dq:
+            while i < len(content) and content[i] != '\n':
+                i += 1
+            cur.append('\n'); continue
+        if ch == "'" and not in_dq:
+            in_sq = not in_sq
+        elif ch == '"' and not in_sq:
+            in_dq = not in_dq
+        if ch == ';' and not in_sq and not in_dq:
+            s = ''.join(cur).strip()
+            if s:
+                stmts.append(s)
+            cur = []
+        else:
+            cur.append(ch)
+        i += 1
+    last = ''.join(cur).strip()
+    if last:
+        stmts.append(last)
+    for stmt in stmts:
+        spark.sql(stmt)
+
+
+_execute_sql_file(spark, '/tmp_workspace/init_db.sql')
+
+tables = spark.sql('SHOW TABLES IN internal_platform_db').collect()
+print(f'Init complete, {len(tables)} tables created')
+for t in tables:
+    print(f'  - {t.namespace}.{t.tableName}')
+spark.stop()

tasks/offline-compute/HiveSQL/hivesql_001_en/init/init_db.sql

@@ -0,0 +1,90 @@
+CREATE DATABASE IF NOT EXISTS internal_platform_db;
+
+CREATE TABLE IF NOT EXISTS internal_platform_db.dim_mq_topic_d_query_engine_001 (
+  business_id STRING,
+  business_name STRING,
+  cluster_set STRING,
+  tenant STRING,
+  namespaces STRING,
+  topic STRING,
+  mq_type STRING,
+  dw_appgroup STRING,
+  in_charge STRING,
+  description STRING,
+  create_time STRING,
+  modify_time STRING,
+  cluster_id STRING,
+  cluster_type STRING,
+  cluster_name STRING,
+  bg STRING,
+  category_name STRING,
+  is_filtered TINYINT,
+  tids STRING,
+  consumed_tids STRING,
+  unconsumed_tids STRING,
+  is_fully_consumed TINYINT,
+  has_unconsumed_tid TINYINT,
+  system_belong STRING
+) PARTITIONED BY (dt STRING) STORED AS ORC;
+
+INSERT INTO TABLE internal_platform_db.dim_mq_topic_d_query_engine_001 PARTITION (dt='20260506') VALUES
+('bid001','BizName1','cluster_set_a','tenant_a','ns_a','topic_mq_1','PULSAR','appgroup1','user1','desc1','2026-01-01','2026-05-01','cid001','消息队列MQ','cluster_p1','BG1','product1',NULL,NULL,NULL,NULL,NULL,NULL,'数据总线'),
+('bid002','BizName2',NULL,NULL,NULL,'topic_tube_1','TUBEMQ','appgroup2','user2','desc2','2026-02-01','2026-05-02','cid002','TubeMQ','cluster_t1','BG2','product2',1,'tid1,tid2','tid1',NULL,0,1,'数据总线'),
+('bid003','BizName3',NULL,'tenant_b','ns_b','topic_inlong_1','PULSAR','appgroup3','user3',NULL,'2026-03-01','2026-05-03',NULL,NULL,'cluster_i1',NULL,'product3',NULL,NULL,NULL,NULL,NULL,NULL,'流处理平台'),
+('bid004','BizName4',NULL,NULL,NULL,'topic_tube_2','TUBEMQ','appgroup4','user4','desc4','2026-04-01','2026-05-04','cid004','TubeMQ','cluster_t2','BG1','product4',0,'tid3,tid4','tid3,tid4',NULL,1,0,'数据总线'),
+('bid005','BizName5','cluster_set_b','tenant_c','ns_c','topic_mq_2','PULSAR','appgroup5','user5','desc5','2026-05-01','2026-05-05','cid005','消息队列MQ','cluster_p2','BG3','product5',NULL,NULL,NULL,NULL,NULL,NULL,'流处理平台');
+
+CREATE TABLE IF NOT EXISTS internal_platform_db.dim_mq_topic_d_copilot_query_engine_001 (
+  business_id STRING,
+  business_name STRING,
+  cluster_set STRING,
+  tenant STRING,
+  namespaces STRING,
+  topic STRING,
+  mq_type STRING,
+  dw_appgroup STRING,
+  in_charge STRING,
+  description STRING,
+  create_time STRING,
+  modify_time STRING,
+  cluster_id STRING,
+  cluster_type STRING,
+  cluster_name STRING,
+  bg STRING,
+  category_name STRING,
+  is_filtered TINYINT,
+  tids STRING,
+  consumed_tids STRING,
+  unconsumed_tids STRING,
+  is_fully_consumed TINYINT,
+  has_unconsumed_tid TINYINT,
+  system_belong STRING
+) PARTITIONED BY (dt STRING) STORED AS ORC;
+
+-- Agent 候选目标表
+CREATE TABLE IF NOT EXISTS internal_platform_db.dim_mq_topic_d_copilot_cand_query_engine_001 (
+  business_id STRING,
+  business_name STRING,
+  cluster_set STRING,
+  tenant STRING,
+  namespaces STRING,
+  topic STRING,
+  mq_type STRING,
+  dw_appgroup STRING,
+  in_charge STRING,
+  description STRING,
+  create_time STRING,
+  modify_time STRING,
+  cluster_id STRING,
+  cluster_type STRING,
+  cluster_name STRING,
+  bg STRING,
+  category_name STRING,
+  is_filtered TINYINT,
+  tids STRING,
+  consumed_tids STRING,
+  unconsumed_tids STRING,
+  is_fully_consumed TINYINT,
+  has_unconsumed_tid TINYINT,
+  system_belong STRING
+) PARTITIONED BY (dt STRING) STORED AS ORC;

tasks/offline-compute/HiveSQL/hivesql_001_en/init/schema.sql

@@ -0,0 +1,27 @@
+CREATE TABLE IF NOT EXISTS internal_platform_db.dim_mq_topic_d_copilot (
+    `dt` STRING COMMENT '天分区',
+    `business_id` STRING COMMENT 'business_id',
+    `business_name` STRING COMMENT 'business_name',
+    `cluster_set` STRING COMMENT 'cluster_set',
+    `tenant` STRING COMMENT 'tenant',
+    `namespaces` STRING COMMENT 'namespaces',
+    `topic` STRING COMMENT 'topic',
+    `mq_type` STRING COMMENT 'MQ类型（消息队列MQ/TubeMQ）',
+    `dw_appgroup` STRING COMMENT 'bid所属应用组',
+    `in_charge` STRING COMMENT 'bid负责人',
+    `description` STRING COMMENT 'bid描述信息',
+    `create_time` STRING COMMENT 'bid创建时间',
+    `modify_time` STRING COMMENT 'bid最后修改时间',
+    `cluster_id` STRING COMMENT 'bid所属集群id',
+    `cluster_type` STRING COMMENT 'bid所属集群类型',
+    `cluster_name` STRING COMMENT 'bid所属集群名',
+    `bg` STRING COMMENT 'bid所属的bg',
+    `category_name` STRING COMMENT 'bid所归属的产品',
+    `is_filtered` TINYINT COMMENT 'Tube-bid是否有消费组是过滤消费(1:是 0:否)',
+    `tids` STRING COMMENT 'Tube-bid下的所有tid',
+    `consumed_tids` STRING COMMENT 'Tube-bid下所有配置消费的tid',
+    `unconsumed_tids` STRING COMMENT 'Tube-bid下所有未配置消费的tid',
+    `is_fully_consumed` TINYINT COMMENT 'Tube-topic是否有消费组是全量消费(1:是 0:否)',
+    `has_unconsumed_tid` TINYINT COMMENT 'Tube-bid是否有没有被消费的tid',
+    `system_belong` STRING COMMENT 'topic所属的系统（数据总线/流处理平台）'
+) STORED AS ORC;

tasks/offline-compute/HiveSQL/hivesql_001_en/task.md

@@ -0,0 +1,25 @@
+---
+id: offline-compute_HiveSQL_hivesql_001
+name: Message Queue Topic Dimension Table internal_platform_db.dim_mq_topic_d_su
+category: offline-compute/HiveSQL
+timeout_seconds: 600
+modality: pure-text
+engine: hivesql
+---
+## Prompt
+**Task Objective**: Read data from the previous day's partition of the input table and copy it as-is to the output table's current day partition.
+
+**Time Variables**: The platform provides these variables for dynamic date computation:
+- `${yyyymmdd}` : current day in YYYYMMDD format
+- `${yyyymmdd-1}` : previous day in YYYYMMDD format
+- You may also use Spark SQL built-in functions like `current_date()` and `date_sub()`.
+
+**Input**: `internal_platform_db.dim_mq_topic_d_query_engine_001` (a partitioned table, partitioned by `dt`).
+
+**Processing Rules**: 1) Select all data where the `dt` partition equals the previous day (use `${yyyymmdd-1}`); 2) No joins, single-table processing; 3) All fields are retained as-is, with no transformations or filtering.
+
+**Output Requirements**: Output all non-partition columns: `business_id`, `business_name`, `cluster_set`, `tenant`, `namespaces`, `topic`, `mq_type`, `dw_appgroup`, `in_charge`, `description`, `create_time`, `modify_time`, `cluster_id`, `cluster_type`, `cluster_name`, `bg`, `category_name`, `is_filtered`, `tids`, `consumed_tids`, `unconsumed_tids`, `is_fully_consumed`, `has_unconsumed_tid`, `system_belong`; partitioned by the `dt` field.
+
+**Write Requirements**: Use `INSERT OVERWRITE` to write to the current day partition (use `${yyyymmdd}`) of `internal_platform_db.dim_mq_topic_d_copilot_cand_query_engine_001`.
+
+Please write the final HiveSQL to `result.sql` and execute it.

tasks/offline-compute/HiveSQL/hivesql_002/gt/expected.csv

@@ -0,0 +1,4 @@
+trace_id,datawd_project_id,datawd_task_id,datawd_task_instance_id,compute_type,status_code,instance_run_time,code_run_time,resource_wait_time,code_start_time,code_end_time,instance_start_time,instance_end_time,serving_id,is_permanent,apply_for_gpu_count,pod_name,pkg_agg_time,gpu_util,gpu_count,p_date,dt
+trace_001,proj_1,task_1,inst_1,ray,0,86400,75600,2400,2025-05-06 01:00:00,2025-05-06 22:00:00,2025-05-06 00:00:00,2025-05-07 00:00:00,1001,true,2,pod_alpha,,,,2025-05-06,20260507
+trace_001,proj_1,task_1,inst_1,ray,0,0,,,,,,2025-05-06 00:00:00,2025-05-07 00:00:00,1001,true,2,pod_alpha,,,,2025-05-07,20260507
+trace_002,proj_2,task_2,inst_2,ray,1,10800,8000,800,2025-05-07 00:13:20,2025-05-07 02:26:40,2025-05-07 00:00:00,2025-05-07 03:00:00,1002,false,2,pod_beta,,,,2025-05-07,20260507

tasks/offline-compute/HiveSQL/hivesql_002/gt/grade.py

@@ -0,0 +1,754 @@
+"""hivesql_005 精细评分脚本
+
+业务场景：Notebook 实例跨天 Pod 详情（多表JOIN + 时间切分）
+难度: HARD | 特征: 6表5JOIN + CTE + UNION + 跨天时间计算
+
+评分范式: 产物100分 = A(10) + B(10) + C(10) + D(55) + F(15)
+  A_executability       (10分): result.sql 能跑通且产出非空
+  B_schema              (10分): 22列 + 列名匹配
+  C_row_alignment     (10分): 行数比例 + key覆盖率
+  D_time_calculation    (30分): instance_run_time, code_run_time 等时间指标
+  D_cross_day_split     (15分): trace_001 应出现在两天中
+  D_dimension_join      (10分): serving_id, pod_name 等维度列匹配
+  F_join_completeness   (5分):  多表都被关联
+  F_insert_overwrite    (5分):  写入模式+分区
+  F_partition_value     (5分):  分区值正确
+
+权重: HARD → product=0.7, process=0.3
+"""
+import os
+import re
+import subprocess
+import tempfile
+import json
+import math
+
+
+def grade(workspace_path, **kwargs):
+
+    # ========== Case 配置 ==========
+    OUTPUT_TABLE = "internal_platform_db.dwd_notebook_instance_pod_cross_day_detail_d_cand_query_engine_005"
+    DIFFICULTY = "HARD"
+    SOURCE_TABLES = [
+        "notebook_span_info_query_engine_005",
+        "dwd_gputj_service_instance_map_query_engine_005",
+        "dwd_ml_platform_instance_podname_query_engine_005",
+        "gputj_gpu_info_parsed_agg_1min_query_engine_005",
+        "notebook_engine_info_query_engine_005",
+    ]
+    KEY_COLUMNS = ["trace_id", "p_date"]
+    EXPECTED_TRACE_DATES = {
+        "trace_001": ["2025-05-06", "2025-05-07"],
+        "trace_002": ["2025-05-07"],
+    }
+    TIME_COLUMNS = ["instance_run_time", "code_run_time", "resource_wait_time"]
+    DIM_COLUMNS = ["serving_id", "is_permanent", "apply_for_gpu_count", "pod_name"]
+    SOURCE_TABLE_SHORT_NAMES = [
+        "notebook_span_info",
+        "dwd_gputj_service_instance_map",
+        "dwd_ml_platform_instance_podname",
+        "gputj_gpu_info_parsed_agg_1min",
+        "notebook_engine_info",
+    ]
+    GT_TABLE = OUTPUT_TABLE.replace("_cand_", "_")
+
+    DIFFICULTY_WEIGHTS = {
+        "EASY": (0.5, 0.5),
+        "MEDIUM": (0.6, 0.4),
+        "HARD": (0.7, 0.3),
+        "EXPERT": (0.8, 0.2),
+    }
+
+    _SPARK_SUBMIT_TIMEOUT = 300
+    _JSON_START = "__GRADE_JSON_START__"
+    _JSON_END = "__GRADE_JSON_END__"
+
+    result = {
+        "overall_score": 0.0,
+        "total_points": 0,
+        "grade": "",
+        "details": {},
+        "diagnostics": [],
+    }
+
+    # ========== 内部辅助函数 ==========
+
+    def values_match(pred_val, gt_val, abs_tol=1e-6, rel_tol=1e-4):
+        if pred_val is None and gt_val is None:
+            return True
+        if pred_val is None or gt_val is None:
+            return False
+        s_pred = str(pred_val).strip()
+        s_gt = str(gt_val).strip()
+        if s_pred == s_gt:
+            return True
+        try:
+            pv = float(s_pred)
+            gv = float(s_gt)
+            if math.isnan(pv) and math.isnan(gv):
+                return True
+            if math.isnan(pv) or math.isnan(gv):
+                return False
+            if abs(gv) < abs_tol:
+                return abs(pv - gv) <= abs_tol
+            return abs(pv - gv) <= abs_tol or abs(pv - gv) / max(abs(gv), 1e-12) <= rel_tol
+        except (ValueError, TypeError):
+            pass
+        return s_pred.lower() == s_gt.lower()
+
+    def _run_spark_script(script_code, timeout=_SPARK_SUBMIT_TIMEOUT):
+        with tempfile.NamedTemporaryFile(mode="w", suffix=".py", delete=False, encoding="utf-8") as f:
+            f.write(script_code)
+            script_path = f.name
+        try:
+            r = subprocess.run(
+                ["spark-submit", script_path],
+                capture_output=True, text=True, timeout=timeout,
+            )
+            stdout = r.stdout or ""
+            if _JSON_START in stdout and _JSON_END in stdout:
+                json_str = stdout.split(_JSON_START)[1].split(_JSON_END)[0].strip()
+                return json.loads(json_str), None
+            else:
+                if r.returncode == 0:
+                    for line in stdout.splitlines():
+                        if line.strip().startswith("Traceback"):
+                            return None, f"spark-submit error: {line}"
+                    return None, "spark-submit 无 JSON 输出"
+                err_msg = (r.stderr or "")[-500:]
+                return None, f"spark-submit failed: {err_msg}"
+        except subprocess.TimeoutExpired:
+            return None, f"spark-submit 超时 ({timeout}s)"
+        except Exception as e:
+            return None, f"spark-submit 异常: {e}"
+        finally:
+            try:
+                os.unlink(script_path)
+            except OSError:
+                pass
+
+    def read_table_via_spark_submit(table_name):
+        """Read table via spark-submit subprocess. Returns (cols, rows_as_lists)."""
+        if not re.match(r'^[a-zA-Z_][a-zA-Z0-9_.]*$', table_name):
+            raise ValueError(f"非���表名: {table_name}")
+        read_script = f'''
+import json
+from pyspark.sql import SparkSession
+spark = SparkSession.builder.appName("grade_read").enableHiveSupport() \\
+    .config("spark.sql.warehouse.dir", "/tmp/hive_warehouse").getOrCreate()
+try:
+    df = spark.sql("SELECT * FROM {table_name}")
+    cols = [c.lower() for c in df.columns]
+    rows = [[str(v) if v is not None else "" for v in row] for row in df.collect()]
+    print("{_JSON_START}")
+    print(json.dumps({{"cols": cols, "rows": rows}}, ensure_ascii=False))
+    print("{_JSON_END}")
+except Exception as e:
+    print("{_JSON_START}")
+    print(json.dumps({{"error": str(e)}}))
+    print("{_JSON_END}")
+finally:
+    spark.stop()
+'''
+        data, err = _run_spark_script(read_script)
+        if err:
+            raise RuntimeError(f"read_table failed: {err}")
+        if "error" in data:
+            raise RuntimeError(f"query failed: {data['error']}")
+        return data["cols"], data["rows"]
+
+    # SQL executor template (self-contained, no external dependency)
+    _SQL_EXEC_TEMPLATE = '''
+import json, re
+from pyspark.sql import SparkSession
+spark = SparkSession.builder.appName("{app_name}").enableHiveSupport() \\
+    .config("spark.sql.warehouse.dir", "/tmp/hive_warehouse").getOrCreate()
+try:
+    with open("{sql_file}", "r", encoding="utf-8") as _f:
+        _sql = _f.read()
+    _sql = re.sub(r"^\\s*set\\s+query_engine\\.\\S+\\n?", "", _sql, flags=re.IGNORECASE)
+    _stmts, _cur, _in_sq, _in_dq, _i = [], [], False, False, 0
+    while _i < len(_sql):
+        _ch = _sql[_i]
+        if _ch == "\\\\" and _i + 1 < len(_sql):
+            _cur.append(_ch); _cur.append(_sql[_i+1]); _i += 2; continue
+        if _ch == "-" and _i+1 < len(_sql) and _sql[_i+1] == "-" and not _in_sq and not _in_dq:
+            while _i < len(_sql) and _sql[_i] != "\\n": _i += 1
+            _cur.append("\\n"); continue
+        if _ch == "'" and not _in_dq: _in_sq = not _in_sq
+        elif _ch == '"' and not _in_sq: _in_dq = not _in_dq
+        if _ch == ";" and not _in_sq and not _in_dq:
+            _s = "".join(_cur).strip()
+            if _s: _stmts.append(_s)
+            _cur = []
+        else:
+            _cur.append(_ch)
+        _i += 1
+    _last = "".join(_cur).strip()
+    if _last: _stmts.append(_last)
+    for _stmt in _stmts:
+        spark.sql(_stmt)
+    print("{_JSON_START}")
+    print(json.dumps({{"ok": True}}))
+    print("{_JSON_END}")
+except Exception as e:
+    print("{_JSON_START}")
+    print(json.dumps({{"ok": False, "error": str(e)}}))
+    print("{_JSON_END}")
+finally:
+    spark.stop()
+'''
+
+    def execute_result_sql():
+        result_sql = os.path.join(workspace_path, "result.sql")
+        if not os.path.exists(result_sql):
+            return False, "no_result_file"
+        # 替换 数据平台WD时间变量（沙箱 spark-sql 不支持 ${...} 语法）
+        with open(result_sql, 'r', encoding='utf-8') as _rf:
+            _sql_text = _rf.read()
+        _bizdate = '20260507'
+        _sql_text = re.sub(r'\${bdp\.system\.bizdate(?:[+-]\d+)?}', _bizdate, _sql_text)
+        _sql_text = re.sub(r'\${yyyymmdd(?:[+-]\d+)?}', _bizdate, _sql_text)
+        _sql_text = re.sub(r'\${bizdate(?:[+-]\d+)?}', _bizdate, _sql_text)
+        _sql_text = re.sub(r'\${[^}]*date[^}]*}', _bizdate, _sql_text)
+        with open(result_sql, 'w', encoding='utf-8') as _wf:
+            _wf.write(_sql_text)
+        script = _SQL_EXEC_TEMPLATE.format(app_name="grade_exec", sql_file=result_sql,
+                                            _JSON_START=_JSON_START, _JSON_END=_JSON_END)
+        data, err = _run_spark_script(script)
+        if err:
+            return False, f"execution_error: {err}"
+        if data and data.get("ok"):
+            return True, None
+        return False, f"execution_error: {data.get('error', 'unknown') if data else 'no output'}"
+
+    def execute_ground_truth_sql():
+        gt_sql = os.path.join(workspace_path, "gt", "ground_truth.sql")
+        if not os.path.exists(gt_sql):
+            return False, "ground_truth.sql not found"
+        script = _SQL_EXEC_TEMPLATE.format(app_name="grade_gt", sql_file=gt_sql,
+                                            _JSON_START=_JSON_START, _JSON_END=_JSON_END)
+        data, err = _run_spark_script(script)
+        if err:
+            return False, f"gt_execution_error: {err}"
+        if data and data.get("ok"):
+            return True, None
+        return False, f"gt_execution_error: {data.get('error', 'unknown') if data else 'no output'}"
+
+    def truncate_table(table_name):
+        script = f'''
+from pyspark.sql import SparkSession
+spark = SparkSession.builder.appName("truncate").enableHiveSupport() \\
+    .config("spark.sql.warehouse.dir", "/tmp/hive_warehouse").getOrCreate()
+spark.sql("TRUNCATE TABLE {table_name}")
+spark.stop()
+'''
+        try:
+            _run_spark_script(script, timeout=120)
+        except Exception:
+            pass
+
+    def restore_hive_site():
+        """Restore hive-site.xml to canonical state (agent may have modified it)."""
+        canonical_hive_site = '''<?xml version="1.0"?>
+<?xml-stylesheet type="text/xsl" href="configuration.xsl"?>
+<configuration>
+  <property>
+    <name>hive.metastore.uris</name>
+    <value>thrift://localhost:9083</value>
+  </property>
+  <property>
+    <name>hive.metastore.warehouse.dir</name>
+    <value>/tmp/hive_warehouse</value>
+  </property>
+  <property>
+    <name>javax.jdo.option.ConnectionURL</name>
+    <value>jdbc:derby:;databaseName=/tmp/hive_metastore_db;create=true</value>
+  </property>
+  <property>
+    <name>javax.jdo.option.ConnectionDriverName</name>
+    <value>org.apache.derby.jdbc.EmbeddedDriver</value>
+  </property>
+  <property>
+    <name>datanucleus.schema.autoCreateAll</name>
+    <value>true</value>
+  </property>
+  <property>
+    <name>hive.metastore.schema.verification</name>
+    <value>false</value>
+  </property>
+</configuration>
+'''
+        hive_site_path = os.path.join(os.environ.get('SPARK_HOME', '/opt/spark'), 'conf', 'hive-site.xml')
+        try:
+            with open(hive_site_path, 'w') as f:
+                f.write(canonical_hive_site)
+        except Exception:
+            pass
+
+    def finalize(result):
+        product_weight, process_weight = DIFFICULTY_WEIGHTS.get(DIFFICULTY, (0.7, 0.3))
+        product_dims = ["A_executability"] + ['B_schema', 'C_row_alignment', 'D_cross_day_split', 'D_dimension_join', 'D_time_calculation', 'F_insert_overwrite', 'F_join_completeness', 'F_partition_value']
+        product_raw = sum(result["details"].get(d, {}).get("score", 0) for d in product_dims)
+        product_ratio = product_raw / 100.0
+        for dim in ["H_efficiency"]:
+            if dim in result["details"]:
+                raw = result["details"][dim].get("score", 0)
+                result["details"][dim]["score_before_scaling"] = raw
+                result["details"][dim]["score"] = round(raw * product_ratio, 2)
+                result["details"][dim]["product_ratio"] = round(product_ratio, 4)
+        process_dims = ["G_exploration", "H_efficiency", "I_self_verification"]
+        process_raw = sum(result["details"].get(d, {}).get("score", 0) for d in process_dims)
+        product_score = round(product_raw * product_weight, 2)
+        process_score = round(process_raw * process_weight, 2)
+        total = round(product_score + process_score, 2)
+        result["total_points"] = total
+        result["product_points"] = product_score
+        result["process_points"] = round(process_score, 2)
+        result["weights"] = {"product": product_weight, "process": process_weight}
+        result["overall_score"] = round(total / 100.0, 4)
+        if total >= 90:
+            result["grade"] = "优秀"
+        elif total >= 75:
+            result["grade"] = "良好"
+        elif total >= 60:
+            result["grade"] = "合格"
+        elif total >= 40:
+            result["grade"] = "偏弱"
+        else:
+            result["grade"] = "不合格"
+        return result
+
+    # Restore hive-site.xml (agent may have modified it)
+    restore_hive_site()
+
+    # ========== A. 可执行性 (10分) ==========
+    a_items = {"A1_exec_ok": 0, "A2_has_data": 0}
+    exec_ok, exec_err = execute_result_sql()
+    if not exec_ok:
+        detail = "未产出 result.sql" if exec_err == "no_result_file" else str(exec_err)[:200]
+        result["details"]["A_executability"] = {"score": 0, "max": 10, "detail": detail}
+        result["error"] = detail
+        for dim in ['B_schema', 'C_row_alignment', 'D_cross_day_split', 'D_dimension_join', 'D_time_calculation', 'F_insert_overwrite', 'F_join_completeness', 'F_partition_value']:
+            result["details"][dim] = {"score": 0, "max": 0, "items": {}}
+        return finalize(result)
+
+    a_items["A1_exec_ok"] = 5
+
+    pred_headers, pred_rows = [], []
+    try:
+        pred_headers, pred_rows = read_table_via_spark_submit(OUTPUT_TABLE)
+    except Exception as e:
+        result["diagnostics"].append(f"read_pred_failed: {e}")
+
+    if not pred_rows:
+        result["details"]["A_executability"] = {"score": 5, "max": 10, "items": a_items}
+        for dim in ['B_schema', 'C_row_alignment', 'D_cross_day_split', 'D_dimension_join', 'D_time_calculation', 'F_insert_overwrite', 'F_join_completeness', 'F_partition_value']:
+            result["details"][dim] = {"score": 0, "max": 0, "items": {}}
+        return finalize(result)
+
+    a_items["A2_has_data"] = 5
+    result["details"]["A_executability"] = {"score": 10, "max": 10, "items": a_items}
+
+    # ========== Execute GT + Read GT ==========
+    truncate_table(GT_TABLE)
+    gt_ok, gt_err = execute_ground_truth_sql()
+    if not gt_ok:
+        result["error"] = f"ground_truth failed: {gt_err}"
+        for dim in ['B_schema', 'C_row_alignment', 'D_cross_day_split', 'D_dimension_join', 'D_time_calculation', 'F_insert_overwrite', 'F_join_completeness', 'F_partition_value']:
+            result["details"][dim] = {"score": 0, "max": 0, "items": {}}
+        return finalize(result)
+
+    gt_headers, gt_rows = [], []
+    try:
+        gt_headers, gt_rows = read_table_via_spark_submit(GT_TABLE)
+    except Exception as e:
+        result["error"] = f"read_gt_failed: {e}"
+        for dim in ['B_schema', 'C_row_alignment', 'D_cross_day_split', 'D_dimension_join', 'D_time_calculation', 'F_insert_overwrite', 'F_join_completeness', 'F_partition_value']:
+            result["details"][dim] = {"score": 0, "max": 0, "items": {}}
+        return finalize(result)
+
+    # ========== B/C/D/F 维度评分 ==========
+    pred_col_map = {h: i for i, h in enumerate(pred_headers)}
+    gt_col_map = {h: i for i, h in enumerate(gt_headers)}
+
+    # ========== B. Schema正确性 (10分) ==========
+    b_items = {}
+    expected_col_count = 22
+    if len(pred_headers) == expected_col_count:
+        b_items["B1_col_count"] = 5
+    elif abs(len(pred_headers) - expected_col_count) <= 2:
+        b_items["B1_col_count"] = 3
+    else:
+        b_items["B1_col_count"] = 0
+
+    gt_col_set = set(gt_headers)
+    pred_col_set = set(pred_headers)
+    name_match_rate = len(gt_col_set & pred_col_set) / max(len(gt_col_set), 1)
+    if name_match_rate >= 0.95:
+        b_items["B2_col_names"] = 5
+    elif name_match_rate >= 0.8:
+        b_items["B2_col_names"] = 3
+    else:
+        b_items["B2_col_names"] = 0
+
+    b_score = sum(b_items.values())
+    result["details"]["B_schema"] = {
+        "score": b_score, "max": 10,
+        "detail": {"col_count": len(pred_headers), "name_match_rate": round(name_match_rate, 4), "items": b_items},
+    }
+
+    # ========== C. 行一致性 (10分) ==========
+    c_items = {}
+    gt_row_count = len(gt_rows)
+    pred_row_count = len(pred_rows)
+
+    if gt_row_count > 0:
+        ratio = pred_row_count / gt_row_count
+        if 0.95 <= ratio <= 1.05:
+            c_items["C1_row_ratio"] = 5
+        elif 0.7 <= ratio <= 1.3:
+            c_items["C1_row_ratio"] = 3
+        else:
+            c_items["C1_row_ratio"] = 0
+    else:
+        c_items["C1_row_ratio"] = 5 if pred_row_count == 0 else 0
+
+    key_cols_avail = [k for k in KEY_COLUMNS if k in gt_col_map and k in pred_col_map]
+    if key_cols_avail and gt_row_count > 0:
+        gt_keys = set()
+        for row in gt_rows:
+            key = tuple(row[gt_col_map[k]] if gt_col_map[k] < len(row) else "" for k in key_cols_avail)
+            gt_keys.add(key)
+        pred_keys = set()
+        for row in pred_rows:
+            key = tuple(row[pred_col_map[k]] if pred_col_map[k] < len(row) else "" for k in key_cols_avail)
+            pred_keys.add(key)
+        coverage = len(gt_keys & pred_keys) / max(len(gt_keys), 1)
+        if coverage >= 0.995:
+            c_items["C2_key_coverage"] = 5
+        elif coverage >= 0.9:
+            c_items["C2_key_coverage"] = 3
+        elif coverage >= 0.7:
+            c_items["C2_key_coverage"] = 2
+        else:
+            c_items["C2_key_coverage"] = round(5 * coverage, 2)
+    else:
+        c_items["C2_key_coverage"] = 0
+
+    c_score = sum(v for v in c_items.values())
+    result["details"]["C_row_alignment"] = {"score": c_score, "max": 10, "detail": c_items}
+
+    # 构建 GT/pred 索引
+    gt_index = {}
+    gt_trace_idx = gt_col_map.get("trace_id")
+    gt_pdate_idx = gt_col_map.get("p_date")
+    if gt_trace_idx is not None and gt_pdate_idx is not None:
+        for row in gt_rows:
+            tid = row[gt_trace_idx].strip() if gt_trace_idx < len(row) else ""
+            pd = row[gt_pdate_idx].strip() if gt_pdate_idx < len(row) else ""
+            gt_index[(tid, pd)] = row
+
+    pred_index = {}
+    trace_idx = pred_col_map.get("trace_id")
+    pdate_idx = pred_col_map.get("p_date")
+    if trace_idx is not None and pdate_idx is not None:
+        for row in pred_rows:
+            if trace_idx >= len(row) or pdate_idx >= len(row):
+                continue
+            tid = row[trace_idx].strip()
+            pd = row[pdate_idx].strip()
+            pred_index[(tid, pd)] = row
+
+    # ========== D. 时间计算正确性 (30分) ==========
+    d_time_items = {}
+    per_col_weight = 30.0 / max(len(TIME_COLUMNS), 1)
+    for col in TIME_COLUMNS:
+        gt_ci = gt_col_map.get(col)
+        pred_ci = pred_col_map.get(col)
+        if gt_ci is None or pred_ci is None:
+            d_time_items[col] = {"pass_rate": 0.0, "score": 0, "reason": "column_missing"}
+            continue
+
+        matches = 0
+        total = 0
+        for key, gt_row in gt_index.items():
+            pred_row = pred_index.get(key)
+            if pred_row is None:
+                total += 1
+                continue
+            gt_val = gt_row[gt_ci].strip() if gt_ci < len(gt_row) else ""
+            pred_val = pred_row[pred_ci].strip() if pred_ci < len(pred_row) else ""
+            if values_match(pred_val, gt_val, abs_tol=1.0, rel_tol=0.01):
+                matches += 1
+            total += 1
+
+        rate = matches / max(total, 1)
+        col_score = rate * per_col_weight
+        d_time_items[col] = {"pass_rate": round(rate, 4), "score": round(col_score, 2)}
+
+    d_time_score = sum(v.get("score", 0) for v in d_time_items.values() if isinstance(v, dict))
+    result["details"]["D_time_calculation"] = {"score": round(d_time_score, 2), "max": 30, "detail": d_time_items}
+
+    # ========== D. 跨天切分正确性 (15分) ==========
+    d_cross_items = {}
+    d_cross_score = 0
+
+    if trace_idx is not None and pdate_idx is not None:
+        pred_trace_dates = {}
+        for row in pred_rows:
+            tid = row[trace_idx].strip() if trace_idx < len(row) else ""
+            pd = row[pdate_idx].strip() if pdate_idx < len(row) else ""
+            if tid not in pred_trace_dates:
+                pred_trace_dates[tid] = set()
+            pred_trace_dates[tid].add(pd)
+
+        # trace_001 应跨两天 (8分)
+        t001_dates = pred_trace_dates.get("trace_001", set())
+        if len(t001_dates) >= 2:
+            d_cross_score += 8
+            d_cross_items["trace_001_cross_day"] = True
+        elif len(t001_dates) == 1:
+            d_cross_score += 3
+            d_cross_items["trace_001_cross_day"] = f"only {t001_dates}"
+        else:
+            d_cross_items["trace_001_cross_day"] = False
+
+        # trace_002 只出现一天 (4分)
+        t002_dates = pred_trace_dates.get("trace_002", set())
+        if len(t002_dates) == 1:
+            d_cross_score += 4
+            d_cross_items["trace_002_single_day"] = True
+        else:
+            d_cross_items["trace_002_single_day"] = f"dates={t002_dates}"
+
+        # 行数正确 (3分)
+        if pred_row_count == gt_row_count:
+            d_cross_score += 3
+            d_cross_items["row_count_correct"] = True
+        elif abs(pred_row_count - gt_row_count) <= 1:
+            d_cross_score += 2
+            d_cross_items["row_count_correct"] = f"off by {abs(pred_row_count - gt_row_count)}"
+    else:
+        d_cross_items["error"] = "trace_id or p_date column missing"
+
+    result["details"]["D_cross_day_split"] = {"score": d_cross_score, "max": 15, "detail": d_cross_items}
+
+    # ========== D. 维度列关联正确性 (10分) ==========
+    d_dim_items = {}
+    for col in DIM_COLUMNS:
+        gt_ci = gt_col_map.get(col)
+        pred_ci = pred_col_map.get(col)
+        if gt_ci is None or pred_ci is None:
+            continue
+        matches = 0
+        total = 0
+        for key, gt_row in gt_index.items():
+            pred_row = pred_index.get(key)
+            if pred_row is None:
+                total += 1
+                continue
+            gt_val = gt_row[gt_ci].strip() if gt_ci < len(gt_row) else ""
+            pred_val = pred_row[pred_ci].strip() if pred_ci < len(pred_row) else ""
+            if values_match(pred_val, gt_val):
+                matches += 1
+            total += 1
+        rate = matches / max(total, 1)
+        d_dim_items[col] = round(rate, 4)
+
+    d_dim_score = 0
+    if d_dim_items:
+        avg_rate = sum(d_dim_items.values()) / len(d_dim_items)
+        d_dim_score = round(10 * avg_rate, 2)
+    result["details"]["D_dimension_join"] = {"score": d_dim_score, "max": 10, "detail": d_dim_items}
+
+    # ========== F. JOIN 完整性 (5分) ==========
+    f_join_items = {}
+    result_sql_path = os.path.join(workspace_path, "result.sql")
+    sql_text = ""
+    try:
+        with open(result_sql_path, "r", encoding="utf-8") as f:
+            sql_text = f.read().lower()
+    except Exception:
+        sql_text = ""
+
+    tables_found = 0
+    for tname in SOURCE_TABLES:
+        if tname.lower() in sql_text:
+            tables_found += 1
+            f_join_items[tname] = True
+        else:
+            f_join_items[tname] = False
+
+    f_join_score = round(5 * tables_found / max(len(SOURCE_TABLES), 1), 2)
+    result["details"]["F_join_completeness"] = {
+        "score": f_join_score, "max": 5, "detail": f_join_items,
+    }
+
+    # ========== F. INSERT OVERWRITE (5分) ==========
+    f_insert_items = {}
+    has_overwrite = "insert overwrite" in sql_text
+    has_partition = "partition" in sql_text
+    if has_overwrite and has_partition:
+        f_insert_items["insert_overwrite_partition"] = 5
+    elif has_overwrite:
+        f_insert_items["insert_overwrite_partition"] = 3
+    else:
+        f_insert_items["insert_overwrite_partition"] = 0
+    result["details"]["F_insert_overwrite"] = {
+        "score": f_insert_items["insert_overwrite_partition"], "max": 5, "detail": f_insert_items,
+    }
+
+    # ========== F. 分区值 (5分) ==========
+    f_part_items = {}
+    if ("dt" in sql_text and ("20260507" in sql_text or "2026050" in sql_text)) or "p_date" in sql_text:
+        f_part_items["partition_value"] = 5
+    else:
+        f_part_items["partition_value"] = 0
+    result["details"]["F_partition_value"] = {
+        "score": f_part_items["partition_value"], "max": 5, "detail": f_part_items,
+    }
+
+    # 汇总
+    total = 10 + b_score + c_score + d_time_score + d_cross_score + d_dim_score + f_join_score + f_insert_items["insert_overwrite_partition"] + f_part_items["partition_value"]
+
+    # ========== G~I 过程评分 ==========
+    TRANSCRIPT_PATH = "/tmp/dataclaw_chat.jsonl"
+    OUTPUT_TABLE_SHORT = OUTPUT_TABLE.split(".")[-1]
+    INPUT_TABLE_SHORT = SOURCE_TABLES[0]
+
+    transcript_entries = []
+    has_transcript = False
+    try:
+        if os.path.exists(TRANSCRIPT_PATH):
+            with open(TRANSCRIPT_PATH, "r", encoding="utf-8", errors="ignore") as f:
+                for line in f:
+                    line = line.strip()
+                    if line:
+                        try:
+                            transcript_entries.append(json.loads(line))
+                        except json.JSONDecodeError:
+                            continue
+            if len(transcript_entries) > 2:
+                has_transcript = True
+    except Exception:
+        pass
+
+    if not has_transcript:
+        result["details"]["G_exploration"] = {"score": 0, "max": 35, "items": {"no_transcript": True}}
+        result["details"]["H_efficiency"] = {"score": 0, "max": 40, "items": {"no_transcript": True}}
+        result["details"]["I_self_verification"] = {"score": 0, "max": 25, "items": {"no_transcript": True}}
+        return finalize(result)
+
+    # Parse transcript into structured events
+    tool_uses = []
+    first_write_result_idx = None
+    last_spark_submit_success_idx = None
+    write_result_count = 0
+    logic_error_retries = 0
+
+    for idx, entry in enumerate(transcript_entries):
+        content = entry.get("content", [])
+        if isinstance(content, str):
+            content = [content]
+
+        for block_str in content:
+            if not isinstance(block_str, str):
+                continue
+            if "ToolUseBlock" in block_str:
+                name_match = re.search(r"name='([^']+)'", block_str)
+                input_match = re.search(r"input=(\{.*\})", block_str)
+                if name_match:
+                    tool_name = name_match.group(1)
+                    tool_input = input_match.group(1) if input_match else ""
+                    tool_uses.append((idx, tool_name, tool_input))
+                    if tool_name == "Write" and "result.sql" in tool_input:
+                        write_result_count += 1
+                        if first_write_result_idx is None:
+                            first_write_result_idx = idx
+            if "ToolResultBlock" in block_str:
+                if "Traceback" in block_str or "Exception" in block_str:
+                    env_errors = ["Derby", "metastore", "HiveMetaStore", "Connection refused",
+                                  "db.lck", "TTransportException", "port 10000"]
+                    is_env_error = any(e in block_str for e in env_errors)
+                    has_spark_submit = any(t[1] == "Bash" and "spark-submit" in t[2] and "result.sql" in t[2]
+                                          for t in tool_uses)
+                    if not is_env_error and has_spark_submit:
+                        logic_error_retries += 1
+                if ("spark-submit" in block_str or "spark-sql" in block_str) and "result.sql" in block_str:
+                    if "Exit Code: 0" in block_str and "Traceback" not in block_str:
+                        last_spark_submit_success_idx = idx
+
+    before_first_write = first_write_result_idx if first_write_result_idx is not None else len(transcript_entries)
+
+    # ===== G. 探索充分性 (35分) =====
+    g_items = {}
+    g1_pass = any(name == "Read" and "schema" in inp.lower()
+                  for idx, name, inp in tool_uses if idx < before_first_write)
+    g_items["G1_source_schema"] = 9 if g1_pass else 0
+
+    g2_pass = any(name == "Bash" and INPUT_TABLE_SHORT in inp
+                  and "SELECT" in inp.upper() and "LIMIT" in inp.upper()
+                  for idx, name, inp in tool_uses if idx < before_first_write)
+    g_items["G2_source_sample"] = 9 if g2_pass else 0
+
+    g3_pass = any(name == "Bash" and INPUT_TABLE_SHORT in inp
+                  and ("GROUP BY" in inp.upper() or "DISTINCT" in inp.upper() or "COUNT" in inp.upper())
+                  for idx, name, inp in tool_uses if idx < before_first_write)
+    g_items["G3_distribution"] = 9 if g3_pass else 0
+
+    g4_pass = any(name == "Bash" and ("DESCRIBE" in inp.upper() or "SHOW CREATE" in inp.upper())
+                  and OUTPUT_TABLE_SHORT in inp
+                  for idx, name, inp in tool_uses if idx < before_first_write)
+    g_items["G4_target_schema"] = 8 if g4_pass else 0
+
+    g_score = sum(g_items.values())
+    result["details"]["G_exploration"] = {"score": g_score, "max": 35, "items": g_items}
+
+    # ===== H. 执行效率 (40分) =====
+    h_items = {}
+    if write_result_count <= 2:
+        h_items["H1_few_submissions"] = 20
+    elif write_result_count <= 4:
+        h_items["H1_few_submissions"] = 13
+    elif write_result_count <= 6:
+        h_items["H1_few_submissions"] = 7
+    else:
+        h_items["H1_few_submissions"] = 0
+
+    if logic_error_retries == 0:
+        h_items["H2_no_logic_errors"] = 13
+    elif logic_error_retries <= 1:
+        h_items["H2_no_logic_errors"] = 7
+    else:
+        h_items["H2_no_logic_errors"] = 0
+
+    h_items["H3_no_redundancy"] = 7
+    h_score = sum(h_items.values())
+    result["details"]["H_efficiency"] = {"score": min(h_score, 40), "max": 40, "items": h_items}
+
+    # ===== I. 自验证行为 (25分) =====
+    i_items = {}
+    post_submit_uses = []
+    if last_spark_submit_success_idx is not None:
+        post_submit_uses = [(idx, name, inp) for idx, name, inp in tool_uses
+                            if idx > last_spark_submit_success_idx]
+
+    i1_pass = any(name == "Bash" and OUTPUT_TABLE_SHORT in inp and "SELECT" in inp.upper()
+                  for _, name, inp in post_submit_uses)
+    i_items["I1_query_output"] = 8 if i1_pass else 0
+
+    i2_pass = any(name == "Bash" and OUTPUT_TABLE_SHORT in inp
+                  and ("COUNT" in inp.upper() or "GROUP BY" in inp.upper())
+                  for _, name, inp in post_submit_uses)
+    i_items["I2_check_count"] = 9 if i2_pass else 0
+
+    i3_pass = any(name == "Bash" and OUTPUT_TABLE_SHORT in inp
+                  and ("LIMIT" in inp.upper() or "SELECT *" in inp.upper())
+                  for _, name, inp in post_submit_uses)
+    i_items["I3_check_values"] = 8 if i3_pass else 0
+
+    i_score = sum(i_items.values())
+    result["details"]["I_self_verification"] = {"score": i_score, "max": 25, "items": i_items}
+
+    return finalize(result)

tasks/offline-compute/HiveSQL/hivesql_002/gt/grade_spec.csv

@@ -0,0 +1,20 @@
+维度,维度全名,子维度,满分,说明
+A,A_executability,executability,10,result.sql 能跑通且产出非空
+B,B_schema,schema,10,22列 + 列名匹配
+C,C_row_alignment,row_consistency,10,行数比例 + key覆盖率
+D,D_time_calculation,time_calculation,30,"instance_run_time, code_run_time 等时间指标"
+D,D_cross_day_split,cross_day_split,15,trace_001 应出现在两天中
+D,D_dimension_join,dimension_join,10,"serving_id, pod_name 等维度列匹配"
+F,F_join_completeness,join_completeness,5,多表都被关联
+F,F_insert_overwrite,insert_overwrite,5,写入模式+分区
+F,F_partition_value,partition_value,5,分区值正确
+
+总计,,,100,
+
+# 评分模板,评分范式: 产物100分 = A(10) + B(10) + C(10) + D(55) + F(15)
+# 难度,HARD
+# 权重,"product=0.7, process=0.3"
+# 范式,new
+# Key列,"trace_id, p_date"
+# 预期列数,
+# 输出表,internal_platform_db.dwd_notebook_instance_pod_cross_day_detail_d_cand_query_engine_005

tasks/offline-compute/HiveSQL/hivesql_002/gt/ground_truth.sql

@@ -0,0 +1,191 @@
+INSERT overwrite TABLE internal_platform_db.dwd_notebook_instance_pod_cross_day_detail_d_query_engine_005 PARTITION (dt = '20260507')
+WITH
+base_data_raw AS (
+    SELECT DISTINCT
+        trace_id,
+        span_name,
+        start_time,
+        end_time,
+        datawd_project_id,
+        datawd_task_id,
+        datawd_task_instance_id,
+        compute_type,
+        status_code
+    FROM internal_platform_db.notebook_span_info_query_engine_005
+    WHERE databus_imp_date >= '2026050400'
+      AND databus_imp_date <= '2026050700'
+      AND trace_id IN (
+        SELECT trace_id
+        FROM internal_platform_db.notebook_span_info_query_engine_005
+        WHERE databus_imp_date >= '2026050700'
+          AND databus_imp_date <= '2026050700'
+          AND compute_type = 'ray'
+          AND service_name = 'notebook-runner'
+          AND span_name = 'runner.execute'
+        GROUP BY trace_id
+      )
+      AND (
+          span_name = 'runner.execute'
+          OR
+          span_name IN ('execute.code', 'execute.code.cell', 'client.execute.code', 'set.permanent.compute', 'create.non.permanent.compute', 'runner.killed')
+      )
+),
+base_data_time_fixed AS (
+    SELECT
+        trace_id,
+        span_name,
+        CASE
+            WHEN span_name = 'runner.killed'
+              THEN MIN(CASE WHEN span_name IN ('execute.code', 'execute.code.cell', 'client.execute.code', 'runner.killed') THEN start_time END)
+                      OVER(PARTITION BY trace_id)
+            ELSE start_time
+        END AS start_time,
+        CASE
+            WHEN span_name = 'runner.execute'
+              THEN MAX(CASE WHEN span_name IN ('runner.execute', 'runner.killed') THEN end_time END)
+                      OVER(PARTITION BY trace_id)
+            ELSE end_time
+        END AS end_time,
+        datawd_project_id,
+        datawd_task_id,
+        datawd_task_instance_id,
+        compute_type,
+        status_code
+    FROM base_data_raw
+),
+base_data AS (
+  SELECT
+      trace_id,
+      span_name,
+      start_time,
+      end_time,
+      datawd_project_id,
+      datawd_task_id,
+      datawd_task_instance_id,
+      compute_type,
+      status_code,
+      from_unixtime(CAST(start_time AS BIGINT) / 1000) as start_date,
+      from_unixtime(CAST(end_time AS BIGINT) / 1000) as end_date,
+      datediff(from_unixtime(CAST(end_time AS BIGINT) / 1000), from_unixtime(CAST(start_time AS BIGINT) / 1000)) AS diff_days
+  FROM base_data_time_fixed
+),
+pos_series AS (
+    SELECT 0 AS pos UNION ALL SELECT 1 AS pos UNION ALL SELECT 2 AS pos UNION ALL SELECT 3 AS pos
+),
+daily_split_spans AS (
+    SELECT
+        trace_id,
+        span_name,
+        datawd_project_id,
+        datawd_task_id,
+        datawd_task_instance_id,
+        compute_type,
+        status_code,
+        start_date,
+        end_date,
+        diff_days,
+        date_add(b.start_date, s.pos) AS calc_date,
+        CASE WHEN s.pos = 0 THEN CAST(b.start_time AS BIGINT)
+             ELSE unix_timestamp(cast(date_add(b.start_date, s.pos) as timestamp)) * 1000
+        END AS start_time,
+        CASE WHEN s.pos = b.diff_days THEN CAST(b.end_time AS BIGINT)
+             ELSE (unix_timestamp(cast(date_add(b.start_date, s.pos + 1) as timestamp)) * 1000) - 1
+        END AS end_time,
+        CAST(b.start_time AS BIGINT) AS span_start_time,
+        CAST(b.end_time AS BIGINT) AS span_end_time
+    FROM base_data b
+    INNER JOIN pos_series s ON s.pos <= b.diff_days
+),
+trace_time_metrics AS (
+    SELECT
+        trace_id,
+        calc_date,
+        MAX(datawd_project_id) AS datawd_project_id,
+        MAX(datawd_task_id) AS datawd_task_id,
+        MAX(datawd_task_instance_id) AS datawd_task_instance_id,
+        MAX(compute_type) AS compute_type,
+        MAX(MAX(CASE WHEN span_name = 'runner.execute' THEN status_code END)) OVER(PARTITION BY trace_id) AS status_code,
+        ROUND((MAX(CASE WHEN span_name = 'runner.execute' THEN end_time END) -
+               MIN(CASE WHEN span_name = 'runner.execute' THEN start_time END)) / 1000.0) AS instance_run_time,
+        ROUND((MAX(CASE WHEN span_name IN ('execute.code', 'client.execute.code', 'execute.code.cell', 'runner.killed' ) THEN end_time END) -
+               MIN(CASE WHEN span_name IN ('execute.code', 'client.execute.code', 'execute.code.cell', 'runner.killed' ) THEN start_time END)) / 1000.0) AS code_run_time,
+        ROUND((MAX(CASE WHEN span_name IN ('set.permanent.compute', 'create.non.permanent.compute') THEN end_time END) -
+               MIN(CASE WHEN span_name IN ('set.permanent.compute', 'create.non.permanent.compute') THEN start_time END)) / 1000.0) AS resource_wait_time,
+        from_unixtime(MIN(CASE WHEN span_name IN ('execute.code', 'client.execute.code', 'execute.code.cell', 'runner.killed' ) THEN start_time END) / 1000) AS code_start_time,
+        from_unixtime(MAX(CASE WHEN span_name IN ('execute.code', 'client.execute.code', 'execute.code.cell', 'runner.killed' ) THEN end_time END) / 1000) AS code_end_time,
+        from_unixtime(MIN(CASE WHEN span_name = 'runner.execute' THEN span_start_time END) / 1000) AS instance_start_time,
+        from_unixtime(MAX(CASE WHEN span_name = 'runner.execute' THEN span_end_time END) / 1000) AS instance_end_time
+    FROM daily_split_spans
+    GROUP BY trace_id, calc_date
+),
+gputj_dim AS (
+  SELECT
+    t1.service_id,
+    t2.pod_name
+  FROM internal_platform_db.dwd_gputj_service_instance_map_query_engine_005 t1
+  INNER JOIN internal_platform_db.dwd_ml_platform_instance_podname_query_engine_005 t2
+    ON t1.instance_uuid = t2.instance_uuid
+    AND t1.dt=if('2026050700'>='2025080900', '2026050700', '2025080900')
+    AND t2.dt=if('2026050700'>='2025060519', '2026050700', '2025060519')
+  GROUP BY t1.service_id, t2.pod_name
+),
+gpu_metrics AS (
+  select
+    pkg_agg_time,
+    pod_name,
+    k8s_container_vgpu_gpu_util_sum / k8s_container_vgpu_gpu_util_count AS gpu_util,
+    k8s_container_resource_request_gpu_sum / k8s_container_resource_request_gpu_count AS gpu_count
+  from internal_platform_db.gputj_gpu_info_parsed_agg_1min_query_engine_005
+  where dt >= '2026050400'
+    AND dt <= '2026050700'
+    AND ( k8s_container_vgpu_gpu_util_count > 0 OR k8s_container_resource_request_gpu_count > 0 )
+    AND pod_name IS NOT NULL
+)
+SELECT
+  t1.trace_id,
+  t1.datawd_project_id,
+  t1.datawd_task_id,
+  t1.datawd_task_instance_id,
+  t1.compute_type,
+  t1.status_code,
+  t1.instance_run_time,
+  t1.code_run_time,
+  t1.resource_wait_time,
+  t1.code_start_time,
+  t1.code_end_time,
+  t1.instance_start_time,
+  t1.instance_end_time,
+  t2.serving_id,
+  t2.is_permanent,
+  t2.apply_for_gpu_count,
+  t3.pod_name,
+  t4.pkg_agg_time,
+  t4.gpu_util,
+  t4.gpu_count,
+  t1.calc_date AS p_date
+FROM trace_time_metrics t1 LEFT JOIN (
+  SELECT
+    trace_id,
+    MAX(serving_id) AS serving_id,
+    MAX(is_permanent) AS is_permanent,
+    SUM(CAST(replicas AS INT) * CAST(num_gpu AS INT)) AS apply_for_gpu_count
+  FROM internal_platform_db.notebook_engine_info_query_engine_005
+  WHERE databus_imp_date >= '2026050400'
+    AND databus_imp_date <= '2026050700'
+    AND compute_type = 'ray'
+    AND service_name = 'notebook-runner'
+  GROUP BY trace_id
+) t2 ON t1.trace_id = t2.trace_id
+LEFT JOIN gputj_dim t3 ON t2.serving_id = t3.service_id
+LEFT JOIN gpu_metrics t4 
+  ON t3.pod_name = t4.pod_name
+  AND (
+    CASE
+      WHEN t1.instance_start_time IS NOT NULL AND t1.instance_end_time IS NOT NULL
+        THEN t4.pkg_agg_time >= t1.instance_start_time AND t4.pkg_agg_time <= t1.instance_end_time
+      WHEN t1.code_start_time IS NOT NULL AND t1.code_end_time IS NOT NULL
+        THEN t4.pkg_agg_time >= t1.code_start_time AND t4.pkg_agg_time <= t1.code_end_time
+      ELSE FALSE
+    END
+  )
+;

tasks/offline-compute/HiveSQL/hivesql_002/init/init_db.py

@@ -0,0 +1,53 @@
+#!/usr/bin/env python3
+import re
+from pyspark.sql import SparkSession
+
+spark = SparkSession.builder \
+    .appName('hivesql_bench_init') \
+    .enableHiveSupport() \
+    .config('spark.sql.warehouse.dir', '/tmp/hive_warehouse') \
+    .getOrCreate()
+
+spark.sql('CREATE DATABASE IF NOT EXISTS internal_platform_db')
+
+
+def _execute_sql_file(spark, sql_path):
+    """Read SQL file, remove SuperSQL SET headers, split by semicolons, execute."""
+    with open(sql_path, 'r', encoding='utf-8') as f:
+        content = f.read()
+    content = re.sub(r'^\s*set\s+query_engine\.\S+\n?', '', content, flags=re.IGNORECASE)
+    stmts, cur, in_sq, in_dq, i = [], [], False, False, 0
+    while i < len(content):
+        ch = content[i]
+        if ch == '\\' and i + 1 < len(content):
+            cur.append(ch); cur.append(content[i + 1]); i += 2; continue
+        if ch == '-' and i + 1 < len(content) and content[i + 1] == '-' and not in_sq and not in_dq:
+            while i < len(content) and content[i] != '\n':
+                i += 1
+            cur.append('\n'); continue
+        if ch == "'" and not in_dq:
+            in_sq = not in_sq
+        elif ch == '"' and not in_sq:
+            in_dq = not in_dq
+        if ch == ';' and not in_sq and not in_dq:
+            s = ''.join(cur).strip()
+            if s:
+                stmts.append(s)
+            cur = []
+        else:
+            cur.append(ch)
+        i += 1
+    last = ''.join(cur).strip()
+    if last:
+        stmts.append(last)
+    for stmt in stmts:
+        spark.sql(stmt)
+
+
+_execute_sql_file(spark, '/tmp_workspace/init_db.sql')
+
+tables = spark.sql('SHOW TABLES IN internal_platform_db').collect()
+print(f'Init complete, {len(tables)} tables created')
+for t in tables:
+    print(f'  - {t.namespace}.{t.tableName}')
+spark.stop()

tasks/offline-compute/HiveSQL/hivesql_002/init/init_db.sql

@@ -0,0 +1,164 @@
+CREATE DATABASE IF NOT EXISTS internal_platform_db;
+
+-- Input table 1: notebook_span_info
+CREATE TABLE IF NOT EXISTS internal_platform_db.notebook_span_info_query_engine_005 (
+    databus_imp_date STRING,
+    trace_id STRING,
+    span_kind STRING,
+    error_code STRING,
+    service_name STRING,
+    message STRING,
+    parent_span_id STRING,
+    span_name STRING,
+    duration BIGINT,
+    span_id STRING,
+    space_id STRING,
+    span_type STRING,
+    start_time STRING,
+    end_time STRING,
+    service_instance STRING,
+    status_code INT,
+    datawd_project_id STRING,
+    datawd_task_id STRING,
+    datawd_task_instance_id STRING,
+    compute_type STRING
+) STORED AS ORC;
+
+INSERT INTO TABLE internal_platform_db.notebook_span_info_query_engine_005 VALUES
+('2026050700','trace_001','server','0','notebook-runner','','','runner.execute',60000,'s1','sp1','','1746460800000','1746547200000','inst1',0,'proj_1','task_1','inst_1','ray'),
+('2026050700','trace_001','server','0','notebook-runner','','','execute.code',50000,'s2','sp1','','1746464400000','1746540000000','inst1',0,'proj_1','task_1','inst_1','ray'),
+('2026050700','trace_001','server','0','notebook-runner','','','set.permanent.compute',5000,'s3','sp1','','1746460800000','1746463200000','inst1',0,'proj_1','task_1','inst_1','ray'),
+('2026050500','trace_001','server','0','notebook-runner','','','runner.execute',60000,'s4','sp1','','1746460800000','1746547200000','inst1',0,'proj_1','task_1','inst_1','ray'),
+('2026050500','trace_001','server','0','notebook-runner','','','execute.code',50000,'s5','sp1','','1746464400000','1746540000000','inst1',0,'proj_1','task_1','inst_1','ray'),
+('2026050700','trace_002','server','0','notebook-runner','','','runner.execute',30000,'s6','sp2','','1746547200000','1746558000000','inst2',1,'proj_2','task_2','inst_2','ray'),
+('2026050700','trace_002','server','0','notebook-runner','','','execute.code',25000,'s7','sp2','','1746548000000','1746556000000','inst2',1,'proj_2','task_2','inst_2','ray'),
+('2026050700','trace_002','server','0','notebook-runner','','','create.non.permanent.compute',3000,'s8','sp2','','1746547200000','1746548000000','inst2',1,'proj_2','task_2','inst_2','ray');
+
+-- Input table 2: dwd_gputj_service_instance_map
+CREATE TABLE IF NOT EXISTS internal_platform_db.dwd_gputj_service_instance_map_query_engine_005 (
+    dt STRING,
+    service_name STRING,
+    instance_uuid STRING,
+    service_id BIGINT,
+    version_id BIGINT,
+    `desc` STRING,
+    create_time DOUBLE,
+    update_time DOUBLE,
+    workload_name STRING
+) STORED AS ORC;
+
+INSERT INTO TABLE internal_platform_db.dwd_gputj_service_instance_map_query_engine_005 VALUES
+('2026050700','svc_a','uuid_001',1001,1,'desc1',1.0,2.0,'wl1'),
+('2026050700','svc_b','uuid_002',1002,1,'desc2',1.0,2.0,'wl2');
+
+-- Input table 3: dwd_ml_platform_instance_podname
+CREATE TABLE IF NOT EXISTS internal_platform_db.dwd_ml_platform_instance_podname_query_engine_005 (
+    dt STRING,
+    instance_uuid STRING,
+    pod_name STRING,
+    pod_phase STRING,
+    pod_create_time BIGINT,
+    pod_update_time BIGINT,
+    namespace STRING,
+    trial_job_name STRING,
+    create_time BIGINT,
+    modify_time BIGINT,
+    fix_modify_time BIGINT
+) STORED AS ORC;
+
+INSERT INTO TABLE internal_platform_db.dwd_ml_platform_instance_podname_query_engine_005 VALUES
+('2026050700','uuid_001','pod_alpha','Running',1746460000,1746461000,'ns1','job1',1746460000,1746461000,1746461000),
+('2026050700','uuid_002','pod_beta','Running',1746460000,1746461000,'ns2','job2',1746460000,1746461000,1746461000);
+
+-- Input table 4: gputj_gpu_info_parsed_agg_1min (MINIMAL - only columns GT uses)
+CREATE TABLE IF NOT EXISTS internal_platform_db.gputj_gpu_info_parsed_agg_1min_query_engine_005 (
+    dt STRING,
+    pod_name STRING,
+    pkg_agg_time STRING,
+    k8s_container_vgpu_gpu_util_sum DOUBLE,
+    k8s_container_vgpu_gpu_util_count BIGINT,
+    k8s_container_resource_request_gpu_sum DOUBLE,
+    k8s_container_resource_request_gpu_count BIGINT
+) STORED AS ORC;
+
+INSERT INTO TABLE internal_platform_db.gputj_gpu_info_parsed_agg_1min_query_engine_005 VALUES
+('2026050700','pod_alpha','2026-05-06 02:00:00',160.0,2,4.0,2),
+('2026050600','pod_alpha','2026-05-05 20:00:00',140.0,2,2.0,2);
+
+-- Input table 5: notebook_engine_info
+CREATE TABLE IF NOT EXISTS internal_platform_db.notebook_engine_info_query_engine_005 (
+    databus_imp_date STRING,
+    trace_id STRING,
+    service_name STRING,
+    compute_type STRING,
+    datawd_project_id STRING,
+    datawd_task_id STRING,
+    datawd_task_instance_id STRING,
+    is_permanent STRING,
+    serving_id STRING,
+    user_name STRING,
+    is_model_service STRING,
+    compute_info STRING,
+    num_gpu STRING,
+    replicas STRING,
+    gpu_type STRING,
+    `timestamp` STRING,
+    ray_job_id STRING,
+    reason STRING,
+    log_url STRING,
+    dashboard_url STRING
+) STORED AS ORC;
+
+INSERT INTO TABLE internal_platform_db.notebook_engine_info_query_engine_005 VALUES
+('2026050700','trace_001','notebook-runner','ray','proj_1','task_1','inst_1','true','1001','user1','false','{}','2','1','A100','1746460800','job1','','',''),
+('2026050700','trace_002','notebook-runner','ray','proj_2','task_2','inst_2','false','1002','user2','false','{}','1','2','V100','1746547200','job2','','','');
+
+-- Target table
+CREATE TABLE IF NOT EXISTS internal_platform_db.dwd_notebook_instance_pod_cross_day_detail_d_query_engine_005 (
+    trace_id STRING,
+    datawd_project_id STRING,
+    datawd_task_id STRING,
+    datawd_task_instance_id STRING,
+    compute_type STRING,
+    status_code INT,
+    instance_run_time INT,
+    code_run_time INT,
+    resource_wait_time INT,
+    code_start_time STRING,
+    code_end_time STRING,
+    instance_start_time STRING,
+    instance_end_time STRING,
+    serving_id STRING,
+    is_permanent STRING,
+    apply_for_gpu_count INT,
+    pod_name STRING,
+    pkg_agg_time STRING,
+    gpu_util DOUBLE,
+    gpu_count INT,
+    p_date STRING
+) PARTITIONED BY (dt STRING) STORED AS ORC;
+
+-- Agent 候选目标表
+CREATE TABLE IF NOT EXISTS internal_platform_db.dwd_notebook_instance_pod_cross_day_detail_d_cand_query_engine_005 (
+    trace_id STRING,
+    datawd_project_id STRING,
+    datawd_task_id STRING,
+    datawd_task_instance_id STRING,
+    compute_type STRING,
+    status_code INT,
+    instance_run_time INT,
+    code_run_time INT,
+    resource_wait_time INT,
+    code_start_time STRING,
+    code_end_time STRING,
+    instance_start_time STRING,
+    instance_end_time STRING,
+    serving_id STRING,
+    is_permanent STRING,
+    apply_for_gpu_count INT,
+    pod_name STRING,
+    pkg_agg_time STRING,
+    gpu_util DOUBLE,
+    gpu_count INT,
+    p_date STRING
+) PARTITIONED BY (dt STRING) STORED AS ORC;

tasks/offline-compute/HiveSQL/hivesql_002/init/schema.sql

@@ -0,0 +1,274 @@
+CREATE TABLE IF NOT EXISTS internal_platform_db.notebook_span_info_copilot (
+    `databus_imp_date` STRING COMMENT 'primary partition',
+    `trace_id` STRING COMMENT 'trace_id',
+    `span_kind` STRING COMMENT 'span_kind',
+    `error_code` STRING COMMENT 'error_code',
+    `service_name` STRING COMMENT 'service_name',
+    `message` STRING COMMENT 'message',
+    `parent_span_id` STRING COMMENT 'parent_span_id',
+    `span_name` STRING COMMENT 'span_name',
+    `duration` BIGINT COMMENT 'duration',
+    `span_id` STRING COMMENT 'span_id',
+    `space_id` STRING COMMENT 'space_id',
+    `span_type` STRING COMMENT 'span_type',
+    `start_time` STRING COMMENT 'start_time',
+    `end_time` STRING COMMENT 'end_time',
+    `service_instance` STRING COMMENT 'service_instance',
+    `status_code` INT COMMENT 'status_code',
+    `datawd_project_id` STRING COMMENT 'datawd_project_id',
+    `datawd_task_id` STRING COMMENT 'datawd_task_id',
+    `datawd_task_instance_id` STRING COMMENT 'datawd_task_instance_id',
+    `compute_type` STRING COMMENT 'compute_type'
+) STORED AS ORC;
+
+CREATE TABLE IF NOT EXISTS internal_platform_db.dwd_gputj_service_instance_map_copilot (
+    `dt` STRING COMMENT '分区字段',
+    `service_name` STRING,
+    `instance_uuid` STRING,
+    `service_id` BIGINT,
+    `version_id` BIGINT,
+    `desc` STRING,
+    `create_time` DOUBLE,
+    `update_time` DOUBLE,
+    `workload_name` STRING
+) STORED AS ORC;
+
+CREATE TABLE IF NOT EXISTS internal_platform_db.dwd_ml_platform_instance_podname_copilot (
+    `dt` STRING COMMENT '分区字段',
+    `instance_uuid` STRING,
+    `pod_name` STRING,
+    `pod_phase` STRING,
+    `pod_create_time` BIGINT,
+    `pod_update_time` BIGINT,
+    `namespace` STRING,
+    `trial_job_name` STRING,
+    `create_time` BIGINT COMMENT '创建时间',
+    `modify_time` BIGINT COMMENT '修改时间',
+    `fix_modify_time` BIGINT COMMENT 'GPU调度平台TJ原始的修改时间较正常少了8小时，因此这里加了回来'
+) STORED AS ORC;
+
+CREATE TABLE IF NOT EXISTS internal_platform_db.gputj_gpu_info_parsed_agg_1min_copilot (
+    `dt` STRING,
+    `thedate` STRING,
+    `pod_name` STRING,
+    `node` STRING,
+    `pkg_agg_time` STRING,
+    `namespace` STRING,
+    `reason` STRING,
+    `instance` STRING,
+    `node_ip` STRING,
+    `source` STRING,
+    `pkg_time` STRING,
+    `mountpoint` STRING,
+    `node_role` STRING,
+    `zone` STRING,
+    `gpu_name` STRING,
+    `karmada` STRING,
+    `state` STRING,
+    `uuid` STRING,
+    `fstype` STRING,
+    `group` STRING,
+    `dedicated` STRING,
+    `resource` STRING,
+    `hostname` STRING,
+    `gpu` STRING,
+    `modelname` STRING,
+    `unit` STRING,
+    `exported_namespace` STRING,
+    `job` STRING,
+    `device` STRING,
+    `instance_type` STRING,
+    `status` STRING,
+    `k8s_dcgm_fi_dev_fb_util_max` DOUBLE,
+    `k8s_dcgm_fi_dev_fb_util_min` DOUBLE,
+    `k8s_dcgm_fi_dev_fb_util_sum` DOUBLE,
+    `k8s_dcgm_fi_dev_fb_util_count` BIGINT,
+    `k8s_pod_bs_fs_usage_bytes_usage_max` DOUBLE,
+    `k8s_pod_bs_fs_usage_bytes_usage_min` DOUBLE,
+    `k8s_pod_bs_fs_usage_bytes_usage_sum` DOUBLE,
+    `k8s_pod_bs_fs_usage_bytes_usage_count` BIGINT,
+    `k8s_pod_bs_fs_usage_bytes_max` DOUBLE,
+    `k8s_pod_bs_fs_usage_bytes_min` DOUBLE,
+    `k8s_pod_bs_fs_usage_bytes_sum` DOUBLE,
+    `k8s_pod_bs_fs_usage_bytes_count` BIGINT,
+    `k8s_container_bs_mem_working_set_bytes_max` DOUBLE,
+    `k8s_container_bs_mem_working_set_bytes_min` DOUBLE,
+    `k8s_container_bs_mem_working_set_bytes_sum` DOUBLE,
+    `k8s_container_bs_mem_working_set_bytes_count` BIGINT,
+    `k8s_container_bs_resource_request_mem_max` DOUBLE,
+    `k8s_container_bs_resource_request_mem_min` DOUBLE,
+    `k8s_container_bs_resource_request_mem_sum` DOUBLE,
+    `k8s_container_bs_resource_request_mem_count` BIGINT,
+    `k8s_container_bs_resource_request_cpu_max` DOUBLE,
+    `k8s_container_bs_resource_request_cpu_min` DOUBLE,
+    `k8s_container_bs_resource_request_cpu_sum` DOUBLE,
+    `k8s_container_bs_resource_request_cpu_count` BIGINT,
+    `k8s_pod_bs_restart_times_max` DOUBLE,
+    `k8s_pod_bs_restart_times_min` DOUBLE,
+    `k8s_pod_bs_restart_times_sum` DOUBLE,
+    `k8s_pod_bs_restart_times_count` BIGINT,
+    `k8s_container_resource_request_gpu_max` DOUBLE,
+    `k8s_container_resource_request_gpu_min` DOUBLE,
+    `k8s_container_resource_request_gpu_sum` DOUBLE,
+    `k8s_container_resource_request_gpu_count` BIGINT,
+    `k8s_container_gpu_used_max` DOUBLE,
+    `k8s_container_gpu_used_min` DOUBLE,
+    `k8s_container_gpu_used_sum` DOUBLE,
+    `k8s_container_gpu_used_count` BIGINT,
+    `k8s_container_gpu_mem_copy_util_max` DOUBLE,
+    `k8s_container_gpu_mem_copy_util_min` DOUBLE,
+    `k8s_container_gpu_mem_copy_util_sum` DOUBLE,
+    `k8s_container_gpu_mem_copy_util_count` BIGINT,
+    `dcgm_fi_prof_nvlink_tx_bytes_max` DOUBLE,
+    `dcgm_fi_prof_nvlink_tx_bytes_min` DOUBLE,
+    `dcgm_fi_prof_nvlink_tx_bytes_sum` DOUBLE,
+    `dcgm_fi_prof_nvlink_tx_bytes_count` BIGINT,
+    `dcgm_fi_prof_nvlink_rx_bytes_max` DOUBLE,
+    `dcgm_fi_prof_nvlink_rx_bytes_min` DOUBLE,
+    `dcgm_fi_prof_nvlink_rx_bytes_sum` DOUBLE,
+    `dcgm_fi_prof_nvlink_rx_bytes_count` BIGINT,
+    `dcgm_fi_prof_pcie_tx_bytes_max` DOUBLE,
+    `dcgm_fi_prof_pcie_tx_bytes_min` DOUBLE,
+    `dcgm_fi_prof_pcie_tx_bytes_sum` DOUBLE,
+    `dcgm_fi_prof_pcie_tx_bytes_count` BIGINT,
+    `dcgm_fi_prof_pcie_rx_bytes_max` DOUBLE,
+    `dcgm_fi_prof_pcie_rx_bytes_min` DOUBLE,
+    `dcgm_fi_prof_pcie_rx_bytes_sum` DOUBLE,
+    `dcgm_fi_prof_pcie_rx_bytes_count` BIGINT,
+    `k8s_namespace_pod_phase_duration_seconds_p95_1h_max` DOUBLE,
+    `k8s_namespace_pod_phase_duration_seconds_p95_1h_min` DOUBLE,
+    `k8s_namespace_pod_phase_duration_seconds_p95_1h_sum` DOUBLE,
+    `k8s_namespace_pod_phase_duration_seconds_p95_1h_count` BIGINT,
+    `k8s_namespace_pod_phase_duration_seconds_p99_1h_max` DOUBLE,
+    `k8s_namespace_pod_phase_duration_seconds_p99_1h_min` DOUBLE,
+    `k8s_namespace_pod_phase_duration_seconds_p99_1h_sum` DOUBLE,
+    `k8s_namespace_pod_phase_duration_seconds_p99_1h_count` BIGINT,
+    `k8s_container_bs_cpu_core_used_max` DOUBLE,
+    `k8s_container_bs_cpu_core_used_min` DOUBLE,
+    `k8s_container_bs_cpu_core_used_sum` DOUBLE,
+    `k8s_container_bs_cpu_core_used_count` BIGINT,
+    `k8s_container_bs_mem_no_cache_bytes_max` DOUBLE,
+    `k8s_container_bs_mem_no_cache_bytes_min` DOUBLE,
+    `k8s_container_bs_mem_no_cache_bytes_sum` DOUBLE,
+    `k8s_container_bs_mem_no_cache_bytes_count` BIGINT,
+    `k8s_container_bs_mem_usage_bytes_max` DOUBLE,
+    `k8s_container_bs_mem_usage_bytes_min` DOUBLE,
+    `k8s_container_bs_mem_usage_bytes_sum` DOUBLE,
+    `k8s_container_bs_mem_usage_bytes_count` BIGINT,
+    `k8s_container_bs_rate_cpu_core_used_request_max` DOUBLE,
+    `k8s_container_bs_rate_cpu_core_used_request_min` DOUBLE,
+    `k8s_container_bs_rate_cpu_core_used_request_sum` DOUBLE,
+    `k8s_container_bs_rate_cpu_core_used_request_count` BIGINT,
+    `k8s_container_rate_mem_working_set_request_max` DOUBLE,
+    `k8s_container_rate_mem_working_set_request_min` DOUBLE,
+    `k8s_container_rate_mem_working_set_request_sum` DOUBLE,
+    `k8s_container_rate_mem_working_set_request_count` BIGINT,
+    `k8s_container_rate_gpu_used_request_max` DOUBLE,
+    `k8s_container_rate_gpu_used_request_min` DOUBLE,
+    `k8s_container_rate_gpu_used_request_sum` DOUBLE,
+    `k8s_container_rate_gpu_used_request_count` BIGINT,
+    `k8s_workload_pod_status_ready_max` DOUBLE,
+    `k8s_workload_pod_status_ready_min` DOUBLE,
+    `k8s_workload_pod_status_ready_sum` DOUBLE,
+    `k8s_workload_pod_status_ready_count` BIGINT,
+    `k8s_workload_pod_num_max` DOUBLE,
+    `k8s_workload_pod_num_min` DOUBLE,
+    `k8s_workload_pod_num_sum` DOUBLE,
+    `k8s_workload_pod_num_count` BIGINT,
+    `k8s_pod_status_ready_max` DOUBLE,
+    `k8s_pod_status_ready_min` DOUBLE,
+    `k8s_pod_status_ready_sum` DOUBLE,
+    `k8s_pod_status_ready_count` BIGINT,
+    `k8s_workload_pod_status_ready_rate_max` DOUBLE,
+    `k8s_workload_pod_status_ready_rate_min` DOUBLE,
+    `k8s_workload_pod_status_ready_rate_sum` DOUBLE,
+    `k8s_workload_pod_status_ready_rate_count` BIGINT,
+    `k8s_star_udf_label_pod_ready_rate_max` DOUBLE,
+    `k8s_star_udf_label_pod_ready_rate_min` DOUBLE,
+    `k8s_star_udf_label_pod_ready_rate_sum` DOUBLE,
+    `k8s_star_udf_label_pod_ready_rate_count` BIGINT,
+    `k8s_container_rate_mem_request_max` DOUBLE,
+    `k8s_container_rate_mem_request_min` DOUBLE,
+    `k8s_container_rate_mem_request_sum` DOUBLE,
+    `k8s_container_rate_mem_request_count` BIGINT,
+    `k8s_container_vgpu_gpu_util_max` DOUBLE,
+    `k8s_container_vgpu_gpu_util_min` DOUBLE,
+    `k8s_container_vgpu_gpu_util_sum` DOUBLE,
+    `k8s_container_vgpu_gpu_util_count` BIGINT,
+    `k8s_dcgm_fi_dev_fb_vgpu_gpu_util_max` DOUBLE,
+    `k8s_dcgm_fi_dev_fb_vgpu_gpu_util_min` DOUBLE,
+    `k8s_dcgm_fi_dev_fb_vgpu_gpu_util_sum` DOUBLE,
+    `k8s_dcgm_fi_dev_fb_vgpu_gpu_util_count` BIGINT,
+    `k8s_dcgm_fi_prof_sm_active_max` DOUBLE,
+    `k8s_dcgm_fi_prof_sm_active_min` DOUBLE,
+    `k8s_dcgm_fi_prof_sm_active_sum` DOUBLE,
+    `k8s_dcgm_fi_prof_sm_active_count` BIGINT,
+    `k8s_pod_bs_status_abnormal_15m_max` DOUBLE,
+    `k8s_pod_bs_status_abnormal_15m_min` DOUBLE,
+    `k8s_pod_bs_status_abnormal_15m_sum` DOUBLE,
+    `k8s_pod_bs_status_abnormal_15m_count` BIGINT,
+    `k8s_pod_bs_status_abnormal_5m_max` DOUBLE,
+    `k8s_pod_bs_status_abnormal_5m_min` DOUBLE,
+    `k8s_pod_bs_status_abnormal_5m_sum` DOUBLE,
+    `k8s_pod_bs_status_abnormal_5m_count` BIGINT,
+    `k8s_pod_bs_status_abnormal_20m_max` DOUBLE,
+    `k8s_pod_bs_status_abnormal_20m_min` DOUBLE,
+    `k8s_pod_bs_status_abnormal_20m_sum` DOUBLE,
+    `k8s_pod_bs_status_abnormal_20m_count` BIGINT,
+    `k8s_pod_bs_status_abnormal_30m_max` DOUBLE,
+    `k8s_pod_bs_status_abnormal_30m_min` DOUBLE,
+    `k8s_pod_bs_status_abnormal_30m_sum` DOUBLE,
+    `k8s_pod_bs_status_abnormal_30m_count` BIGINT,
+    `k8s_pod_bs_status_abnormal_10m_max` DOUBLE,
+    `k8s_pod_bs_status_abnormal_10m_min` DOUBLE,
+    `k8s_pod_bs_status_abnormal_10m_sum` DOUBLE,
+    `k8s_pod_bs_status_abnormal_10m_count` BIGINT,
+    `dcgm_fi_dev_row_remap_failure_max` DOUBLE,
+    `dcgm_fi_dev_row_remap_failure_min` DOUBLE,
+    `dcgm_fi_dev_row_remap_failure_sum` DOUBLE,
+    `dcgm_fi_dev_row_remap_pending_max` DOUBLE,
+    `dcgm_fi_dev_row_remap_pending_min` DOUBLE,
+    `dcgm_fi_dev_row_remap_pending_sum` DOUBLE,
+    `dcgm_fi_dev_row_remap_pending_count` BIGINT,
+    `dcgm_fi_dev_xid_errors_max` DOUBLE,
+    `dcgm_fi_dev_xid_errors_min` DOUBLE,
+    `dcgm_fi_dev_xid_errors_sum` DOUBLE,
+    `dcgm_fi_dev_xid_errors_count` BIGINT,
+    `dcgm_fi_dev_uncorrectable_remapped_rows_max` DOUBLE,
+    `dcgm_fi_dev_uncorrectable_remapped_rows_min` DOUBLE,
+    `dcgm_fi_dev_uncorrectable_remapped_rows_sum` DOUBLE,
+    `dcgm_fi_dev_uncorrectable_remapped_rows_count` BIGINT,
+    `starrocks_timestamp` BIGINT,
+    `priority_class` STRING,
+    `k8s_container_vgpu_gpu_mem_usage_max` DOUBLE,
+    `k8s_container_vgpu_gpu_mem_usage_min` DOUBLE,
+    `k8s_container_vgpu_gpu_mem_usage_sum` DOUBLE,
+    `k8s_container_vgpu_gpu_mem_usage_count` BIGINT,
+    `k8s_container_vgpu_gpu_mem_total_max` DOUBLE,
+    `k8s_container_vgpu_gpu_mem_total_min` DOUBLE,
+    `k8s_container_vgpu_gpu_mem_total_sum` DOUBLE,
+    `k8s_container_vgpu_gpu_mem_total_count` BIGINT
+) STORED AS ORC;
+
+CREATE TABLE IF NOT EXISTS internal_platform_db.notebook_engine_info_copilot (
+    `databus_imp_date` STRING COMMENT 'primary partition',
+    `trace_id` STRING COMMENT 'trace_id',
+    `service_name` STRING COMMENT 'service_name',
+    `compute_type` STRING COMMENT 'compute_type',
+    `datawd_project_id` STRING COMMENT 'datawd_project_id',
+    `datawd_task_id` STRING COMMENT 'datawd_task_id',
+    `datawd_task_instance_id` STRING COMMENT 'datawd_task_instance_id',
+    `is_permanent` STRING COMMENT 'is_permanent',
+    `serving_id` STRING COMMENT 'serving_id',
+    `user_name` STRING COMMENT 'user_name',
+    `is_model_service` STRING COMMENT 'is_model_service',
+    `compute_info` STRING COMMENT 'compute_info',
+    `num_gpu` STRING COMMENT 'num_gpu',
+    `replicas` STRING COMMENT 'replicas',
+    `gpu_type` STRING COMMENT 'gpu_type',
+    `timestamp` STRING COMMENT 'timestamp',
+    `ray_job_id` STRING,
+    `reason` STRING,
+    `log_url` STRING,
+    `dashboard_url` STRING
+) STORED AS ORC;

tasks/offline-compute/HiveSQL/hivesql_002/task.md

@@ -0,0 +1,42 @@
+---
+id: offline-compute_HiveSQL_hivesql_002
+name: 统计数据平台WDNotebook Ray类型管道任务中运行时间跨自然天的实例明细。从`wedat
+category: offline-compute/HiveSQL
+timeout_seconds: 600
+modality: pure-text
+engine: hivesql
+---
+## Prompt
+**任务目标**：统计数据平台WDNotebook Ray类型管道任务中运行时间跨自然天的实例明细，按自然天拆分并关联GPU指标。
+
+**输入**：
+- `internal_platform_db.notebook_span_info_query_engine_005`
+- `internal_platform_db.dwd_gputj_service_instance_map_query_engine_005`
+- `internal_platform_db.dwd_ml_platform_instance_podname_query_engine_005`
+- `internal_platform_db.gputj_gpu_info_parsed_agg_1min_query_engine_005`
+- `internal_platform_db.notebook_engine_info_query_engine_005`
+
+**处理规则**：
+1. 时间过滤：`20260504`-`20260507`，从`notebook_span_info`筛选在`20260507`有记录且跨天的`trace_id`。
+2. 业务过滤：计算类型='ray'，服务名='notebook-runner'；span名称限定：'runner.execute'、'execute.code'、'execute.code.cell'、'client.execute.code'、'runner.killed'、'set.permanent.compute'、'create.non.permanent.compute'；GPU表过滤无效记录。
+3. 跨天拆分：将跨天实例按自然天拆分，每条记录对应一天内的开始和结束时间。
+4. 派生字段：
+   - `instance_run_time`：(end_time - start_time)/1000（秒），基于'runner.execute'
+   - `code_run_time`：代码执行相关span总时长（秒）
+   - `resource_wait_time`：资源创建相关span总时长（秒）
+   - `apply_for_gpu_count`：SUM(replicas * num_gpu)，按trace_id
+   - `gpu_util`：k8s_container_vgpu_gpu_util_sum / k8s_container_vgpu_gpu_util_count
+   - `gpu_count`：k8s_container_resource_request_gpu_sum / k8s_container_resource_request_gpu_count
+5. 表关联：
+   - 拆分结果按`trace_id`左关联`notebook_engine_info`获取`serving_id`
+   - 通过`instance_uuid`关联`dwd_gputj_service_instance_map`与`dwd_ml_platform_instance_podname`，再按`serving_id=service_id`左关联得到`pod_name`
+   - 按`pod_name`左关联`gputj_gpu_info_parsed_agg_1min`，条件为`pkg_agg_time`落在实例运行窗口内（优先`instance_start_time`到`instance_end_time`，否则`code_start_time`到`code_end_time`）
+
+**输出要求**：
+- 输出表：`internal_platform_db.dwd_notebook_instance_pod_cross_day_detail_d_cand_query_engine_005`
+- 字段顺序：dt:STRING; trace_id:STRING; datawd_project_id:STRING; datawd_task_id:STRING; datawd_task_instance_id:STRING; compute_type:STRING; status_code:INT; instance_run_time:INT; code_run_time:INT; resource_wait_time:INT; code_start_time:STRING; code_end_time:STRING; instance_start_time:STRING; instance_end_time:STRING; serving_id:STRING; is_permanent:BOOLEAN; apply_for_gpu_count:INT; pod_name:STRING; pkg_agg_time:STRING; gpu_util:DOUBLE; gpu_count:INT; p_date:STRING
+- 分区字段：`dt`
+
+**写入要求**：写入`dt='20260507'`分区。
+
+请将最终 HiveSQL 写入 result.sql 并执行。

tasks/offline-compute/HiveSQL/hivesql_003/gt/expected.csv

@@ -0,0 +1,3 @@
+p_date,trace_id,datawd_project_id,datawd_task_id,datawd_task_instance_id,compute_type,status_code,instance_run_time,code_run_time,resource_wait_time,code_start_time,code_end_time,instance_start_time,instance_end_time,serving_id,is_permanent,apply_for_gpu_count,dt
+2025-05-07,trace_001,proj_01,task_01,inst_01,ray,2,15,13,2,2025-05-07 03:59:52,2025-05-07 04:00:05,2025-05-07 03:59:50,2025-05-07 04:00:05,srv_001,True,2,20260507
+2025-05-07,trace_002,proj_02,task_02,inst_02,ray,2,18,16,3,2025-05-07 04:01:32,2025-05-07 04:01:48,2025-05-07 04:01:30,2025-05-07 04:01:48,srv_002,False,8,20260507

tasks/offline-compute/HiveSQL/hivesql_003/gt/grade.py

@@ -0,0 +1,753 @@
+"""hivesql_007 精细评分脚本
+
+业务场景：Killed 实例按天拆分运行时长统计（2表 + CTE + UNION + 跨天拆分）
+难度: HARD | 特征: CTE|UNION_ALL|CROSS_DAY_SPLIT|TIME_CALCULATION|LEFT_JOIN
+
+评分范式: 产物100分 = A(10) + B(10) + C(10) + D(55) + F(15)
+  A_executability       (10分): result.sql 能跑通且产出非空
+  B_schema              (10分): 18列 + 列名匹配
+  C_row_alignment     (10分): 行数比例 + key覆盖率
+  D_time_calculation    (30分): instance_run_time, code_run_time 等时间指标
+  D_cross_day_split     (15分): 跨天 trace 按天拆分正确性
+  D_engine_join         (10分): serving_id, is_permanent 等 engine 维度列匹配
+  F_join_completeness   (5分):  2张源表都被关联
+  F_insert_overwrite    (5分):  写入模式+分区
+  F_partition_value     (5分):  dt=20260507
+
+权重: HARD → product=0.7, process=0.3
+"""
+import os
+import re
+import subprocess
+import tempfile
+import json
+import math
+
+
+def grade(workspace_path, **kwargs):
+
+    # ========== Case 配置 ==========
+    OUTPUT_TABLE = "internal_platform_db.dwd_notebook_killed_instance_detail_d_copilot_cand_query_engine_007"
+    DIFFICULTY = "HARD"
+    SOURCE_TABLES = [
+        "notebook_span_info_query_engine_007",
+        "notebook_engine_info_query_engine_007",
+        "dwd_notebook_killed_instance_detail_d_copilot_query_engine_007",
+    ]
+    KEY_COLUMNS = ["trace_id", "p_date"]
+    TIME_COLUMNS = ["instance_run_time", "code_run_time", "resource_wait_time"]
+    ENGINE_DIM_COLUMNS = ["serving_id", "is_permanent", "apply_for_gpu_count"]
+    SOURCE_TABLE_SHORT_NAMES = [
+        "notebook_span_info",
+        "notebook_engine_info",
+    ]
+    GT_TABLE = OUTPUT_TABLE.replace("_cand_", "_")
+
+    DIFFICULTY_WEIGHTS = {
+        "EASY": (0.5, 0.5),
+        "MEDIUM": (0.6, 0.4),
+        "HARD": (0.7, 0.3),
+        "EXPERT": (0.8, 0.2),
+    }
+
+    _SPARK_SUBMIT_TIMEOUT = 300
+    _JSON_START = "__GRADE_JSON_START__"
+    _JSON_END = "__GRADE_JSON_END__"
+
+    result = {
+        "overall_score": 0.0,
+        "total_points": 0,
+        "grade": "",
+        "details": {},
+        "diagnostics": [],
+    }
+
+    # ========== 内部辅助函数 ==========
+
+    def values_match(pred_val, gt_val, abs_tol=1e-6, rel_tol=1e-4):
+        if pred_val is None and gt_val is None:
+            return True
+        if pred_val is None or gt_val is None:
+            return False
+        s_pred = str(pred_val).strip()
+        s_gt = str(gt_val).strip()
+        if s_pred == s_gt:
+            return True
+        try:
+            pv = float(s_pred)
+            gv = float(s_gt)
+            if math.isnan(pv) and math.isnan(gv):
+                return True
+            if math.isnan(pv) or math.isnan(gv):
+                return False
+            if abs(gv) < abs_tol:
+                return abs(pv - gv) <= abs_tol
+            return abs(pv - gv) <= abs_tol or abs(pv - gv) / max(abs(gv), 1e-12) <= rel_tol
+        except (ValueError, TypeError):
+            pass
+        return s_pred.lower() == s_gt.lower()
+
+    def _run_spark_script(script_code, timeout=_SPARK_SUBMIT_TIMEOUT):
+        with tempfile.NamedTemporaryFile(mode="w", suffix=".py", delete=False, encoding="utf-8") as f:
+            f.write(script_code)
+            script_path = f.name
+        try:
+            r = subprocess.run(
+                ["spark-submit", script_path],
+                capture_output=True, text=True, timeout=timeout,
+            )
+            stdout = r.stdout or ""
+            if _JSON_START in stdout and _JSON_END in stdout:
+                json_str = stdout.split(_JSON_START)[1].split(_JSON_END)[0].strip()
+                return json.loads(json_str), None
+            else:
+                if r.returncode == 0:
+                    for line in stdout.splitlines():
+                        if line.strip().startswith("Traceback"):
+                            return None, f"spark-submit error: {line}"
+                    return None, "spark-submit 无 JSON 输出"
+                err_msg = (r.stderr or "")[-500:]
+                return None, f"spark-submit failed: {err_msg}"
+        except subprocess.TimeoutExpired:
+            return None, f"spark-submit 超时 ({timeout}s)"
+        except Exception as e:
+            return None, f"spark-submit 异常: {e}"
+        finally:
+            try:
+                os.unlink(script_path)
+            except OSError:
+                pass
+
+    def read_table_via_spark_submit(table_name):
+        """Read table via spark-submit subprocess. Returns (cols, rows_as_lists)."""
+        if not re.match(r'^[a-zA-Z_][a-zA-Z0-9_.]*$', table_name):
+            raise ValueError(f"非法表名: {table_name}")
+        read_script = f'''
+import json
+from pyspark.sql import SparkSession
+spark = SparkSession.builder.appName("grade_read").enableHiveSupport() \\
+    .config("spark.sql.warehouse.dir", "/tmp/hive_warehouse").getOrCreate()
+try:
+    df = spark.sql("SELECT * FROM {table_name}")
+    cols = [c.lower() for c in df.columns]
+    rows = [[str(v) if v is not None else "" for v in row] for row in df.collect()]
+    print("{_JSON_START}")
+    print(json.dumps({{"cols": cols, "rows": rows}}, ensure_ascii=False))
+    print("{_JSON_END}")
+except Exception as e:
+    print("{_JSON_START}")
+    print(json.dumps({{"error": str(e)}}))
+    print("{_JSON_END}")
+finally:
+    spark.stop()
+'''
+        data, err = _run_spark_script(read_script)
+        if err:
+            raise RuntimeError(f"read_table failed: {err}")
+        if "error" in data:
+            raise RuntimeError(f"query failed: {data['error']}")
+        return data["cols"], data["rows"]
+
+    # SQL executor template (self-contained, no external dependency)
+    _SQL_EXEC_TEMPLATE = '''
+import json, re
+from pyspark.sql import SparkSession
+spark = SparkSession.builder.appName("{app_name}").enableHiveSupport() \\
+    .config("spark.sql.warehouse.dir", "/tmp/hive_warehouse").getOrCreate()
+try:
+    with open("{sql_file}", "r", encoding="utf-8") as _f:
+        _sql = _f.read()
+    _sql = re.sub(r"^\\s*set\\s+query_engine\\.\\S+\\n?", "", _sql, flags=re.IGNORECASE)
+    _stmts, _cur, _in_sq, _in_dq, _i = [], [], False, False, 0
+    while _i < len(_sql):
+        _ch = _sql[_i]
+        if _ch == "\\\\" and _i + 1 < len(_sql):
+            _cur.append(_ch); _cur.append(_sql[_i+1]); _i += 2; continue
+        if _ch == "-" and _i+1 < len(_sql) and _sql[_i+1] == "-" and not _in_sq and not _in_dq:
+            while _i < len(_sql) and _sql[_i] != "\\n": _i += 1
+            _cur.append("\\n"); continue
+        if _ch == "'" and not _in_dq: _in_sq = not _in_sq
+        elif _ch == '"' and not _in_sq: _in_dq = not _in_dq
+        if _ch == ";" and not _in_sq and not _in_dq:
+            _s = "".join(_cur).strip()
+            if _s: _stmts.append(_s)
+            _cur = []
+        else:
+            _cur.append(_ch)
+        _i += 1
+    _last = "".join(_cur).strip()
+    if _last: _stmts.append(_last)
+    for _stmt in _stmts:
+        spark.sql(_stmt)
+    print("{_JSON_START}")
+    print(json.dumps({{"ok": True}}))
+    print("{_JSON_END}")
+except Exception as e:
+    print("{_JSON_START}")
+    print(json.dumps({{"ok": False, "error": str(e)}}))
+    print("{_JSON_END}")
+finally:
+    spark.stop()
+'''
+
+    def execute_result_sql():
+        result_sql = os.path.join(workspace_path, "result.sql")
+        if not os.path.exists(result_sql):
+            return False, "no_result_file"
+        # 替换 数据平台WD时间变量（沙箱 spark-sql 不支持 ${...} 语法）
+        with open(result_sql, 'r', encoding='utf-8') as _rf:
+            _sql_text = _rf.read()
+        _bizdate = '20260507'
+        _sql_text = re.sub(r'\${bdp\.system\.bizdate(?:[+-]\d+)?}', _bizdate, _sql_text)
+        _sql_text = re.sub(r'\${yyyymmdd(?:[+-]\d+)?}', _bizdate, _sql_text)
+        _sql_text = re.sub(r'\${bizdate(?:[+-]\d+)?}', _bizdate, _sql_text)
+        _sql_text = re.sub(r'\${[^}]*date[^}]*}', _bizdate, _sql_text)
+        with open(result_sql, 'w', encoding='utf-8') as _wf:
+            _wf.write(_sql_text)
+        script = _SQL_EXEC_TEMPLATE.format(app_name="grade_exec", sql_file=result_sql,
+                                            _JSON_START=_JSON_START, _JSON_END=_JSON_END)
+        data, err = _run_spark_script(script)
+        if err:
+            return False, f"execution_error: {err}"
+        if data and data.get("ok"):
+            return True, None
+        return False, f"execution_error: {data.get('error', 'unknown') if data else 'no output'}"
+
+    def execute_ground_truth_sql():
+        gt_sql = os.path.join(workspace_path, "gt", "ground_truth.sql")
+        if not os.path.exists(gt_sql):
+            return False, "ground_truth.sql not found"
+        script = _SQL_EXEC_TEMPLATE.format(app_name="grade_gt", sql_file=gt_sql,
+                                            _JSON_START=_JSON_START, _JSON_END=_JSON_END)
+        data, err = _run_spark_script(script)
+        if err:
+            return False, f"gt_execution_error: {err}"
+        if data and data.get("ok"):
+            return True, None
+        return False, f"gt_execution_error: {data.get('error', 'unknown') if data else 'no output'}"
+
+    def truncate_table(table_name):
+        script = f'''
+from pyspark.sql import SparkSession
+spark = SparkSession.builder.appName("truncate").enableHiveSupport() \\
+    .config("spark.sql.warehouse.dir", "/tmp/hive_warehouse").getOrCreate()
+spark.sql("TRUNCATE TABLE {table_name}")
+spark.stop()
+'''
+        try:
+            _run_spark_script(script, timeout=120)
+        except Exception:
+            pass
+
+    def restore_hive_site():
+        """Restore hive-site.xml to canonical state (agent may have modified it)."""
+        canonical_hive_site = '''<?xml version="1.0"?>
+<?xml-stylesheet type="text/xsl" href="configuration.xsl"?>
+<configuration>
+  <property>
+    <name>hive.metastore.uris</name>
+    <value>thrift://localhost:9083</value>
+  </property>
+  <property>
+    <name>hive.metastore.warehouse.dir</name>
+    <value>/tmp/hive_warehouse</value>
+  </property>
+  <property>
+    <name>javax.jdo.option.ConnectionURL</name>
+    <value>jdbc:derby:;databaseName=/tmp/hive_metastore_db;create=true</value>
+  </property>
+  <property>
+    <name>javax.jdo.option.ConnectionDriverName</name>
+    <value>org.apache.derby.jdbc.EmbeddedDriver</value>
+  </property>
+  <property>
+    <name>datanucleus.schema.autoCreateAll</name>
+    <value>true</value>
+  </property>
+  <property>
+    <name>hive.metastore.schema.verification</name>
+    <value>false</value>
+  </property>
+</configuration>
+'''
+        hive_site_path = os.path.join(os.environ.get('SPARK_HOME', '/opt/spark'), 'conf', 'hive-site.xml')
+        try:
+            with open(hive_site_path, 'w') as f:
+                f.write(canonical_hive_site)
+        except Exception:
+            pass
+
+    def finalize(result):
+        product_weight, process_weight = DIFFICULTY_WEIGHTS.get(DIFFICULTY, (0.7, 0.3))
+        product_dims = ["A_executability"] + ['B_schema', 'C_row_alignment', 'D_cross_day_split', 'D_engine_join', 'D_time_calculation', 'F_insert_overwrite', 'F_join_completeness', 'F_partition_value']
+        product_raw = sum(result["details"].get(d, {}).get("score", 0) for d in product_dims)
+        product_ratio = product_raw / 100.0
+        for dim in ["H_efficiency"]:
+            if dim in result["details"]:
+                raw = result["details"][dim].get("score", 0)
+                result["details"][dim]["score_before_scaling"] = raw
+                result["details"][dim]["score"] = round(raw * product_ratio, 2)
+                result["details"][dim]["product_ratio"] = round(product_ratio, 4)
+        process_dims = ["G_exploration", "H_efficiency", "I_self_verification"]
+        process_raw = sum(result["details"].get(d, {}).get("score", 0) for d in process_dims)
+        product_score = round(product_raw * product_weight, 2)
+        process_score = round(process_raw * process_weight, 2)
+        total = round(product_score + process_score, 2)
+        result["total_points"] = total
+        result["product_points"] = product_score
+        result["process_points"] = round(process_score, 2)
+        result["weights"] = {"product": product_weight, "process": process_weight}
+        result["overall_score"] = round(total / 100.0, 4)
+        if total >= 90:
+            result["grade"] = "优秀"
+        elif total >= 75:
+            result["grade"] = "良好"
+        elif total >= 60:
+            result["grade"] = "合格"
+        elif total >= 40:
+            result["grade"] = "偏弱"
+        else:
+            result["grade"] = "不合格"
+        return result
+
+    # Restore hive-site.xml (agent may have modified it)
+    restore_hive_site()
+
+    # ========== A. 可执行性 (10分) ==========
+    a_items = {"A1_exec_ok": 0, "A2_has_data": 0}
+    exec_ok, exec_err = execute_result_sql()
+    if not exec_ok:
+        detail = "未产出 result.sql" if exec_err == "no_result_file" else str(exec_err)[:200]
+        result["details"]["A_executability"] = {"score": 0, "max": 10, "detail": detail}
+        result["error"] = detail
+        for dim in ['B_schema', 'C_row_alignment', 'D_cross_day_split', 'D_engine_join', 'D_time_calculation', 'F_insert_overwrite', 'F_join_completeness', 'F_partition_value']:
+            result["details"][dim] = {"score": 0, "max": 0, "items": {}}
+        return finalize(result)
+
+    a_items["A1_exec_ok"] = 5
+
+    pred_headers, pred_rows = [], []
+    try:
+        pred_headers, pred_rows = read_table_via_spark_submit(OUTPUT_TABLE)
+    except Exception as e:
+        result["diagnostics"].append(f"read_pred_failed: {e}")
+
+    if not pred_rows:
+        result["details"]["A_executability"] = {"score": 5, "max": 10, "items": a_items}
+        for dim in ['B_schema', 'C_row_alignment', 'D_cross_day_split', 'D_engine_join', 'D_time_calculation', 'F_insert_overwrite', 'F_join_completeness', 'F_partition_value']:
+            result["details"][dim] = {"score": 0, "max": 0, "items": {}}
+        return finalize(result)
+
+    a_items["A2_has_data"] = 5
+    result["details"]["A_executability"] = {"score": 10, "max": 10, "items": a_items}
+
+    # ========== Execute GT + Read GT ==========
+    truncate_table(GT_TABLE)
+    gt_ok, gt_err = execute_ground_truth_sql()
+    if not gt_ok:
+        result["error"] = f"ground_truth failed: {gt_err}"
+        for dim in ['B_schema', 'C_row_alignment', 'D_cross_day_split', 'D_engine_join', 'D_time_calculation', 'F_insert_overwrite', 'F_join_completeness', 'F_partition_value']:
+            result["details"][dim] = {"score": 0, "max": 0, "items": {}}
+        return finalize(result)
+
+    gt_headers, gt_rows = [], []
+    try:
+        gt_headers, gt_rows = read_table_via_spark_submit(GT_TABLE)
+    except Exception as e:
+        result["error"] = f"read_gt_failed: {e}"
+        for dim in ['B_schema', 'C_row_alignment', 'D_cross_day_split', 'D_engine_join', 'D_time_calculation', 'F_insert_overwrite', 'F_join_completeness', 'F_partition_value']:
+            result["details"][dim] = {"score": 0, "max": 0, "items": {}}
+        return finalize(result)
+
+    # ========== B/C/D/F 维度评分 ==========
+    pred_col_map = {h: i for i, h in enumerate(pred_headers)}
+    gt_col_map = {h: i for i, h in enumerate(gt_headers)}
+
+    # ========== B. Schema正确性 (10分) ==========
+    b_items = {}
+    expected_col_count = 18
+    if len(pred_headers) == expected_col_count:
+        b_items["B1_col_count"] = 5
+    elif abs(len(pred_headers) - expected_col_count) <= 2:
+        b_items["B1_col_count"] = 3
+    else:
+        b_items["B1_col_count"] = 0
+
+    gt_col_set = set(gt_headers)
+    pred_col_set = set(pred_headers)
+    name_match_rate = len(gt_col_set & pred_col_set) / max(len(gt_col_set), 1)
+    if name_match_rate >= 0.95:
+        b_items["B2_col_names"] = 5
+    elif name_match_rate >= 0.8:
+        b_items["B2_col_names"] = 3
+    else:
+        b_items["B2_col_names"] = 0
+
+    b_score = sum(b_items.values())
+    result["details"]["B_schema"] = {
+        "score": b_score, "max": 10,
+        "detail": {"col_count": len(pred_headers), "name_match_rate": round(name_match_rate, 4), "items": b_items},
+    }
+
+    # ========== C. 行一致性 (10分) ==========
+    c_items = {}
+    gt_row_count = len(gt_rows)
+    pred_row_count = len(pred_rows)
+
+    if gt_row_count > 0:
+        ratio = pred_row_count / gt_row_count
+        if 0.95 <= ratio <= 1.05:
+            c_items["C1_row_ratio"] = 5
+        elif 0.7 <= ratio <= 1.3:
+            c_items["C1_row_ratio"] = 3
+        else:
+            c_items["C1_row_ratio"] = 0
+    else:
+        c_items["C1_row_ratio"] = 5 if pred_row_count == 0 else 0
+
+    key_cols_avail = [k for k in KEY_COLUMNS if k in gt_col_map and k in pred_col_map]
+    if key_cols_avail and gt_row_count > 0:
+        gt_keys = set()
+        for row in gt_rows:
+            key = tuple(row[gt_col_map[k]] if gt_col_map[k] < len(row) else "" for k in key_cols_avail)
+            gt_keys.add(key)
+        pred_keys = set()
+        for row in pred_rows:
+            key = tuple(row[pred_col_map[k]] if pred_col_map[k] < len(row) else "" for k in key_cols_avail)
+            pred_keys.add(key)
+        coverage = len(gt_keys & pred_keys) / max(len(gt_keys), 1)
+        if coverage >= 0.995:
+            c_items["C2_key_coverage"] = 5
+        elif coverage >= 0.9:
+            c_items["C2_key_coverage"] = 3
+        elif coverage >= 0.7:
+            c_items["C2_key_coverage"] = 2
+        else:
+            c_items["C2_key_coverage"] = round(5 * coverage, 2)
+    else:
+        c_items["C2_key_coverage"] = 0
+
+    c_score = sum(v for v in c_items.values())
+    result["details"]["C_row_alignment"] = {"score": c_score, "max": 10, "detail": c_items}
+
+    # 构建索引
+    pred_index = {}
+    if key_cols_avail:
+        for row in pred_rows:
+            key = tuple(row[pred_col_map[k]] if pred_col_map[k] < len(row) else "" for k in key_cols_avail)
+            pred_index[key] = row
+
+    gt_index = {}
+    for row in gt_rows:
+        key = tuple(row[gt_col_map[k]] if gt_col_map[k] < len(row) else "" for k in key_cols_avail)
+        gt_index[key] = row
+
+    # ========== D. 时间计算正确性 (30分) ==========
+    d_time_items = {}
+    per_col_weight = 30.0 / max(len(TIME_COLUMNS), 1)
+    for col in TIME_COLUMNS:
+        gt_ci = gt_col_map.get(col)
+        pred_ci = pred_col_map.get(col)
+        if gt_ci is None or pred_ci is None:
+            d_time_items[col] = {"pass_rate": 0.0, "score": 0, "reason": "column_missing"}
+            continue
+
+        matches = 0
+        total = 0
+        for key, gt_row in gt_index.items():
+            pred_row = pred_index.get(key)
+            if pred_row is None:
+                total += 1
+                continue
+            gt_val = gt_row[gt_ci].strip() if gt_ci < len(gt_row) else ""
+            pred_val = pred_row[pred_ci].strip() if pred_ci < len(pred_row) else ""
+            if values_match(pred_val, gt_val, abs_tol=1.0, rel_tol=0.01):
+                matches += 1
+            total += 1
+
+        rate = matches / max(total, 1)
+        col_score = rate * per_col_weight
+        d_time_items[col] = {"pass_rate": round(rate, 4), "score": round(col_score, 2)}
+
+    d_time_score = sum(v.get("score", 0) for v in d_time_items.values() if isinstance(v, dict))
+    result["details"]["D_time_calculation"] = {"score": round(d_time_score, 2), "max": 30, "detail": d_time_items}
+
+    # ========== D. 跨天切分正确性 (15分) ==========
+    d_cross_items = {}
+    d_cross_score = 0
+    trace_idx = pred_col_map.get("trace_id")
+    pdate_idx = pred_col_map.get("p_date")
+
+    if trace_idx is not None and pdate_idx is not None:
+        pred_trace_dates = {}
+        for row in pred_rows:
+            tid = row[trace_idx].strip() if trace_idx < len(row) else ""
+            pd = row[pdate_idx].strip() if pdate_idx < len(row) else ""
+            if tid not in pred_trace_dates:
+                pred_trace_dates[tid] = set()
+            pred_trace_dates[tid].add(pd)
+
+        # 检查是否有 trace 跨天（跨天 trace 应出现多个 p_date）
+        gt_trace_idx = gt_col_map.get("trace_id")
+        gt_pdate_idx = gt_col_map.get("p_date")
+        gt_trace_dates = {}
+        if gt_trace_idx is not None and gt_pdate_idx is not None:
+            for row in gt_rows:
+                tid = row[gt_trace_idx].strip() if gt_trace_idx < len(row) else ""
+                pd = row[gt_pdate_idx].strip() if gt_pdate_idx < len(row) else ""
+                if tid not in gt_trace_dates:
+                    gt_trace_dates[tid] = set()
+                gt_trace_dates[tid].add(pd)
+
+        # 跨天 trace 拆分正确 (10分)
+        cross_traces = {tid: dates for tid, dates in gt_trace_dates.items() if len(dates) > 1}
+        if cross_traces:
+            correct = 0
+            for tid, expected_dates in cross_traces.items():
+                pred_dates = pred_trace_dates.get(tid, set())
+                if pred_dates >= expected_dates:
+                    correct += 1
+                elif len(pred_dates & expected_dates) > 0:
+                    correct += 0.5
+            cross_rate = correct / len(cross_traces)
+            d_cross_score += round(10 * cross_rate, 2)
+            d_cross_items["cross_day_traces"] = {"rate": round(cross_rate, 4), "count": len(cross_traces)}
+        else:
+            # 没有跨天 trace，给满分
+            d_cross_score += 10
+            d_cross_items["cross_day_traces"] = "no cross-day traces in GT"
+
+        # 行数正确 (5分)
+        if pred_row_count == gt_row_count:
+            d_cross_score += 5
+            d_cross_items["row_count_correct"] = True
+        elif abs(pred_row_count - gt_row_count) <= 1:
+            d_cross_score += 3
+            d_cross_items["row_count_correct"] = f"off by {abs(pred_row_count - gt_row_count)}"
+        else:
+            d_cross_items["row_count_correct"] = f"pred={pred_row_count}, gt={gt_row_count}"
+    else:
+        d_cross_items["error"] = "trace_id or p_date column missing"
+
+    result["details"]["D_cross_day_split"] = {"score": round(d_cross_score, 2), "max": 15, "detail": d_cross_items}
+
+    # ========== D. Engine 维度列匹配 (10分) ==========
+    d_engine_items = {}
+    for col in ENGINE_DIM_COLUMNS:
+        gt_ci = gt_col_map.get(col)
+        pred_ci = pred_col_map.get(col)
+        if gt_ci is None or pred_ci is None:
+            continue
+        matches = 0
+        total = 0
+        for key, gt_row in gt_index.items():
+            pred_row = pred_index.get(key)
+            if pred_row is None:
+                total += 1
+                continue
+            gt_val = gt_row[gt_ci].strip() if gt_ci < len(gt_row) else ""
+            pred_val = pred_row[pred_ci].strip() if pred_ci < len(pred_row) else ""
+            if values_match(pred_val, gt_val):
+                matches += 1
+            total += 1
+        rate = matches / max(total, 1)
+        d_engine_items[col] = round(rate, 4)
+
+    d_engine_score = 0
+    if d_engine_items:
+        avg_rate = sum(d_engine_items.values()) / len(d_engine_items)
+        d_engine_score = round(10 * avg_rate, 2)
+    result["details"]["D_engine_join"] = {"score": d_engine_score, "max": 10, "detail": d_engine_items}
+
+    # ========== F. JOIN 完整性 (5分) ==========
+    f_join_items = {}
+    result_sql_path = os.path.join(workspace_path, "result.sql")
+    sql_text = ""
+    try:
+        with open(result_sql_path, "r", encoding="utf-8") as f:
+            sql_text = f.read().lower()
+    except Exception:
+        sql_text = ""
+
+    tables_found = 0
+    for tname in SOURCE_TABLES:
+        if tname.lower() in sql_text:
+            tables_found += 1
+            f_join_items[tname] = True
+        else:
+            f_join_items[tname] = False
+
+    f_join_score = round(5 * tables_found / max(len(SOURCE_TABLES), 1), 2)
+    result["details"]["F_join_completeness"] = {
+        "score": f_join_score, "max": 5, "detail": f_join_items,
+    }
+
+    # ========== F. INSERT OVERWRITE (5分) ==========
+    f_insert_items = {}
+    has_overwrite = "insert overwrite" in sql_text
+    has_partition = "partition" in sql_text
+    if has_overwrite and has_partition:
+        f_insert_items["insert_overwrite_partition"] = 5
+    elif has_overwrite:
+        f_insert_items["insert_overwrite_partition"] = 3
+    else:
+        f_insert_items["insert_overwrite_partition"] = 0
+    result["details"]["F_insert_overwrite"] = {
+        "score": f_insert_items["insert_overwrite_partition"], "max": 5, "detail": f_insert_items,
+    }
+
+    # ========== F. 分区值 (5分) ==========
+    f_part_items = {}
+    if "dt" in sql_text and "20260507" in sql_text:
+        f_part_items["partition_value"] = 5
+    else:
+        f_part_items["partition_value"] = 0
+    result["details"]["F_partition_value"] = {
+        "score": f_part_items["partition_value"], "max": 5, "detail": f_part_items,
+    }
+
+    # 汇总
+    total = (10 + b_score + c_score + d_time_score + d_cross_score + d_engine_score
+             + f_join_score + f_insert_items["insert_overwrite_partition"]
+             + f_part_items["partition_value"])
+
+    # ========== G~I 过程评分 ==========
+    TRANSCRIPT_PATH = "/tmp/dataclaw_chat.jsonl"
+    OUTPUT_TABLE_SHORT = OUTPUT_TABLE.split(".")[-1]
+    INPUT_TABLE_SHORT = SOURCE_TABLES[0]
+
+    transcript_entries = []
+    has_transcript = False
+    try:
+        if os.path.exists(TRANSCRIPT_PATH):
+            with open(TRANSCRIPT_PATH, "r", encoding="utf-8", errors="ignore") as f:
+                for line in f:
+                    line = line.strip()
+                    if line:
+                        try:
+                            transcript_entries.append(json.loads(line))
+                        except json.JSONDecodeError:
+                            continue
+            if len(transcript_entries) > 2:
+                has_transcript = True
+    except Exception:
+        pass
+
+    if not has_transcript:
+        result["details"]["G_exploration"] = {"score": 0, "max": 35, "items": {"no_transcript": True}}
+        result["details"]["H_efficiency"] = {"score": 0, "max": 40, "items": {"no_transcript": True}}
+        result["details"]["I_self_verification"] = {"score": 0, "max": 25, "items": {"no_transcript": True}}
+        return finalize(result)
+
+    # Parse transcript into structured events
+    tool_uses = []
+    first_write_result_idx = None
+    last_spark_submit_success_idx = None
+    write_result_count = 0
+    logic_error_retries = 0
+
+    for idx, entry in enumerate(transcript_entries):
+        content = entry.get("content", [])
+        if isinstance(content, str):
+            content = [content]
+
+        for block_str in content:
+            if not isinstance(block_str, str):
+                continue
+            if "ToolUseBlock" in block_str:
+                name_match = re.search(r"name='([^']+)'", block_str)
+                input_match = re.search(r"input=(\{.*\})", block_str)
+                if name_match:
+                    tool_name = name_match.group(1)
+                    tool_input = input_match.group(1) if input_match else ""
+                    tool_uses.append((idx, tool_name, tool_input))
+                    if tool_name == "Write" and "result.sql" in tool_input:
+                        write_result_count += 1
+                        if first_write_result_idx is None:
+                            first_write_result_idx = idx
+            if "ToolResultBlock" in block_str:
+                if "Traceback" in block_str or "Exception" in block_str:
+                    env_errors = ["Derby", "metastore", "HiveMetaStore", "Connection refused",
+                                  "db.lck", "TTransportException", "port 10000"]
+                    is_env_error = any(e in block_str for e in env_errors)
+                    has_spark_submit = any(t[1] == "Bash" and "spark-submit" in t[2] and "result.sql" in t[2]
+                                          for t in tool_uses)
+                    if not is_env_error and has_spark_submit:
+                        logic_error_retries += 1
+                if ("spark-submit" in block_str or "spark-sql" in block_str) and "result.sql" in block_str:
+                    if "Exit Code: 0" in block_str and "Traceback" not in block_str:
+                        last_spark_submit_success_idx = idx
+
+    before_first_write = first_write_result_idx if first_write_result_idx is not None else len(transcript_entries)
+
+    # ===== G. 探索充分性 (35分) =====
+    g_items = {}
+    g1_pass = any(name == "Read" and "schema" in inp.lower()
+                  for idx, name, inp in tool_uses if idx < before_first_write)
+    g_items["G1_source_schema"] = 9 if g1_pass else 0
+
+    g2_pass = any(name == "Bash" and INPUT_TABLE_SHORT in inp
+                  and "SELECT" in inp.upper() and "LIMIT" in inp.upper()
+                  for idx, name, inp in tool_uses if idx < before_first_write)
+    g_items["G2_source_sample"] = 9 if g2_pass else 0
+
+    g3_pass = any(name == "Bash" and INPUT_TABLE_SHORT in inp
+                  and ("GROUP BY" in inp.upper() or "DISTINCT" in inp.upper() or "COUNT" in inp.upper())
+                  for idx, name, inp in tool_uses if idx < before_first_write)
+    g_items["G3_distribution"] = 9 if g3_pass else 0
+
+    g4_pass = any(name == "Bash" and ("DESCRIBE" in inp.upper() or "SHOW CREATE" in inp.upper())
+                  and OUTPUT_TABLE_SHORT in inp
+                  for idx, name, inp in tool_uses if idx < before_first_write)
+    g_items["G4_target_schema"] = 8 if g4_pass else 0
+
+    g_score = sum(g_items.values())
+    result["details"]["G_exploration"] = {"score": g_score, "max": 35, "items": g_items}
+
+    # ===== H. 执行效率 (40分) =====
+    h_items = {}
+    if write_result_count <= 2:
+        h_items["H1_few_submissions"] = 20
+    elif write_result_count <= 4:
+        h_items["H1_few_submissions"] = 13
+    elif write_result_count <= 6:
+        h_items["H1_few_submissions"] = 7
+    else:
+        h_items["H1_few_submissions"] = 0
+
+    if logic_error_retries == 0:
+        h_items["H2_no_logic_errors"] = 13
+    elif logic_error_retries <= 1:
+        h_items["H2_no_logic_errors"] = 7
+    else:
+        h_items["H2_no_logic_errors"] = 0
+
+    h_items["H3_no_redundancy"] = 7
+    h_score = sum(h_items.values())
+    result["details"]["H_efficiency"] = {"score": min(h_score, 40), "max": 40, "items": h_items}
+
+    # ===== I. 自验证行为 (25分) =====
+    i_items = {}
+    post_submit_uses = []
+    if last_spark_submit_success_idx is not None:
+        post_submit_uses = [(idx, name, inp) for idx, name, inp in tool_uses
+                            if idx > last_spark_submit_success_idx]
+
+    i1_pass = any(name == "Bash" and OUTPUT_TABLE_SHORT in inp and "SELECT" in inp.upper()
+                  for _, name, inp in post_submit_uses)
+    i_items["I1_query_output"] = 8 if i1_pass else 0
+
+    i2_pass = any(name == "Bash" and OUTPUT_TABLE_SHORT in inp
+                  and ("COUNT" in inp.upper() or "GROUP BY" in inp.upper())
+                  for _, name, inp in post_submit_uses)
+    i_items["I2_check_count"] = 9 if i2_pass else 0
+
+    i3_pass = any(name == "Bash" and OUTPUT_TABLE_SHORT in inp
+                  and ("LIMIT" in inp.upper() or "SELECT *" in inp.upper())
+                  for _, name, inp in post_submit_uses)
+    i_items["I3_check_values"] = 8 if i3_pass else 0
+
+    i_score = sum(i_items.values())
+    result["details"]["I_self_verification"] = {"score": i_score, "max": 25, "items": i_items}
+
+    return finalize(result)

tasks/offline-compute/HiveSQL/hivesql_003/gt/grade_spec.csv

@@ -0,0 +1,20 @@
+维度,维度全名,子维度,满分,说明
+A,A_executability,executability,10,result.sql 能跑通且产出非空
+B,B_schema,schema,10,18列 + 列名匹配
+C,C_row_alignment,row_consistency,10,行数比例 + key覆盖率
+D,D_time_calculation,time_calculation,30,"instance_run_time, code_run_time 等时间指标"
+D,D_cross_day_split,cross_day_split,15,跨天 trace 按天拆分正确性
+D,D_engine_join,engine_join,10,"serving_id, is_permanent 等 engine 维度列匹配"
+F,F_join_completeness,join_completeness,5,2张源表都被关联
+F,F_insert_overwrite,insert_overwrite,5,写入模式+分区
+F,F_partition_value,partition_value,5,dt=20260507
+
+总计,,,100,
+
+# 评分模板,评分范式: 产物100分 = A(10) + B(10) + C(10) + D(55) + F(15)
+# 难度,HARD
+# 权重,"product=0.7, process=0.3"
+# 范式,new
+# Key列,"trace_id, p_date"
+# 预期列数,
+# 输出表,internal_platform_db.dwd_notebook_killed_instance_detail_d_copilot_cand_query_engine_007

tasks/offline-compute/HiveSQL/hivesql_003/gt/ground_truth.sql

@@ -0,0 +1,166 @@
+INSERT overwrite TABLE internal_platform_db.dwd_notebook_killed_instance_detail_d_copilot_query_engine_007 PARTITION (dt = '20260507')
+WITH base_trace AS (
+    SELECT trace_id
+    FROM internal_platform_db.notebook_span_info_query_engine_007
+    WHERE databus_imp_date >= '2026050700'
+      AND databus_imp_date <= '2026050700'
+      AND compute_type = 'ray'
+      AND service_name = 'notebook-runner'
+      AND span_name IN ('runner.execute', 'runner.killed')
+    GROUP BY trace_id
+    HAVING
+        COUNT(CASE WHEN span_name = 'runner.killed' THEN 1 END) > 0
+        AND
+        COUNT(CASE WHEN span_name = 'runner.execute' THEN 1 END) = 0
+),
+combined_spans AS (
+  SELECT
+      trace_id,
+      span_name,
+      start_time,
+      end_time,
+      datawd_project_id,
+      datawd_task_id,
+      datawd_task_instance_id,
+      compute_type,
+      status_code
+  FROM internal_platform_db.notebook_span_info_query_engine_007
+  WHERE databus_imp_date >= '2026050400'
+    AND databus_imp_date <= '2026050700'
+    AND trace_id IN (SELECT trace_id FROM base_trace)
+    AND span_name IN ('runner.killed', 'runner.start', 'execute.code', 'execute.code.cell', 'client.execute.code', 'set.permanent.compute', 'create.non.permanent.compute')
+
+  UNION ALL
+
+  SELECT
+      trace_id,
+      'runner.execute' AS span_name,
+      MAX(CASE WHEN span_name = 'runner.start' THEN start_time END) AS start_time,
+      MAX(CASE WHEN span_name = 'runner.killed' THEN end_time END) AS end_time,
+      MAX(CASE WHEN span_name = 'runner.killed' THEN datawd_project_id END) AS datawd_project_id,
+      MAX(CASE WHEN span_name = 'runner.killed' THEN datawd_task_id END) AS datawd_task_id,
+      MAX(CASE WHEN span_name = 'runner.killed' THEN datawd_task_instance_id END) AS datawd_task_instance_id,
+      MAX(CASE WHEN span_name = 'runner.killed' THEN compute_type END) AS compute_type,
+      2 AS status_code
+  FROM internal_platform_db.notebook_span_info_query_engine_007
+  WHERE databus_imp_date >= '2026050400'
+    AND databus_imp_date <= '2026050700'
+    AND trace_id IN (SELECT trace_id FROM base_trace)
+    AND span_name IN ('runner.start', 'runner.killed')
+  GROUP BY trace_id
+),
+base_data_time_fixed AS (
+  SELECT
+      trace_id,
+      span_name,
+      CASE
+          WHEN span_name = 'runner.killed'
+            THEN MIN(CASE WHEN span_name IN ('execute.code', 'execute.code.cell', 'client.execute.code', 'runner.killed') THEN start_time END)
+                    OVER(PARTITION BY trace_id)
+          ELSE start_time
+      END AS start_time,
+      end_time,
+      datawd_project_id,
+      datawd_task_id,
+      datawd_task_instance_id,
+      compute_type,
+      status_code
+  FROM combined_spans
+),
+base_data AS (
+  SELECT
+      trace_id,
+      span_name,
+      start_time,
+      end_time,
+      datawd_project_id,
+      datawd_task_id,
+      datawd_task_instance_id,
+      compute_type,
+      status_code,
+      from_unixtime(CAST(start_time AS BIGINT) / 1000) as start_date,
+      from_unixtime(CAST(end_time AS BIGINT) / 1000) as end_date,
+      datediff(from_unixtime(CAST(end_time AS BIGINT) / 1000), from_unixtime(CAST(start_time AS BIGINT) / 1000)) AS diff_days
+  FROM base_data_time_fixed
+),
+pos_series AS (
+    SELECT 0 AS pos UNION ALL SELECT 1 AS pos UNION ALL SELECT 2 AS pos UNION ALL SELECT 3 AS pos
+),
+daily_split_spans AS (
+    SELECT
+        trace_id,
+        span_name,
+        datawd_project_id,
+        datawd_task_id,
+        datawd_task_instance_id,
+        compute_type,
+        status_code,
+        start_date,
+        end_date,
+        diff_days,
+        date_add(b.start_date, s.pos) AS calc_date,
+        CASE WHEN s.pos = 0 THEN CAST(b.start_time AS BIGINT)
+             ELSE unix_timestamp(cast(date_add(b.start_date, s.pos) as timestamp)) * 1000
+        END AS start_time,
+        CASE WHEN s.pos = b.diff_days THEN CAST(b.end_time AS BIGINT)
+             ELSE (unix_timestamp(cast(date_add(b.start_date, s.pos + 1) as timestamp)) * 1000) - 1
+        END AS end_time,
+        CAST(b.start_time AS BIGINT) AS span_start_time,
+        CAST(b.end_time AS BIGINT) AS span_end_time
+    FROM base_data b
+    INNER JOIN pos_series s ON s.pos <= b.diff_days
+),
+trace_time_metrics AS (
+    SELECT
+        trace_id,
+        calc_date,
+        MAX(datawd_project_id) AS datawd_project_id,
+        MAX(datawd_task_id) AS datawd_task_id,
+        MAX(datawd_task_instance_id) AS datawd_task_instance_id,
+        MAX(compute_type) AS compute_type,
+        MAX(MAX(CASE WHEN span_name = 'runner.execute' THEN status_code END)) OVER(PARTITION BY trace_id) AS status_code,
+        CAST(ROUND((MAX(CASE WHEN span_name = 'runner.execute' THEN end_time END) -
+               MIN(CASE WHEN span_name = 'runner.execute' THEN start_time END)) / 1000.0) AS INT) AS instance_run_time,
+        CAST(ROUND((MAX(CASE WHEN span_name IN ('execute.code', 'client.execute.code', 'execute.code.cell', 'runner.killed' ) THEN end_time END) -
+               MIN(CASE WHEN span_name IN ('execute.code', 'client.execute.code', 'execute.code.cell', 'runner.killed' ) THEN start_time END)) / 1000.0) AS INT) AS code_run_time,
+        CAST(ROUND((MAX(CASE WHEN span_name IN ('set.permanent.compute', 'create.non.permanent.compute') THEN end_time END) -
+               MIN(CASE WHEN span_name IN ('set.permanent.compute', 'create.non.permanent.compute') THEN start_time END)) / 1000.0) AS INT) AS resource_wait_time,
+        from_unixtime(MIN(CASE WHEN span_name IN ('execute.code', 'client.execute.code', 'execute.code.cell', 'runner.killed' ) THEN start_time END) / 1000) AS code_start_time,
+        from_unixtime(MAX(CASE WHEN span_name IN ('execute.code', 'client.execute.code', 'execute.code.cell', 'runner.killed' ) THEN end_time END) / 1000) AS code_end_time,
+        from_unixtime(MIN(CASE WHEN span_name = 'runner.execute' THEN span_start_time END) / 1000) AS instance_start_time,
+        from_unixtime(MAX(CASE WHEN span_name = 'runner.execute' THEN span_end_time END) / 1000) AS instance_end_time
+    FROM daily_split_spans
+    GROUP BY trace_id, calc_date
+)
+SELECT
+  t1.calc_date AS p_date,
+  t1.trace_id,
+  t1.datawd_project_id,
+  t1.datawd_task_id,
+  t1.datawd_task_instance_id,
+  t1.compute_type,
+  t1.status_code,
+  t1.instance_run_time,
+  t1.code_run_time,
+  t1.resource_wait_time,
+  t1.code_start_time,
+  t1.code_end_time,
+  t1.instance_start_time,
+  t1.instance_end_time,
+  t2.serving_id,
+  CAST(t2.is_permanent AS BOOLEAN) AS is_permanent,
+  CAST(t2.apply_for_gpu_count AS INT) AS apply_for_gpu_count
+FROM trace_time_metrics t1 LEFT JOIN (
+  SELECT
+    trace_id,
+    MAX(serving_id) AS serving_id,
+    MAX(is_permanent) AS is_permanent,
+    SUM(CAST(replicas AS INT) * CAST(num_gpu AS INT)) AS apply_for_gpu_count
+  FROM internal_platform_db.notebook_engine_info_query_engine_007
+  WHERE databus_imp_date >= '2026050400'
+    AND databus_imp_date <= '2026050700'
+    AND compute_type = 'ray'
+    AND service_name = 'notebook-runner'
+  GROUP BY trace_id
+) t2 ON t1.trace_id = t2.trace_id
+;

tasks/offline-compute/HiveSQL/hivesql_003/init/init_db.py

@@ -0,0 +1,53 @@
+#!/usr/bin/env python3
+import re
+from pyspark.sql import SparkSession
+
+spark = SparkSession.builder \
+    .appName('hivesql_bench_init') \
+    .enableHiveSupport() \
+    .config('spark.sql.warehouse.dir', '/tmp/hive_warehouse') \
+    .getOrCreate()
+
+spark.sql('CREATE DATABASE IF NOT EXISTS internal_platform_db')
+
+
+def _execute_sql_file(spark, sql_path):
+    """Read SQL file, remove SuperSQL SET headers, split by semicolons, execute."""
+    with open(sql_path, 'r', encoding='utf-8') as f:
+        content = f.read()
+    content = re.sub(r'^\s*set\s+query_engine\.\S+\n?', '', content, flags=re.IGNORECASE)
+    stmts, cur, in_sq, in_dq, i = [], [], False, False, 0
+    while i < len(content):
+        ch = content[i]
+        if ch == '\\' and i + 1 < len(content):
+            cur.append(ch); cur.append(content[i + 1]); i += 2; continue
+        if ch == '-' and i + 1 < len(content) and content[i + 1] == '-' and not in_sq and not in_dq:
+            while i < len(content) and content[i] != '\n':
+                i += 1
+            cur.append('\n'); continue
+        if ch == "'" and not in_dq:
+            in_sq = not in_sq
+        elif ch == '"' and not in_sq:
+            in_dq = not in_dq
+        if ch == ';' and not in_sq and not in_dq:
+            s = ''.join(cur).strip()
+            if s:
+                stmts.append(s)
+            cur = []
+        else:
+            cur.append(ch)
+        i += 1
+    last = ''.join(cur).strip()
+    if last:
+        stmts.append(last)
+    for stmt in stmts:
+        spark.sql(stmt)
+
+
+_execute_sql_file(spark, '/tmp_workspace/init_db.sql')
+
+tables = spark.sql('SHOW TABLES IN internal_platform_db').collect()
+print(f'Init complete, {len(tables)} tables created')
+for t in tables:
+    print(f'  - {t.namespace}.{t.tableName}')
+spark.stop()

tasks/offline-compute/HiveSQL/hivesql_003/init/init_db.sql

@@ -0,0 +1,104 @@
+CREATE DATABASE IF NOT EXISTS internal_platform_db;
+
+CREATE TABLE IF NOT EXISTS internal_platform_db.notebook_span_info_query_engine_007 (
+    databus_imp_date STRING,
+    trace_id STRING,
+    span_kind STRING,
+    error_code STRING,
+    service_name STRING,
+    message STRING,
+    parent_span_id STRING,
+    span_name STRING,
+    duration BIGINT,
+    span_id STRING,
+    space_id STRING,
+    span_type STRING,
+    start_time STRING,
+    end_time STRING,
+    service_instance STRING,
+    status_code INT,
+    datawd_project_id STRING,
+    datawd_task_id STRING,
+    datawd_task_instance_id STRING,
+    compute_type STRING
+) STORED AS ORC;
+
+INSERT INTO TABLE internal_platform_db.notebook_span_info_query_engine_007 VALUES
+('2026050700','trace_001','SERVER','0','notebook-runner','','','runner.killed',5000,'s1','','','1746590400000','1746590405000','inst1',2,'proj_01','task_01','inst_01','ray'),
+('2026050700','trace_001','SERVER','0','notebook-runner','','','runner.start',1000,'s2','','','1746590390000','1746590391000','inst1',0,'proj_01','task_01','inst_01','ray'),
+('2026050500','trace_001','SERVER','0','notebook-runner','','','execute.code.cell',3000,'s3','','','1746590392000','1746590395000','inst1',0,'proj_01','task_01','inst_01','ray'),
+('2026050500','trace_001','SERVER','0','notebook-runner','','','set.permanent.compute',2000,'s4','','','1746590388000','1746590390000','inst1',0,'proj_01','task_01','inst_01','ray'),
+('2026050700','trace_002','SERVER','0','notebook-runner','','','runner.killed',8000,'s5','','','1746590500000','1746590508000','inst2',2,'proj_02','task_02','inst_02','ray'),
+('2026050700','trace_002','SERVER','0','notebook-runner','','','runner.start',1000,'s6','','','1746590490000','1746590491000','inst2',0,'proj_02','task_02','inst_02','ray'),
+('2026050600','trace_002','SERVER','0','notebook-runner','','','client.execute.code',6000,'s7','','','1746590492000','1746590498000','inst2',0,'proj_02','task_02','inst_02','ray'),
+('2026050600','trace_002','SERVER','0','notebook-runner','','','create.non.permanent.compute',3000,'s8','','','1746590487000','1746590490000','inst2',0,'proj_02','task_02','inst_02','ray'),
+('2026050700','trace_003','SERVER','0','notebook-runner','','','runner.killed',4000,'s9','','','1746590600000','1746590604000','inst3',2,'proj_03','task_03','inst_03','ray'),
+('2026050700','trace_003','SERVER','0','notebook-runner','','','runner.execute',10000,'s10','','','1746590580000','1746590610000','inst3',0,'proj_03','task_03','inst_03','ray');
+
+CREATE TABLE IF NOT EXISTS internal_platform_db.notebook_engine_info_query_engine_007 (
+    databus_imp_date STRING,
+    trace_id STRING,
+    service_name STRING,
+    compute_type STRING,
+    datawd_project_id STRING,
+    datawd_task_id STRING,
+    datawd_task_instance_id STRING,
+    is_permanent STRING,
+    serving_id STRING,
+    user_name STRING,
+    is_model_service STRING,
+    compute_info STRING,
+    num_gpu STRING,
+    replicas STRING,
+    gpu_type STRING,
+    `timestamp` STRING,
+    ray_job_id STRING,
+    reason STRING,
+    log_url STRING,
+    dashboard_url STRING
+) STORED AS ORC;
+
+INSERT INTO TABLE internal_platform_db.notebook_engine_info_query_engine_007 VALUES
+('2026050700','trace_001','notebook-runner','ray','proj_01','task_01','inst_01','true','srv_001','user1','false','{}','2','1','A100','1746590400000','job1','','',''),
+('2026050700','trace_002','notebook-runner','ray','proj_02','task_02','inst_02','false','srv_002','user2','false','{}','4','2','V100','1746590500000','job2','','','');
+
+CREATE TABLE IF NOT EXISTS internal_platform_db.dwd_notebook_killed_instance_detail_d_copilot_query_engine_007 (
+    p_date STRING,
+    trace_id STRING,
+    datawd_project_id STRING,
+    datawd_task_id STRING,
+    datawd_task_instance_id STRING,
+    compute_type STRING,
+    status_code INT,
+    instance_run_time INT,
+    code_run_time INT,
+    resource_wait_time INT,
+    code_start_time STRING,
+    code_end_time STRING,
+    instance_start_time STRING,
+    instance_end_time STRING,
+    serving_id STRING,
+    is_permanent BOOLEAN,
+    apply_for_gpu_count INT
+) PARTITIONED BY (dt STRING) STORED AS ORC;
+
+-- Agent 候选目标表
+CREATE TABLE IF NOT EXISTS internal_platform_db.dwd_notebook_killed_instance_detail_d_copilot_cand_query_engine_007 (
+    p_date STRING,
+    trace_id STRING,
+    datawd_project_id STRING,
+    datawd_task_id STRING,
+    datawd_task_instance_id STRING,
+    compute_type STRING,
+    status_code INT,
+    instance_run_time INT,
+    code_run_time INT,
+    resource_wait_time INT,
+    code_start_time STRING,
+    code_end_time STRING,
+    instance_start_time STRING,
+    instance_end_time STRING,
+    serving_id STRING,
+    is_permanent BOOLEAN,
+    apply_for_gpu_count INT
+) PARTITIONED BY (dt STRING) STORED AS ORC;

tasks/offline-compute/HiveSQL/hivesql_003/init/schema.sql

@@ -0,0 +1,45 @@
+CREATE TABLE IF NOT EXISTS internal_platform_db.notebook_span_info_copilot (
+    `databus_imp_date` STRING COMMENT 'primary partition',
+    `trace_id` STRING COMMENT 'trace_id',
+    `span_kind` STRING COMMENT 'span_kind',
+    `error_code` STRING COMMENT 'error_code',
+    `service_name` STRING COMMENT 'service_name',
+    `message` STRING COMMENT 'message',
+    `parent_span_id` STRING COMMENT 'parent_span_id',
+    `span_name` STRING COMMENT 'span_name',
+    `duration` BIGINT COMMENT 'duration',
+    `span_id` STRING COMMENT 'span_id',
+    `space_id` STRING COMMENT 'space_id',
+    `span_type` STRING COMMENT 'span_type',
+    `start_time` STRING COMMENT 'start_time',
+    `end_time` STRING COMMENT 'end_time',
+    `service_instance` STRING COMMENT 'service_instance',
+    `status_code` INT COMMENT 'status_code',
+    `datawd_project_id` STRING COMMENT 'datawd_project_id',
+    `datawd_task_id` STRING COMMENT 'datawd_task_id',
+    `datawd_task_instance_id` STRING COMMENT 'datawd_task_instance_id',
+    `compute_type` STRING COMMENT 'compute_type'
+) STORED AS ORC;
+
+CREATE TABLE IF NOT EXISTS internal_platform_db.notebook_engine_info_copilot (
+    `databus_imp_date` STRING COMMENT 'primary partition',
+    `trace_id` STRING COMMENT 'trace_id',
+    `service_name` STRING COMMENT 'service_name',
+    `compute_type` STRING COMMENT 'compute_type',
+    `datawd_project_id` STRING COMMENT 'datawd_project_id',
+    `datawd_task_id` STRING COMMENT 'datawd_task_id',
+    `datawd_task_instance_id` STRING COMMENT 'datawd_task_instance_id',
+    `is_permanent` STRING COMMENT 'is_permanent',
+    `serving_id` STRING COMMENT 'serving_id',
+    `user_name` STRING COMMENT 'user_name',
+    `is_model_service` STRING COMMENT 'is_model_service',
+    `compute_info` STRING COMMENT 'compute_info',
+    `num_gpu` STRING COMMENT 'num_gpu',
+    `replicas` STRING COMMENT 'replicas',
+    `gpu_type` STRING COMMENT 'gpu_type',
+    `timestamp` STRING COMMENT 'timestamp',
+    `ray_job_id` STRING,
+    `reason` STRING,
+    `log_url` STRING,
+    `dashboard_url` STRING
+) STORED AS ORC;

tasks/offline-compute/HiveSQL/hivesql_003/task.md

@@ -0,0 +1,34 @@
+---
+id: offline-compute_HiveSQL_hivesql_003
+name: 从 `internal_platform_db.notebook_span_info_query_engine
+category: offline-compute/HiveSQL
+timeout_seconds: 600
+modality: pure-text
+engine: hivesql
+---
+## Prompt
+1) 任务目标
+识别并统计 Notebook 管道中被异常 kill 且缺失正常完成信号的 Ray 实例，按天拆分计算运行时长、代码执行时长、资源等待时长及 GPU 申请数。
+
+2) 输入
+- `internal_platform_db.notebook_span_info_query_engine_007`
+- `internal_platform_db.notebook_engine_info_query_engine_007`
+
+3) 处理规则
+- 目标实例筛选：从 span_info 筛选 `compute_type='ray'` 且 `service_name='notebook-runner'` 的实例，选取有 `span_name='runner.killed'` 但无 `span_name='runner.execute'` 的 `trace_id`。
+- 合成运行区间：对每个目标 trace_id，取 `span_name='runner.start'` 的 start_time 作为实例开始时间，取 `span_name='runner.killed'` 的 end_time 作为实例结束时间，合成 `runner.execute` 区间，`status_code` 固定为 2。
+- 跨天拆分：将合成区间按自然日拆分为多条记录，每条对应一天内的部分，`calc_date` 为拆分后的日期。
+- 时长计算（按 trace_id 与 calc_date）：
+  - instance_run_time：合成区间在当天的起止时间差（毫秒转秒）。
+  - code_run_time：`span_name` 为 'execute.code', 'execute.code.cell', 'client.execute.code', 'runner.killed' 在当天的起止时间差（毫秒转秒）。
+  - resource_wait_time：`span_name` 为 'set.permanent.compute', 'create.non.permanent.compute' 在当天的起止时间差（毫秒转秒）。
+  - 输出 code_start_time/code_end_time/instance_start_time/instance_end_time（时间戳转字符串）。
+- 关联引擎信息：左关联 engine_info（过滤 `compute_type='ray'` 且 `service_name='notebook-runner'`），关联键 `trace_id`，获取 serving_id、is_permanent，计算 `apply_for_gpu_count = SUM(replicas * num_gpu)`。
+
+4) 输出要求
+输出字段顺序：dt（STRING，分区）、p_date（STRING，calc_date）、trace_id（STRING）、datawd_project_id（STRING）、datawd_task_id（STRING）、datawd_task_instance_id（STRING）、compute_type（STRING）、status_code（INT）、instance_run_time（INT）、code_run_time（INT）、resource_wait_time（INT）、code_start_time（STRING）、code_end_time（STRING）、instance_start_time（STRING）、instance_end_time（STRING）、serving_id（STRING）、is_permanent（BOOLEAN）、apply_for_gpu_count（INT）。
+
+5) 写入要求
+目标表：`internal_platform_db.dwd_notebook_killed_instance_detail_d_copilot_cand_query_engine_007`。分区字段：dt。写入分区：`dt='20260507'`。写入模式：覆盖。
+
+请将最终 HiveSQL 写入 result.sql 并执行。

tasks/offline-compute/HiveSQL/hivesql_004/gt/expected.csv

@@ -0,0 +1,6 @@
+username,application_group,application_group_owner,product_id,product_name,product_owner,obs_product_id,obs_product_name,obs_product_owner,department,plan_product_id,plan_product_name,plan_product_owner,application_group_numbers,application_group_owners,product_owners,obs_product_owners,plan_product_owners,bg,id,dt
+user_e,group_delta,4,105,prod_v,205,305,obs_prod_e,405,dept_finance,505,plan_e,605,2,owner5,powner5,obs_owner5,plan_owner5,BG2,5,20260507
+user_d,group_alpha,1,104,prod_w,204,304,obs_prod_d,404,dept_eng,504,plan_d,604,7,owner4,powner4,obs_owner4,plan_owner4,BG3,4,20260507
+user_a,group_alpha,1,101,prod_x,201,301,obs_prod_a,401,dept_eng,501,plan_a,601,5,owner1,powner1,obs_owner1,plan_owner1,BG1,1,20260507
+user_b,group_beta,2,102,prod_y,202,302,obs_prod_b,402,dept_sales,502,plan_b,602,10,owner2,powner2,obs_owner2,plan_owner2,BG2,2,20260507
+user_c,group_gamma,3,103,prod_z,203,303,obs_prod_c,403,dept_hr,503,plan_c,603,3,owner3,powner3,obs_owner3,plan_owner3,BG1,3,20260507

tasks/offline-compute/HiveSQL/hivesql_004/gt/grade.py

@@ -0,0 +1,710 @@
+"""hivesql_011 精细评分脚本
+
+业务场景：应用组与产品归属关系按天做历史快照，追加 dt 分区列全量覆盖写入
+难度: EASY | 特征: INSERT_OVERWRITE|PARTITION|SINGLE_TABLE|APPEND_DT_COLUMN
+
+评分范式: 产物100分 = A(15) + B(10) + C(15) + D(40) + F(20)
+  A_executability       (15分): result.sql 能跑通且产出非空
+  B_schema              (10分): 21列(5) + 列名匹配(5)
+  C_row_alignment     (15分): 行数比例(7) + key覆盖率(8)
+  D_field_value_match   (25分): 非key字段逐列值匹配率
+  D_field_completeness  (15分): 关键字段非空/非空串比例
+  F_insert_overwrite     (5分):  INSERT OVERWRITE + PARTITION
+  F_partition_value      (5分):  dt 分区值 = 20260507
+  F_dt_column           (10分):  dt 派生列存在于输出且值正确
+
+权重: EASY -> product=0.5, process=0.5
+"""
+import os
+import re
+import subprocess
+import tempfile
+import json
+import math
+
+
+def grade(workspace_path, **kwargs):
+
+    # ========== Case 配置 ==========
+    OUTPUT_TABLE = "internal_platform_db.app_group_product_info_history_cand_query_engine_011"
+    DIFFICULTY = "EASY"
+    SOURCE_TABLES = ["app_group_product_info_query_engine_011", "app_group_product_info_history_query_engine_011"]
+    KEY_COLUMNS = ["product_id", "obs_product_id", "plan_product_id", "id", "dt"]
+    EXPECTED_COL_COUNT = 21
+    KEY_FIELDS = ["username", "application_group", "product_id", "product_name", "bg", "id"]
+    GT_TABLE = OUTPUT_TABLE.replace("_cand_", "_")
+
+    DIFFICULTY_WEIGHTS = {
+        "EASY": (0.5, 0.5),
+        "MEDIUM": (0.6, 0.4),
+        "HARD": (0.7, 0.3),
+        "EXPERT": (0.8, 0.2),
+    }
+
+    _SPARK_SUBMIT_TIMEOUT = 300
+    _JSON_START = "__GRADE_JSON_START__"
+    _JSON_END = "__GRADE_JSON_END__"
+
+    result = {
+        "overall_score": 0.0,
+        "total_points": 0,
+        "grade": "",
+        "details": {},
+        "diagnostics": [],
+    }
+
+    # ========== 内部辅助函数 ==========
+
+    def values_match(pred_val, gt_val, abs_tol=1e-6, rel_tol=1e-4):
+        if pred_val is None and gt_val is None:
+            return True
+        if pred_val is None or gt_val is None:
+            return False
+        s_pred = str(pred_val).strip()
+        s_gt = str(gt_val).strip()
+        if s_pred == s_gt:
+            return True
+        try:
+            pv = float(s_pred)
+            gv = float(s_gt)
+            if math.isnan(pv) and math.isnan(gv):
+                return True
+            if math.isnan(pv) or math.isnan(gv):
+                return False
+            if abs(gv) < abs_tol:
+                return abs(pv - gv) <= abs_tol
+            return abs(pv - gv) <= abs_tol or abs(pv - gv) / max(abs(gv), 1e-12) <= rel_tol
+        except (ValueError, TypeError):
+            pass
+        return s_pred.lower() == s_gt.lower()
+
+    def _run_spark_script(script_code, timeout=_SPARK_SUBMIT_TIMEOUT):
+        with tempfile.NamedTemporaryFile(mode="w", suffix=".py", delete=False, encoding="utf-8") as f:
+            f.write(script_code)
+            script_path = f.name
+        try:
+            r = subprocess.run(
+                ["spark-submit", script_path],
+                capture_output=True, text=True, timeout=timeout,
+            )
+            stdout = r.stdout or ""
+            if _JSON_START in stdout and _JSON_END in stdout:
+                json_str = stdout.split(_JSON_START)[1].split(_JSON_END)[0].strip()
+                return json.loads(json_str), None
+            else:
+                if r.returncode == 0:
+                    for line in stdout.splitlines():
+                        if line.strip().startswith("Traceback"):
+                            return None, f"spark-submit error: {line}"
+                    return None, "spark-submit 无 JSON 输出"
+                err_msg = (r.stderr or "")[-500:]
+                return None, f"spark-submit failed: {err_msg}"
+        except subprocess.TimeoutExpired:
+            return None, f"spark-submit 超时 ({timeout}s)"
+        except Exception as e:
+            return None, f"spark-submit 异常: {e}"
+        finally:
+            try:
+                os.unlink(script_path)
+            except OSError:
+                pass
+
+    def read_table_via_spark_submit(table_name):
+        """Read table via spark-submit subprocess. Returns (cols, rows_as_lists)."""
+        if not re.match(r'^[a-zA-Z_][a-zA-Z0-9_.]*$', table_name):
+            raise ValueError(f"非法表名: {table_name}")
+        read_script = f'''
+import json
+from pyspark.sql import SparkSession
+spark = SparkSession.builder.appName("grade_read").enableHiveSupport() \\
+    .config("spark.sql.warehouse.dir", "/tmp/hive_warehouse").getOrCreate()
+try:
+    df = spark.sql("SELECT * FROM {table_name}")
+    cols = [c.lower() for c in df.columns]
+    rows = [[str(v) if v is not None else "" for v in row] for row in df.collect()]
+    print("{_JSON_START}")
+    print(json.dumps({{"cols": cols, "rows": rows}}, ensure_ascii=False))
+    print("{_JSON_END}")
+except Exception as e:
+    print("{_JSON_START}")
+    print(json.dumps({{"error": str(e)}}))
+    print("{_JSON_END}")
+finally:
+    spark.stop()
+'''
+        data, err = _run_spark_script(read_script)
+        if err:
+            raise RuntimeError(f"read_table failed: {err}")
+        if "error" in data:
+            raise RuntimeError(f"query failed: {data['error']}")
+        return data["cols"], data["rows"]
+
+    # SQL executor template (self-contained, no external dependency)
+    _SQL_EXEC_TEMPLATE = '''
+import json, re
+from pyspark.sql import SparkSession
+spark = SparkSession.builder.appName("{app_name}").enableHiveSupport() \\
+    .config("spark.sql.warehouse.dir", "/tmp/hive_warehouse").getOrCreate()
+try:
+    with open("{sql_file}", "r", encoding="utf-8") as _f:
+        _sql = _f.read()
+    _sql = re.sub(r"^\\s*set\\s+query_engine\\.\\S+\\n?", "", _sql, flags=re.IGNORECASE)
+    _stmts, _cur, _in_sq, _in_dq, _i = [], [], False, False, 0
+    while _i < len(_sql):
+        _ch = _sql[_i]
+        if _ch == "\\\\" and _i + 1 < len(_sql):
+            _cur.append(_ch); _cur.append(_sql[_i+1]); _i += 2; continue
+        if _ch == "-" and _i+1 < len(_sql) and _sql[_i+1] == "-" and not _in_sq and not _in_dq:
+            while _i < len(_sql) and _sql[_i] != "\\n": _i += 1
+            _cur.append("\\n"); continue
+        if _ch == "'" and not _in_dq: _in_sq = not _in_sq
+        elif _ch == '"' and not _in_sq: _in_dq = not _in_dq
+        if _ch == ";" and not _in_sq and not _in_dq:
+            _s = "".join(_cur).strip()
+            if _s: _stmts.append(_s)
+            _cur = []
+        else:
+            _cur.append(_ch)
+        _i += 1
+    _last = "".join(_cur).strip()
+    if _last: _stmts.append(_last)
+    for _stmt in _stmts:
+        spark.sql(_stmt)
+    print("{_JSON_START}")
+    print(json.dumps({{"ok": True}}))
+    print("{_JSON_END}")
+except Exception as e:
+    print("{_JSON_START}")
+    print(json.dumps({{"ok": False, "error": str(e)}}))
+    print("{_JSON_END}")
+finally:
+    spark.stop()
+'''
+
+    def execute_result_sql():
+        result_sql = os.path.join(workspace_path, "result.sql")
+        if not os.path.exists(result_sql):
+            return False, "no_result_file"
+        # 替换 数据平台WD时间变量（沙箱 spark-sql 不支持 ${...} 语法）
+        with open(result_sql, 'r', encoding='utf-8') as _rf:
+            _sql_text = _rf.read()
+        _bizdate = '20260507'
+        _sql_text = re.sub(r'\${bdp\.system\.bizdate(?:[+-]\d+)?}', _bizdate, _sql_text)
+        _sql_text = re.sub(r'\${yyyymmdd(?:[+-]\d+)?}', _bizdate, _sql_text)
+        _sql_text = re.sub(r'\${bizdate(?:[+-]\d+)?}', _bizdate, _sql_text)
+        _sql_text = re.sub(r'\${[^}]*date[^}]*}', _bizdate, _sql_text)
+        with open(result_sql, 'w', encoding='utf-8') as _wf:
+            _wf.write(_sql_text)
+        script = _SQL_EXEC_TEMPLATE.format(app_name="grade_exec", sql_file=result_sql,
+                                            _JSON_START=_JSON_START, _JSON_END=_JSON_END)
+        data, err = _run_spark_script(script)
+        if err:
+            return False, f"execution_error: {err}"
+        if data and data.get("ok"):
+            return True, None
+        return False, f"execution_error: {data.get('error', 'unknown') if data else 'no output'}"
+
+    def execute_ground_truth_sql():
+        gt_sql = os.path.join(workspace_path, "gt", "ground_truth.sql")
+        if not os.path.exists(gt_sql):
+            return False, "ground_truth.sql not found"
+        script = _SQL_EXEC_TEMPLATE.format(app_name="grade_gt", sql_file=gt_sql,
+                                            _JSON_START=_JSON_START, _JSON_END=_JSON_END)
+        data, err = _run_spark_script(script)
+        if err:
+            return False, f"gt_execution_error: {err}"
+        if data and data.get("ok"):
+            return True, None
+        return False, f"gt_execution_error: {data.get('error', 'unknown') if data else 'no output'}"
+
+    def truncate_table(table_name):
+        script = f'''
+from pyspark.sql import SparkSession
+spark = SparkSession.builder.appName("truncate").enableHiveSupport() \\
+    .config("spark.sql.warehouse.dir", "/tmp/hive_warehouse").getOrCreate()
+spark.sql("TRUNCATE TABLE {table_name}")
+spark.stop()
+'''
+        try:
+            _run_spark_script(script, timeout=120)
+        except Exception:
+            pass
+
+    def restore_hive_site():
+        """Restore hive-site.xml to canonical state (agent may have modified it)."""
+        canonical_hive_site = '''<?xml version="1.0"?>
+<?xml-stylesheet type="text/xsl" href="configuration.xsl"?>
+<configuration>
+  <property>
+    <name>hive.metastore.uris</name>
+    <value>thrift://localhost:9083</value>
+  </property>
+  <property>
+    <name>hive.metastore.warehouse.dir</name>
+    <value>/tmp/hive_warehouse</value>
+  </property>
+  <property>
+    <name>javax.jdo.option.ConnectionURL</name>
+    <value>jdbc:derby:;databaseName=/tmp/hive_metastore_db;create=true</value>
+  </property>
+  <property>
+    <name>javax.jdo.option.ConnectionDriverName</name>
+    <value>org.apache.derby.jdbc.EmbeddedDriver</value>
+  </property>
+  <property>
+    <name>datanucleus.schema.autoCreateAll</name>
+    <value>true</value>
+  </property>
+  <property>
+    <name>hive.metastore.schema.verification</name>
+    <value>false</value>
+  </property>
+</configuration>
+'''
+        hive_site_path = os.path.join(os.environ.get('SPARK_HOME', '/opt/spark'), 'conf', 'hive-site.xml')
+        try:
+            with open(hive_site_path, 'w') as f:
+                f.write(canonical_hive_site)
+        except Exception:
+            pass
+
+    def finalize(result):
+        product_weight, process_weight = DIFFICULTY_WEIGHTS.get(DIFFICULTY, (0.7, 0.3))
+        product_dims = ["A_executability"] + ['B_schema', 'C_row_alignment', 'D_field_completeness', 'D_field_value_match', 'F_dt_column', 'F_insert_overwrite', 'F_partition_value']
+        product_raw = sum(result["details"].get(d, {}).get("score", 0) for d in product_dims)
+        product_ratio = product_raw / 100.0
+        for dim in ["H_efficiency"]:
+            if dim in result["details"]:
+                raw = result["details"][dim].get("score", 0)
+                result["details"][dim]["score_before_scaling"] = raw
+                result["details"][dim]["score"] = round(raw * product_ratio, 2)
+                result["details"][dim]["product_ratio"] = round(product_ratio, 4)
+        process_dims = ["G_exploration", "H_efficiency", "I_self_verification"]
+        process_raw = sum(result["details"].get(d, {}).get("score", 0) for d in process_dims)
+        product_score = round(product_raw * product_weight, 2)
+        process_score = round(process_raw * process_weight, 2)
+        total = round(product_score + process_score, 2)
+        result["total_points"] = total
+        result["product_points"] = product_score
+        result["process_points"] = round(process_score, 2)
+        result["weights"] = {"product": product_weight, "process": process_weight}
+        result["overall_score"] = round(total / 100.0, 4)
+        if total >= 90:
+            result["grade"] = "优秀"
+        elif total >= 75:
+            result["grade"] = "良好"
+        elif total >= 60:
+            result["grade"] = "合格"
+        elif total >= 40:
+            result["grade"] = "偏弱"
+        else:
+            result["grade"] = "不合格"
+        return result
+
+    # Restore hive-site.xml (agent may have modified it)
+    restore_hive_site()
+
+    # ========== A. 可执行性 (15分) ==========
+    a_items = {"A1_exec_ok": 0, "A2_has_data": 0}
+    exec_ok, exec_err = execute_result_sql()
+    if not exec_ok:
+        detail = "未产出 result.sql" if exec_err == "no_result_file" else str(exec_err)[:200]
+        result["details"]["A_executability"] = {"score": 0, "max": 15, "detail": detail}
+        result["error"] = detail
+        for dim in ['B_schema', 'C_row_alignment', 'D_field_completeness', 'D_field_value_match', 'F_dt_column', 'F_insert_overwrite', 'F_partition_value']:
+            result["details"][dim] = {"score": 0, "max": 0, "items": {}}
+        return finalize(result)
+
+    a_items["A1_exec_ok"] = 8
+
+    pred_headers, pred_rows = [], []
+    try:
+        pred_headers, pred_rows = read_table_via_spark_submit(OUTPUT_TABLE)
+    except Exception as e:
+        result["diagnostics"].append(f"read_pred_failed: {e}")
+
+    if not pred_rows:
+        result["details"]["A_executability"] = {"score": 8, "max": 15, "items": a_items}
+        for dim in ['B_schema', 'C_row_alignment', 'D_field_completeness', 'D_field_value_match', 'F_dt_column', 'F_insert_overwrite', 'F_partition_value']:
+            result["details"][dim] = {"score": 0, "max": 0, "items": {}}
+        return finalize(result)
+
+    a_items["A2_has_data"] = 7
+    result["details"]["A_executability"] = {"score": 15, "max": 15, "items": a_items}
+
+    # ========== Execute GT + Read GT ==========
+    truncate_table(GT_TABLE)
+    gt_ok, gt_err = execute_ground_truth_sql()
+    if not gt_ok:
+        result["error"] = f"ground_truth failed: {gt_err}"
+        for dim in ['B_schema', 'C_row_alignment', 'D_field_completeness', 'D_field_value_match', 'F_dt_column', 'F_insert_overwrite', 'F_partition_value']:
+            result["details"][dim] = {"score": 0, "max": 0, "items": {}}
+        return finalize(result)
+
+    gt_headers, gt_rows = [], []
+    try:
+        gt_headers, gt_rows = read_table_via_spark_submit(GT_TABLE)
+    except Exception as e:
+        result["error"] = f"read_gt_failed: {e}"
+        for dim in ['B_schema', 'C_row_alignment', 'D_field_completeness', 'D_field_value_match', 'F_dt_column', 'F_insert_overwrite', 'F_partition_value']:
+            result["details"][dim] = {"score": 0, "max": 0, "items": {}}
+        return finalize(result)
+
+    # ========== B/C/D/F 维度评分 ==========
+    pred_col_map = {h: i for i, h in enumerate(pred_headers)}
+    gt_col_map = {h: i for i, h in enumerate(gt_headers)}
+
+    # ========== B. Schema正确性 (10分) ==========
+    b_items = {}
+    # B1: ���数 (5分)
+    if len(pred_headers) == EXPECTED_COL_COUNT:
+        b_items["B1_col_count"] = 5
+    elif abs(len(pred_headers) - EXPECTED_COL_COUNT) <= 2:
+        b_items["B1_col_count"] = 3
+    else:
+        b_items["B1_col_count"] = 0
+
+    # B2: 列名匹配 (5分)
+    gt_col_set = set(gt_headers)
+    pred_col_set = set(pred_headers)
+    name_match_rate = len(gt_col_set & pred_col_set) / max(len(gt_col_set), 1)
+    if name_match_rate >= 0.95:
+        b_items["B2_col_names"] = 5
+    elif name_match_rate >= 0.8:
+        b_items["B2_col_names"] = 3
+    else:
+        b_items["B2_col_names"] = 0
+
+    b_score = sum(b_items.values())
+    result["details"]["B_schema"] = {
+        "score": b_score, "max": 10,
+        "detail": {"col_count": len(pred_headers), "name_match_rate": round(name_match_rate, 4), "items": b_items},
+    }
+
+    # ========== C. 行一致性 (15分) ==========
+    c_items = {}
+    gt_row_count = len(gt_rows)
+    pred_row_count = len(pred_rows)
+
+    # C1: 行数比例 (7分)
+    if gt_row_count > 0:
+        ratio = pred_row_count / gt_row_count
+        if 0.95 <= ratio <= 1.05:
+            c_items["C1_row_ratio"] = 7
+        elif 0.7 <= ratio <= 1.3:
+            c_items["C1_row_ratio"] = 4
+        else:
+            c_items["C1_row_ratio"] = 0
+    else:
+        c_items["C1_row_ratio"] = 7 if pred_row_count == 0 else 0
+
+    # C2: key覆盖率 (8分)
+    key_cols_avail = [k for k in KEY_COLUMNS if k in gt_col_map and k in pred_col_map]
+    if key_cols_avail and gt_row_count > 0:
+        gt_keys = set()
+        for row in gt_rows:
+            key = tuple(row[gt_col_map[k]] if gt_col_map[k] < len(row) else "" for k in key_cols_avail)
+            gt_keys.add(key)
+        pred_keys = set()
+        for row in pred_rows:
+            key = tuple(row[pred_col_map[k]] if pred_col_map[k] < len(row) else "" for k in key_cols_avail)
+            pred_keys.add(key)
+        coverage = len(gt_keys & pred_keys) / max(len(gt_keys), 1)
+        if coverage >= 0.995:
+            c_items["C2_key_coverage"] = 8
+        elif coverage >= 0.9:
+            c_items["C2_key_coverage"] = 6
+        elif coverage >= 0.7:
+            c_items["C2_key_coverage"] = 3
+        else:
+            c_items["C2_key_coverage"] = round(8 * coverage, 2)
+    else:
+        c_items["C2_key_coverage"] = 0
+
+    c_score = sum(v for v in c_items.values())
+    result["details"]["C_row_alignment"] = {"score": c_score, "max": 15, "detail": c_items}
+
+    # 构建索引
+    if key_cols_avail:
+        pred_index = {}
+        for row in pred_rows:
+            key = tuple(row[pred_col_map[k]] if pred_col_map[k] < len(row) else "" for k in key_cols_avail)
+            pred_index[key] = row
+        gt_index = {}
+        for row in gt_rows:
+            key = tuple(row[gt_col_map[k]] if gt_col_map[k] < len(row) else "" for k in key_cols_avail)
+            gt_index[key] = row
+    else:
+        pred_index = {}
+        gt_index = {}
+
+    # ========== D. 字段值匹配 (25分) ==========
+    value_cols = [c for c in gt_headers if c in pred_col_map and c not in KEY_COLUMNS]
+    if not value_cols:
+        value_cols = [c for c in gt_headers if c in pred_col_map]
+
+    d_val_items = {}
+    per_col_weight = 25.0 / max(len(value_cols), 1)
+    d_val_score = 0
+    for col in value_cols:
+        gt_ci = gt_col_map.get(col)
+        pred_ci = pred_col_map.get(col)
+        if gt_ci is None or pred_ci is None:
+            d_val_items[col] = {"pass_rate": 0.0, "score": 0, "reason": "column_missing"}
+            continue
+        matches = 0
+        total = 0
+        for key, gt_row in gt_index.items():
+            pred_row = pred_index.get(key)
+            if pred_row is None:
+                total += 1
+                continue
+            gt_val = gt_row[gt_ci].strip() if gt_ci < len(gt_row) else ""
+            pred_val = pred_row[pred_ci].strip() if pred_ci < len(pred_row) else ""
+            if values_match(pred_val, gt_val):
+                matches += 1
+            total += 1
+        rate = matches / max(total, 1)
+        col_score = rate * per_col_weight
+        d_val_score += col_score
+        d_val_items[col] = {"pass_rate": round(rate, 4), "score": round(col_score, 2)}
+
+    result["details"]["D_field_value_match"] = {
+        "score": round(d_val_score, 2), "max": 25, "detail": d_val_items,
+    }
+
+    # ========== D. 字段完整性 (15分) ==========
+    d_comp_items = {}
+    d_comp_score = 0
+    per_field_weight = 15.0 / max(len(KEY_FIELDS), 1)
+    for field in KEY_FIELDS:
+        gt_ci = gt_col_map.get(field)
+        pred_ci = pred_col_map.get(field)
+        if gt_ci is None or pred_ci is None:
+            d_comp_items[field] = {"score": 0, "reason": "column_missing"}
+            continue
+        gt_nonempty = 0
+        pred_nonempty = 0
+        total = 0
+        for key, gt_row in gt_index.items():
+            gt_val = gt_row[gt_ci].strip() if gt_ci < len(gt_row) else ""
+            if gt_val:
+                total += 1
+                pred_row = pred_index.get(key)
+                if pred_row is not None:
+                    pred_val = pred_row[pred_ci].strip() if pred_ci < len(pred_row) else ""
+                    if pred_val:
+                        pred_nonempty += 1
+        gt_nonempty = total
+        rate = pred_nonempty / max(gt_nonempty, 1)
+        field_score = rate * per_field_weight
+        d_comp_score += field_score
+        d_comp_items[field] = {"nonempty_rate": round(rate, 4), "score": round(field_score, 2)}
+
+    result["details"]["D_field_completeness"] = {
+        "score": round(d_comp_score, 2), "max": 15, "detail": d_comp_items,
+    }
+
+    # ========== F. INSERT OVERWRITE (5分) ==========
+    result_sql_path = os.path.join(workspace_path, "result.sql")
+    sql_text = ""
+    try:
+        with open(result_sql_path, "r", encoding="utf-8") as f:
+            sql_text = f.read().lower()
+    except Exception:
+        sql_text = ""
+
+    f_insert_items = {}
+    has_overwrite = "insert overwrite" in sql_text
+    has_partition = "partition" in sql_text
+    if has_overwrite and has_partition:
+        f_insert_items["insert_overwrite_partition"] = 5
+    elif has_overwrite:
+        f_insert_items["insert_overwrite_partition"] = 3
+    else:
+        f_insert_items["insert_overwrite_partition"] = 0
+    result["details"]["F_insert_overwrite"] = {
+        "score": f_insert_items["insert_overwrite_partition"], "max": 5, "detail": f_insert_items,
+    }
+
+    # ========== F. 分区值 (5分) ==========
+    f_part_items = {}
+    if "dt" in sql_text and "20260507" in sql_text:
+        f_part_items["partition_value"] = 5
+    else:
+        f_part_items["partition_value"] = 0
+    result["details"]["F_partition_value"] = {
+        "score": f_part_items["partition_value"], "max": 5, "detail": f_part_items,
+    }
+
+    # ========== F. dt 派生列 (10分) ==========
+    f_dt_items = {}
+    dt_col_exists = "dt" in pred_col_map
+    dt_values_correct = False
+    if dt_col_exists:
+        dt_ci = pred_col_map["dt"]
+        correct_count = 0
+        total_count = 0
+        for row in pred_rows:
+            if dt_ci < len(row):
+                total_count += 1
+                if row[dt_ci].strip() == "20260507":
+                    correct_count += 1
+        dt_values_correct = total_count > 0 and correct_count == total_count
+
+    if dt_col_exists and dt_values_correct:
+        f_dt_items["dt_column"] = 10
+    elif dt_col_exists:
+        f_dt_items["dt_column"] = 5
+    else:
+        f_dt_items["dt_column"] = 0
+    result["details"]["F_dt_column"] = {
+        "score": f_dt_items["dt_column"], "max": 10, "detail": f_dt_items,
+    }
+
+    # 汇总
+    total = (15 + b_score + c_score + d_val_score + d_comp_score
+             + f_insert_items["insert_overwrite_partition"]
+             + f_part_items["partition_value"]
+             + f_dt_items["dt_column"])
+
+    # ========== G~I 过程评分 ==========
+    TRANSCRIPT_PATH = "/tmp/dataclaw_chat.jsonl"
+    OUTPUT_TABLE_SHORT = OUTPUT_TABLE.split(".")[-1]
+    INPUT_TABLE_SHORT = SOURCE_TABLES[0]
+
+    transcript_entries = []
+    has_transcript = False
+    try:
+        if os.path.exists(TRANSCRIPT_PATH):
+            with open(TRANSCRIPT_PATH, "r", encoding="utf-8", errors="ignore") as f:
+                for line in f:
+                    line = line.strip()
+                    if line:
+                        try:
+                            transcript_entries.append(json.loads(line))
+                        except json.JSONDecodeError:
+                            continue
+            if len(transcript_entries) > 2:
+                has_transcript = True
+    except Exception:
+        pass
+
+    if not has_transcript:
+        result["details"]["G_exploration"] = {"score": 0, "max": 35, "items": {"no_transcript": True}}
+        result["details"]["H_efficiency"] = {"score": 0, "max": 40, "items": {"no_transcript": True}}
+        result["details"]["I_self_verification"] = {"score": 0, "max": 25, "items": {"no_transcript": True}}
+        return finalize(result)
+
+    # Parse transcript into structured events
+    tool_uses = []
+    first_write_result_idx = None
+    last_spark_submit_success_idx = None
+    write_result_count = 0
+    logic_error_retries = 0
+
+    for idx, entry in enumerate(transcript_entries):
+        content = entry.get("content", [])
+        if isinstance(content, str):
+            content = [content]
+
+        for block_str in content:
+            if not isinstance(block_str, str):
+                continue
+            if "ToolUseBlock" in block_str:
+                name_match = re.search(r"name='([^']+)'", block_str)
+                input_match = re.search(r"input=(\{.*\})", block_str)
+                if name_match:
+                    tool_name = name_match.group(1)
+                    tool_input = input_match.group(1) if input_match else ""
+                    tool_uses.append((idx, tool_name, tool_input))
+                    if tool_name == "Write" and "result.sql" in tool_input:
+                        write_result_count += 1
+                        if first_write_result_idx is None:
+                            first_write_result_idx = idx
+            if "ToolResultBlock" in block_str:
+                if "Traceback" in block_str or "Exception" in block_str:
+                    env_errors = ["Derby", "metastore", "HiveMetaStore", "Connection refused",
+                                  "db.lck", "TTransportException", "port 10000"]
+                    is_env_error = any(e in block_str for e in env_errors)
+                    has_spark_submit = any(t[1] == "Bash" and "spark-submit" in t[2] and "result.sql" in t[2]
+                                          for t in tool_uses)
+                    if not is_env_error and has_spark_submit:
+                        logic_error_retries += 1
+                if ("spark-submit" in block_str or "spark-sql" in block_str) and "result.sql" in block_str:
+                    if "Exit Code: 0" in block_str and "Traceback" not in block_str:
+                        last_spark_submit_success_idx = idx
+
+    before_first_write = first_write_result_idx if first_write_result_idx is not None else len(transcript_entries)
+
+    # ===== G. 探索充分性 (35分) =====
+    g_items = {}
+    g1_pass = any(name == "Read" and "schema" in inp.lower()
+                  for idx, name, inp in tool_uses if idx < before_first_write)
+    g_items["G1_source_schema"] = 9 if g1_pass else 0
+
+    g2_pass = any(name == "Bash" and INPUT_TABLE_SHORT in inp
+                  and "SELECT" in inp.upper() and "LIMIT" in inp.upper()
+                  for idx, name, inp in tool_uses if idx < before_first_write)
+    g_items["G2_source_sample"] = 9 if g2_pass else 0
+
+    g3_pass = any(name == "Bash" and INPUT_TABLE_SHORT in inp
+                  and ("GROUP BY" in inp.upper() or "DISTINCT" in inp.upper() or "COUNT" in inp.upper())
+                  for idx, name, inp in tool_uses if idx < before_first_write)
+    g_items["G3_distribution"] = 9 if g3_pass else 0
+
+    g4_pass = any(name == "Bash" and ("DESCRIBE" in inp.upper() or "SHOW CREATE" in inp.upper())
+                  and OUTPUT_TABLE_SHORT in inp
+                  for idx, name, inp in tool_uses if idx < before_first_write)
+    g_items["G4_target_schema"] = 8 if g4_pass else 0
+
+    g_score = sum(g_items.values())
+    result["details"]["G_exploration"] = {"score": g_score, "max": 35, "items": g_items}
+
+    # ===== H. 执行效率 (40分) =====
+    h_items = {}
+    if write_result_count <= 2:
+        h_items["H1_few_submissions"] = 20
+    elif write_result_count <= 4:
+        h_items["H1_few_submissions"] = 13
+    elif write_result_count <= 6:
+        h_items["H1_few_submissions"] = 7
+    else:
+        h_items["H1_few_submissions"] = 0
+
+    if logic_error_retries == 0:
+        h_items["H2_no_logic_errors"] = 13
+    elif logic_error_retries <= 1:
+        h_items["H2_no_logic_errors"] = 7
+    else:
+        h_items["H2_no_logic_errors"] = 0
+
+    h_items["H3_no_redundancy"] = 7
+    h_score = sum(h_items.values())
+    result["details"]["H_efficiency"] = {"score": min(h_score, 40), "max": 40, "items": h_items}
+
+    # ===== I. 自验证行为 (25分) =====
+    i_items = {}
+    post_submit_uses = []
+    if last_spark_submit_success_idx is not None:
+        post_submit_uses = [(idx, name, inp) for idx, name, inp in tool_uses
+                            if idx > last_spark_submit_success_idx]
+
+    i1_pass = any(name == "Bash" and OUTPUT_TABLE_SHORT in inp and "SELECT" in inp.upper()
+                  for _, name, inp in post_submit_uses)
+    i_items["I1_query_output"] = 8 if i1_pass else 0
+
+    i2_pass = any(name == "Bash" and OUTPUT_TABLE_SHORT in inp
+                  and ("COUNT" in inp.upper() or "GROUP BY" in inp.upper())
+                  for _, name, inp in post_submit_uses)
+    i_items["I2_check_count"] = 9 if i2_pass else 0
+
+    i3_pass = any(name == "Bash" and OUTPUT_TABLE_SHORT in inp
+                  and ("LIMIT" in inp.upper() or "SELECT *" in inp.upper())
+                  for _, name, inp in post_submit_uses)
+    i_items["I3_check_values"] = 8 if i3_pass else 0
+
+    i_score = sum(i_items.values())
+    result["details"]["I_self_verification"] = {"score": i_score, "max": 25, "items": i_items}
+
+    return finalize(result)

tasks/offline-compute/HiveSQL/hivesql_004/gt/grade_spec.csv

@@ -0,0 +1,19 @@
+维度,维度全名,子维度,满分,说明
+A,A_executability,executability,15,result.sql 能跑通且产出非空
+B,B_schema,schema,10,21列(5) + 列名匹配(5)
+C,C_row_alignment,row_consistency,15,行数比例(7) + key覆盖率(8)
+D,D_field_value_match,field_value_match,25,非key字段逐列值匹配率
+D,D_field_completeness,field_completeness,15,关键字段非空/非空串比例
+F,F_insert_overwrite,insert_overwrite,5,INSERT OVERWRITE + PARTITION
+F,F_partition_value,partition_value,5,dt 分区值 = 20260507
+F,F_dt_column,dt_column,10,dt 派生列存在于输出且值正确
+
+总计,,,100,
+
+# 评分模板,评分范式: 产物100分 = A(15) + B(10) + C(15) + D(40) + F(20)
+# 难度,EASY
+# 权重,"product=0.5, process=0.5"
+# 范式,new
+# Key列,"product_id, obs_product_id, plan_product_id, id, dt"
+# 预期列数,21
+# 输出表,internal_platform_db.app_group_product_info_history_cand_query_engine_011

tasks/offline-compute/HiveSQL/hivesql_004/gt/ground_truth.sql

@@ -0,0 +1,23 @@
+insert overwrite TABLE internal_platform_db.app_group_product_info_history_query_engine_011 PARTITION (dt = '20260507')
+SELECT
+username,
+application_group,
+application_group_owner,
+product_id,
+product_name,
+product_owner,
+obs_product_id,
+obs_product_name,
+obs_product_owner,
+department,
+plan_product_id,
+plan_product_name,
+plan_product_owner,
+application_group_numbers,
+application_group_owners,
+product_owners,
+obs_product_owners,
+plan_product_owners,
+bg,
+id
+FROM internal_platform_db.app_group_product_info_query_engine_011;

tasks/offline-compute/HiveSQL/hivesql_004/init/init_db.py

@@ -0,0 +1,53 @@
+#!/usr/bin/env python3
+import re
+from pyspark.sql import SparkSession
+
+spark = SparkSession.builder \
+    .appName('hivesql_bench_init') \
+    .enableHiveSupport() \
+    .config('spark.sql.warehouse.dir', '/tmp/hive_warehouse') \
+    .getOrCreate()
+
+spark.sql('CREATE DATABASE IF NOT EXISTS internal_platform_db')
+
+
+def _execute_sql_file(spark, sql_path):
+    """Read SQL file, remove SuperSQL SET headers, split by semicolons, execute."""
+    with open(sql_path, 'r', encoding='utf-8') as f:
+        content = f.read()
+    content = re.sub(r'^\s*set\s+query_engine\.\S+\n?', '', content, flags=re.IGNORECASE)
+    stmts, cur, in_sq, in_dq, i = [], [], False, False, 0
+    while i < len(content):
+        ch = content[i]
+        if ch == '\\' and i + 1 < len(content):
+            cur.append(ch); cur.append(content[i + 1]); i += 2; continue
+        if ch == '-' and i + 1 < len(content) and content[i + 1] == '-' and not in_sq and not in_dq:
+            while i < len(content) and content[i] != '\n':
+                i += 1
+            cur.append('\n'); continue
+        if ch == "'" and not in_dq:
+            in_sq = not in_sq
+        elif ch == '"' and not in_sq:
+            in_dq = not in_dq
+        if ch == ';' and not in_sq and not in_dq:
+            s = ''.join(cur).strip()
+            if s:
+                stmts.append(s)
+            cur = []
+        else:
+            cur.append(ch)
+        i += 1
+    last = ''.join(cur).strip()
+    if last:
+        stmts.append(last)
+    for stmt in stmts:
+        spark.sql(stmt)
+
+
+_execute_sql_file(spark, '/tmp_workspace/init_db.sql')
+
+tables = spark.sql('SHOW TABLES IN internal_platform_db').collect()
+print(f'Init complete, {len(tables)} tables created')
+for t in tables:
+    print(f'  - {t.namespace}.{t.tableName}')
+spark.stop()

tasks/offline-compute/HiveSQL/hivesql_004/init/init_db.sql

@@ -0,0 +1,78 @@
+CREATE DATABASE IF NOT EXISTS internal_platform_db;
+
+CREATE TABLE IF NOT EXISTS internal_platform_db.app_group_product_info_query_engine_011 (
+    username STRING,
+    application_group STRING,
+    application_group_owner INT,
+    product_id INT,
+    product_name STRING,
+    product_owner INT,
+    obs_product_id INT,
+    obs_product_name STRING,
+    obs_product_owner INT,
+    department STRING,
+    plan_product_id INT,
+    plan_product_name STRING,
+    plan_product_owner INT,
+    application_group_numbers STRING,
+    application_group_owners STRING,
+    product_owners STRING,
+    obs_product_owners STRING,
+    plan_product_owners STRING,
+    bg STRING,
+    id INT
+) STORED AS ORC;
+
+INSERT INTO TABLE internal_platform_db.app_group_product_info_query_engine_011 VALUES
+('user_a', 'group_alpha', 1, 101, 'prod_x', 201, 301, 'obs_prod_a', 401, 'dept_eng', 501, 'plan_a', 601, '5', 'owner1', 'powner1', 'obs_owner1', 'plan_owner1', 'BG1', 1),
+('user_b', 'group_beta', 2, 102, 'prod_y', 202, 302, 'obs_prod_b', 402, 'dept_sales', 502, 'plan_b', 602, '10', 'owner2', 'powner2', 'obs_owner2', 'plan_owner2', 'BG2', 2),
+('user_c', 'group_gamma', 3, 103, 'prod_z', 203, 303, 'obs_prod_c', 403, 'dept_hr', 503, 'plan_c', 603, '3', 'owner3', 'powner3', 'obs_owner3', 'plan_owner3', 'BG1', 3),
+('user_d', 'group_alpha', 1, 104, 'prod_w', 204, 304, 'obs_prod_d', 404, 'dept_eng', 504, 'plan_d', 604, '7', 'owner4', 'powner4', 'obs_owner4', 'plan_owner4', 'BG3', 4),
+('user_e', 'group_delta', 4, 105, 'prod_v', 205, 305, 'obs_prod_e', 405, 'dept_finance', 505, 'plan_e', 605, '2', 'owner5', 'powner5', 'obs_owner5', 'plan_owner5', 'BG2', 5);
+
+CREATE TABLE IF NOT EXISTS internal_platform_db.app_group_product_info_history_query_engine_011 (
+    username STRING,
+    application_group STRING,
+    application_group_owner INT,
+    product_id INT,
+    product_name STRING,
+    product_owner INT,
+    obs_product_id INT,
+    obs_product_name STRING,
+    obs_product_owner INT,
+    department STRING,
+    plan_product_id INT,
+    plan_product_name STRING,
+    plan_product_owner INT,
+    application_group_numbers STRING,
+    application_group_owners STRING,
+    product_owners STRING,
+    obs_product_owners STRING,
+    plan_product_owners STRING,
+    bg STRING,
+    id INT
+) PARTITIONED BY (dt STRING) STORED AS ORC;
+
+-- Agent 候选目标表
+CREATE TABLE IF NOT EXISTS internal_platform_db.app_group_product_info_history_cand_query_engine_011 (
+    username STRING,
+    application_group STRING,
+    application_group_owner INT,
+    product_id INT,
+    product_name STRING,
+    product_owner INT,
+    obs_product_id INT,
+    obs_product_name STRING,
+    obs_product_owner INT,
+    department STRING,
+    plan_product_id INT,
+    plan_product_name STRING,
+    plan_product_owner INT,
+    application_group_numbers STRING,
+    application_group_owners STRING,
+    product_owners STRING,
+    obs_product_owners STRING,
+    plan_product_owners STRING,
+    bg STRING,
+    id INT
+) PARTITIONED BY (dt STRING) STORED AS ORC;

tasks/offline-compute/HiveSQL/hivesql_004/init/schema.sql

@@ -0,0 +1,22 @@
+CREATE TABLE IF NOT EXISTS internal_platform_db.app_group_product_info (
+    `username` STRING,
+    `application_group` STRING,
+    `application_group_owner` INT,
+    `product_id` INT,
+    `product_name` STRING,
+    `product_owner` INT,
+    `obs_product_id` INT,
+    `obs_product_name` STRING,
+    `obs_product_owner` INT,
+    `department` STRING,
+    `plan_product_id` INT,
+    `plan_product_name` STRING,
+    `plan_product_owner` INT,
+    `application_group_numbers` STRING,
+    `application_group_owners` STRING,
+    `product_owners` STRING,
+    `obs_product_owners` STRING,
+    `plan_product_owners` STRING,
+    `bg` STRING,
+    `id` INT
+) STORED AS ORC;

tasks/offline-compute/HiveSQL/hivesql_004/task.md

@@ -0,0 +1,31 @@
+---
+id: offline-compute_HiveSQL_hivesql_004
+name: 将源表 internal_platform_db.app_group_product_info_sup
+category: offline-compute/HiveSQL
+timeout_seconds: 600
+modality: pure-text
+engine: hivesql
+---
+## Prompt
+任务目标：将应用组与产品归属关系按天做历史快照，按分区全量覆盖写入目标表。
+
+输入：internal_platform_db.app_group_product_info_query_engine_011
+- 字段：username、application_group、application_group_owner、product_id、product_name、product_owner、obs_product_id、obs_product_name、obs_product_owner、department、plan_product_id、plan_product_name、plan_product_owner、application_group_numbers、application_group_owners、product_owners、obs_product_owners、plan_product_owners、bg、id
+
+处理规则：
+1. 无 Join，单表处理
+2. 无过滤条件，全量读取源表所有记录
+3. 无聚合操作
+4. 追加派生列 dt（STRING 类型），填充当天日期字符串（格式 YYYYMMDD，如 '20260507'），置于输出字段最前
+
+输出要求：
+- 输出字段顺序：dt、username、application_group、application_group_owner、product_id、product_name、product_owner、obs_product_id、obs_product_name、obs_product_owner、department、plan_product_id、plan_product_name、plan_product_owner、application_group_numbers、application_group_owners、product_owners、obs_product_owners、plan_product_owners、bg、id
+- 无需去重
+
+写入要求：
+- 目标表：internal_platform_db.app_group_product_info_history_cand_query_engine_011
+- 分区字段：dt（STRING）
+- 写入方式：INSERT OVERWRITE 按分区写入
+- 若目标表不存在，先按 Hive 标准建表（ORC 存储、按 dt 分区），再写入
+
+请将最终 HiveSQL 写入 result.sql 并执行。

tasks/offline-compute/HiveSQL/hivesql_005/gt/expected.csv

@@ -0,0 +1,3 @@
+business_id,business_name,topic,cluster_set,mq_full_topic,system_belong,app_group,bid_incharge,bid_description,bid_create_time,bid_modify_time,cluster_id,cluster_type,cluster_name,bg,category_name,total_produce_pkg_d,production_days_last_7d,total_produce_pkg_last_7d,production_days_last_30d,total_produce_pkg_last_30d,production_days_last_90d,total_produce_pkg_last_90d,is_consumed,has_consumption_days_last_7d,has_consumption_days_last_30d,has_consumption_days_last_90d,hitted_gov_items,governance_benefit_estimate,dt
+bid_001,account_alpha,topic_x,cluster_a,persistent://tenant1/ns1/topic_x,数据总线,appgrp1,user_a,desc1,2025-01-01,2025-06-01,c001,dedicated,cluster_alpha,DEPT_I,product_a,5000,7,35000,30,150000,90,450000,1,7,30,90,,,20260507
+bid_002,account_beta,topic_y,cluster_b,topic_y,流处理平台,appgrp2,user_b,desc2,2025-02-15,2025-05-20,c002,shared,cluster_beta,DEPT_C,product_b,1000,5,7000,20,30000,60,90000,0,3,10,30,,,20260507

tasks/offline-compute/HiveSQL/hivesql_005/gt/grade.py

@@ -0,0 +1,700 @@
+"""hivesql_013 精细评分脚本
+
+业务场景：汇总当天有生产量的 消息队列MQ topic 维度信息，派生 mq_full_topic 等字段，落地为 topic 治理项明细表
+难度: EASY | 特征: INSERT_OVERWRITE|PARTITION|SINGLE_TABLE|FIELD_MAPPING|DERIVED_COLUMN
+
+评分范式: 产物100分 = A(15) + B(10) + C(15) + D(40) + F(20)
+  A_executability                (15分): result.sql 能跑通且产出非空
+  B_schema                       (10分): 30列(5) + 列名匹配(5)
+  C_row_alignment              (15分): 行数比例(7) + key覆盖率(8)
+  D_field_mapping                (25分): 字段别名映射正确性（app_group/bid_*等）
+  D_derived_mq_full_topic    (15分): mq_full_topic 派生列拼接逻辑正确性
+  F_insert_overwrite              (5分):  INSERT OVERWRITE + PARTITION
+  F_partition_value               (5分):  dt 分区值 = 20260507
+  F_filter_condition             (10分):  WHERE dt='20260507' AND total_produce_pkg_last_90d>0
+
+权重: EASY -> product=0.5, process=0.5
+"""
+import os
+import re
+import subprocess
+import tempfile
+import json
+import math
+
+
+def grade(workspace_path, **kwargs):
+
+    # ========== Case 配置 ==========
+    OUTPUT_TABLE = "internal_platform_db.ads_mq_topic_governance_item_d_cand_query_engine_013"
+    DIFFICULTY = "EASY"
+    SOURCE_TABLES = ["dws_mq_production_feature_d_increase_query_engine_013", "ads_mq_topic_governance_item_d_query_engine_013"]
+    KEY_COLUMNS = ["business_id", "bid_incharge", "bid_description", "bid_create_time", "bid_modify_time", "cluster_id", "dt"]
+    EXPECTED_COL_COUNT = 30
+    FIELD_MAPPINGS = {
+        "dw_appgroup": "app_group",
+        "in_charge": "bid_incharge",
+        "description": "bid_description",
+        "create_time": "bid_create_time",
+        "modify_time": "bid_modify_time",
+    }
+    GT_TABLE = OUTPUT_TABLE.replace("_cand_", "_")
+
+    DIFFICULTY_WEIGHTS = {
+        "EASY": (0.5, 0.5),
+        "MEDIUM": (0.6, 0.4),
+        "HARD": (0.7, 0.3),
+        "EXPERT": (0.8, 0.2),
+    }
+
+    _SPARK_SUBMIT_TIMEOUT = 300
+    _JSON_START = "__GRADE_JSON_START__"
+    _JSON_END = "__GRADE_JSON_END__"
+
+    result = {
+        "overall_score": 0.0,
+        "total_points": 0,
+        "grade": "",
+        "details": {},
+        "diagnostics": [],
+    }
+
+    # ========== 内部辅助函数 ==========
+
+    def values_match(pred_val, gt_val, abs_tol=1e-6, rel_tol=1e-4):
+        if pred_val is None and gt_val is None:
+            return True
+        if pred_val is None or gt_val is None:
+            return False
+        s_pred = str(pred_val).strip()
+        s_gt = str(gt_val).strip()
+        if s_pred == s_gt:
+            return True
+        try:
+            pv = float(s_pred)
+            gv = float(s_gt)
+            if math.isnan(pv) and math.isnan(gv):
+                return True
+            if math.isnan(pv) or math.isnan(gv):
+                return False
+            if abs(gv) < abs_tol:
+                return abs(pv - gv) <= abs_tol
+            return abs(pv - gv) <= abs_tol or abs(pv - gv) / max(abs(gv), 1e-12) <= rel_tol
+        except (ValueError, TypeError):
+            pass
+        return s_pred.lower() == s_gt.lower()
+
+    def _run_spark_script(script_code, timeout=_SPARK_SUBMIT_TIMEOUT):
+        with tempfile.NamedTemporaryFile(mode="w", suffix=".py", delete=False, encoding="utf-8") as f:
+            f.write(script_code)
+            script_path = f.name
+        try:
+            r = subprocess.run(
+                ["spark-submit", script_path],
+                capture_output=True, text=True, timeout=timeout,
+            )
+            stdout = r.stdout or ""
+            if _JSON_START in stdout and _JSON_END in stdout:
+                json_str = stdout.split(_JSON_START)[1].split(_JSON_END)[0].strip()
+                return json.loads(json_str), None
+            else:
+                if r.returncode == 0:
+                    for line in stdout.splitlines():
+                        if line.strip().startswith("Traceback"):
+                            return None, f"spark-submit error: {line}"
+                    return None, "spark-submit 无 JSON 输出"
+                err_msg = (r.stderr or "")[-500:]
+                return None, f"spark-submit failed: {err_msg}"
+        except subprocess.TimeoutExpired:
+            return None, f"spark-submit 超时 ({timeout}s)"
+        except Exception as e:
+            return None, f"spark-submit 异常: {e}"
+        finally:
+            try:
+                os.unlink(script_path)
+            except OSError:
+                pass
+
+    def read_table_via_spark_submit(table_name):
+        """Read table via spark-submit subprocess. Returns (cols, rows_as_lists)."""
+        if not re.match(r'^[a-zA-Z_][a-zA-Z0-9_.]*$', table_name):
+            raise ValueError(f"非法表名: {table_name}")
+        read_script = f'''
+import json
+from pyspark.sql import SparkSession
+spark = SparkSession.builder.appName("grade_read").enableHiveSupport() \\
+    .config("spark.sql.warehouse.dir", "/tmp/hive_warehouse").getOrCreate()
+try:
+    df = spark.sql("SELECT * FROM {table_name}")
+    cols = [c.lower() for c in df.columns]
+    rows = [[str(v) if v is not None else "" for v in row] for row in df.collect()]
+    print("{_JSON_START}")
+    print(json.dumps({{"cols": cols, "rows": rows}}, ensure_ascii=False))
+    print("{_JSON_END}")
+except Exception as e:
+    print("{_JSON_START}")
+    print(json.dumps({{"error": str(e)}}))
+    print("{_JSON_END}")
+finally:
+    spark.stop()
+'''
+        data, err = _run_spark_script(read_script)
+        if err:
+            raise RuntimeError(f"read_table failed: {err}")
+        if "error" in data:
+            raise RuntimeError(f"query failed: {data['error']}")
+        return data["cols"], data["rows"]
+
+    # SQL executor template (self-contained, no external dependency)
+    _SQL_EXEC_TEMPLATE = '''
+import json, re
+from pyspark.sql import SparkSession
+spark = SparkSession.builder.appName("{app_name}").enableHiveSupport() \\
+    .config("spark.sql.warehouse.dir", "/tmp/hive_warehouse").getOrCreate()
+try:
+    with open("{sql_file}", "r", encoding="utf-8") as _f:
+        _sql = _f.read()
+    _sql = re.sub(r"^\\s*set\\s+query_engine\\.\\S+\\n?", "", _sql, flags=re.IGNORECASE)
+    _stmts, _cur, _in_sq, _in_dq, _i = [], [], False, False, 0
+    while _i < len(_sql):
+        _ch = _sql[_i]
+        if _ch == "\\\\" and _i + 1 < len(_sql):
+            _cur.append(_ch); _cur.append(_sql[_i+1]); _i += 2; continue
+        if _ch == "-" and _i+1 < len(_sql) and _sql[_i+1] == "-" and not _in_sq and not _in_dq:
+            while _i < len(_sql) and _sql[_i] != "\\n": _i += 1
+            _cur.append("\\n"); continue
+        if _ch == "'" and not _in_dq: _in_sq = not _in_sq
+        elif _ch == '"' and not _in_sq: _in_dq = not _in_dq
+        if _ch == ";" and not _in_sq and not _in_dq:
+            _s = "".join(_cur).strip()
+            if _s: _stmts.append(_s)
+            _cur = []
+        else:
+            _cur.append(_ch)
+        _i += 1
+    _last = "".join(_cur).strip()
+    if _last: _stmts.append(_last)
+    for _stmt in _stmts:
+        spark.sql(_stmt)
+    print("{_JSON_START}")
+    print(json.dumps({{"ok": True}}))
+    print("{_JSON_END}")
+except Exception as e:
+    print("{_JSON_START}")
+    print(json.dumps({{"ok": False, "error": str(e)}}))
+    print("{_JSON_END}")
+finally:
+    spark.stop()
+'''
+
+    def execute_result_sql():
+        result_sql = os.path.join(workspace_path, "result.sql")
+        if not os.path.exists(result_sql):
+            return False, "no_result_file"
+        # 替换 数据平台WD时间变量（沙箱 spark-sql 不支持 ${...} 语法）
+        with open(result_sql, 'r', encoding='utf-8') as _rf:
+            _sql_text = _rf.read()
+        _bizdate = '20260507'
+        _sql_text = re.sub(r'\${bdp\.system\.bizdate(?:[+-]\d+)?}', _bizdate, _sql_text)
+        _sql_text = re.sub(r'\${yyyymmdd(?:[+-]\d+)?}', _bizdate, _sql_text)
+        _sql_text = re.sub(r'\${bizdate(?:[+-]\d+)?}', _bizdate, _sql_text)
+        _sql_text = re.sub(r'\${[^}]*date[^}]*}', _bizdate, _sql_text)
+        with open(result_sql, 'w', encoding='utf-8') as _wf:
+            _wf.write(_sql_text)
+        script = _SQL_EXEC_TEMPLATE.format(app_name="grade_exec", sql_file=result_sql,
+                                            _JSON_START=_JSON_START, _JSON_END=_JSON_END)
+        data, err = _run_spark_script(script)
+        if err:
+            return False, f"execution_error: {err}"
+        if data and data.get("ok"):
+            return True, None
+        return False, f"execution_error: {data.get('error', 'unknown') if data else 'no output'}"
+
+    def execute_ground_truth_sql():
+        gt_sql = os.path.join(workspace_path, "gt", "ground_truth.sql")
+        if not os.path.exists(gt_sql):
+            return False, "ground_truth.sql not found"
+        script = _SQL_EXEC_TEMPLATE.format(app_name="grade_gt", sql_file=gt_sql,
+                                            _JSON_START=_JSON_START, _JSON_END=_JSON_END)
+        data, err = _run_spark_script(script)
+        if err:
+            return False, f"gt_execution_error: {err}"
+        if data and data.get("ok"):
+            return True, None
+        return False, f"gt_execution_error: {data.get('error', 'unknown') if data else 'no output'}"
+
+    def truncate_table(table_name):
+        script = f'''
+from pyspark.sql import SparkSession
+spark = SparkSession.builder.appName("truncate").enableHiveSupport() \\
+    .config("spark.sql.warehouse.dir", "/tmp/hive_warehouse").getOrCreate()
+spark.sql("TRUNCATE TABLE {table_name}")
+spark.stop()
+'''
+        try:
+            _run_spark_script(script, timeout=120)
+        except Exception:
+            pass
+
+    def restore_hive_site():
+        """Restore hive-site.xml to canonical state (agent may have modified it)."""
+        canonical_hive_site = '''<?xml version="1.0"?>
+<?xml-stylesheet type="text/xsl" href="configuration.xsl"?>
+<configuration>
+  <property>
+    <name>hive.metastore.uris</name>
+    <value>thrift://localhost:9083</value>
+  </property>
+  <property>
+    <name>hive.metastore.warehouse.dir</name>
+    <value>/tmp/hive_warehouse</value>
+  </property>
+  <property>
+    <name>javax.jdo.option.ConnectionURL</name>
+    <value>jdbc:derby:;databaseName=/tmp/hive_metastore_db;create=true</value>
+  </property>
+  <property>
+    <name>javax.jdo.option.ConnectionDriverName</name>
+    <value>org.apache.derby.jdbc.EmbeddedDriver</value>
+  </property>
+  <property>
+    <name>datanucleus.schema.autoCreateAll</name>
+    <value>true</value>
+  </property>
+  <property>
+    <name>hive.metastore.schema.verification</name>
+    <value>false</value>
+  </property>
+</configuration>
+'''
+        hive_site_path = os.path.join(os.environ.get('SPARK_HOME', '/opt/spark'), 'conf', 'hive-site.xml')
+        try:
+            with open(hive_site_path, 'w') as f:
+                f.write(canonical_hive_site)
+        except Exception:
+            pass
+
+    def finalize(result):
+        product_weight, process_weight = DIFFICULTY_WEIGHTS.get(DIFFICULTY, (0.7, 0.3))
+        product_dims = ["A_executability"] + ['B_schema', 'C_row_alignment', 'D_derived_mq_full_topic', 'D_field_mapping', 'F_filter_condition', 'F_insert_overwrite', 'F_partition_value']
+        product_raw = sum(result["details"].get(d, {}).get("score", 0) for d in product_dims)
+        product_ratio = product_raw / 100.0
+        for dim in ["H_efficiency"]:
+            if dim in result["details"]:
+                raw = result["details"][dim].get("score", 0)
+                result["details"][dim]["score_before_scaling"] = raw
+                result["details"][dim]["score"] = round(raw * product_ratio, 2)
+                result["details"][dim]["product_ratio"] = round(product_ratio, 4)
+        process_dims = ["G_exploration", "H_efficiency", "I_self_verification"]
+        process_raw = sum(result["details"].get(d, {}).get("score", 0) for d in process_dims)
+        product_score = round(product_raw * product_weight, 2)
+        process_score = round(process_raw * process_weight, 2)
+        total = round(product_score + process_score, 2)
+        result["total_points"] = total
+        result["product_points"] = product_score
+        result["process_points"] = round(process_score, 2)
+        result["weights"] = {"product": product_weight, "process": process_weight}
+        result["overall_score"] = round(total / 100.0, 4)
+        if total >= 90:
+            result["grade"] = "优秀"
+        elif total >= 75:
+            result["grade"] = "良好"
+        elif total >= 60:
+            result["grade"] = "合格"
+        elif total >= 40:
+            result["grade"] = "偏弱"
+        else:
+            result["grade"] = "不合格"
+        return result
+
+    # Restore hive-site.xml (agent may have modified it)
+    restore_hive_site()
+
+    # ========== A. 可执行性 (15分) ==========
+    a_items = {"A1_exec_ok": 0, "A2_has_data": 0}
+    exec_ok, exec_err = execute_result_sql()
+    if not exec_ok:
+        detail = "未产出 result.sql" if exec_err == "no_result_file" else str(exec_err)[:200]
+        result["details"]["A_executability"] = {"score": 0, "max": 15, "detail": detail}
+        result["error"] = detail
+        for dim in ['B_schema', 'C_row_alignment', 'D_derived_mq_full_topic', 'D_field_mapping', 'F_filter_condition', 'F_insert_overwrite', 'F_partition_value']:
+            result["details"][dim] = {"score": 0, "max": 0, "items": {}}
+        return finalize(result)
+
+    a_items["A1_exec_ok"] = 8
+
+    pred_headers, pred_rows = [], []
+    try:
+        pred_headers, pred_rows = read_table_via_spark_submit(OUTPUT_TABLE)
+    except Exception as e:
+        result["diagnostics"].append(f"read_pred_failed: {e}")
+
+    if not pred_rows:
+        result["details"]["A_executability"] = {"score": 8, "max": 15, "items": a_items}
+        for dim in ['B_schema', 'C_row_alignment', 'D_derived_mq_full_topic', 'D_field_mapping', 'F_filter_condition', 'F_insert_overwrite', 'F_partition_value']:
+            result["details"][dim] = {"score": 0, "max": 0, "items": {}}
+        return finalize(result)
+
+    a_items["A2_has_data"] = 7
+    result["details"]["A_executability"] = {"score": 15, "max": 15, "items": a_items}
+
+    # ========== Execute GT + Read GT ==========
+    truncate_table(GT_TABLE)
+    gt_ok, gt_err = execute_ground_truth_sql()
+    if not gt_ok:
+        result["error"] = f"ground_truth failed: {gt_err}"
+        for dim in ['B_schema', 'C_row_alignment', 'D_derived_mq_full_topic', 'D_field_mapping', 'F_filter_condition', 'F_insert_overwrite', 'F_partition_value']:
+            result["details"][dim] = {"score": 0, "max": 0, "items": {}}
+        return finalize(result)
+
+    gt_headers, gt_rows = [], []
+    try:
+        gt_headers, gt_rows = read_table_via_spark_submit(GT_TABLE)
+    except Exception as e:
+        result["error"] = f"read_gt_failed: {e}"
+        for dim in ['B_schema', 'C_row_alignment', 'D_derived_mq_full_topic', 'D_field_mapping', 'F_filter_condition', 'F_insert_overwrite', 'F_partition_value']:
+            result["details"][dim] = {"score": 0, "max": 0, "items": {}}
+        return finalize(result)
+
+    # ========== B/C/D/F 维度评分 ==========
+    pred_col_map = {h: i for i, h in enumerate(pred_headers)}
+    gt_col_map = {h: i for i, h in enumerate(gt_headers)}
+
+    # ========== B. Schema正确性 (10分) ==========
+    b_items = {}
+    # B1: 列数 (5分)
+    if len(pred_headers) == EXPECTED_COL_COUNT:
+        b_items["B1_col_count"] = 5
+    elif abs(len(pred_headers) - EXPECTED_COL_COUNT) <= 2:
+        b_items["B1_col_count"] = 3
+    else:
+        b_items["B1_col_count"] = 0
+
+    # B2: 列名匹配 (5分)
+    gt_col_set = set(gt_headers)
+    pred_col_set = set(pred_headers)
+    name_match_rate = len(gt_col_set & pred_col_set) / max(len(gt_col_set), 1)
+    if name_match_rate >= 0.95:
+        b_items["B2_col_names"] = 5
+    elif name_match_rate >= 0.8:
+        b_items["B2_col_names"] = 3
+    else:
+        b_items["B2_col_names"] = 0
+
+    b_score = sum(b_items.values())
+    result["details"]["B_schema"] = {
+        "score": b_score, "max": 10,
+        "detail": {"col_count": len(pred_headers), "name_match_rate": round(name_match_rate, 4), "items": b_items},
+    }
+
+    # ========== C. 行一致性 (15分) ==========
+    c_items = {}
+    gt_row_count = len(gt_rows)
+    pred_row_count = len(pred_rows)
+
+    # C1: 行数比例 (7分)
+    if gt_row_count > 0:
+        ratio = pred_row_count / gt_row_count
+        if 0.95 <= ratio <= 1.05:
+            c_items["C1_row_ratio"] = 7
+        elif 0.7 <= ratio <= 1.3:
+            c_items["C1_row_ratio"] = 4
+        else:
+            c_items["C1_row_ratio"] = 0
+    else:
+        c_items["C1_row_ratio"] = 7 if pred_row_count == 0 else 0
+
+    # C2: key覆盖率 (8分)
+    key_cols_avail = [k for k in KEY_COLUMNS if k in gt_col_map and k in pred_col_map]
+    if key_cols_avail and gt_row_count > 0:
+        gt_keys = set()
+        for row in gt_rows:
+            key = tuple(row[gt_col_map[k]] if gt_col_map[k] < len(row) else "" for k in key_cols_avail)
+            gt_keys.add(key)
+        pred_keys = set()
+        for row in pred_rows:
+            key = tuple(row[pred_col_map[k]] if pred_col_map[k] < len(row) else "" for k in key_cols_avail)
+            pred_keys.add(key)
+        coverage = len(gt_keys & pred_keys) / max(len(gt_keys), 1)
+        if coverage >= 0.995:
+            c_items["C2_key_coverage"] = 8
+        elif coverage >= 0.9:
+            c_items["C2_key_coverage"] = 6
+        elif coverage >= 0.7:
+            c_items["C2_key_coverage"] = 3
+        else:
+            c_items["C2_key_coverage"] = round(8 * coverage, 2)
+    else:
+        c_items["C2_key_coverage"] = 0
+
+    c_score = sum(v for v in c_items.values())
+    result["details"]["C_row_alignment"] = {"score": c_score, "max": 15, "detail": c_items}
+
+    # 构建索引
+    if key_cols_avail:
+        pred_index = {}
+        for row in pred_rows:
+            key = tuple(row[pred_col_map[k]] if pred_col_map[k] < len(row) else "" for k in key_cols_avail)
+            pred_index[key] = row
+        gt_index = {}
+        for row in gt_rows:
+            key = tuple(row[gt_col_map[k]] if gt_col_map[k] < len(row) else "" for k in key_cols_avail)
+            gt_index[key] = row
+    else:
+        pred_index = {}
+        gt_index = {}
+
+    # ========== D. 字段映射正确性 (25分) ==========
+    d_map_items = {}
+    per_mapping_weight = 25.0 / max(len(FIELD_MAPPINGS), 1)
+    d_map_score = 0
+    for src_col, dst_col in FIELD_MAPPINGS.items():
+        # GT 中该列用 dst_col 名
+        gt_ci = gt_col_map.get(dst_col)
+        pred_ci = pred_col_map.get(dst_col)
+        if gt_ci is None or pred_ci is None:
+            d_map_items[dst_col] = {"pass_rate": 0.0, "score": 0, "reason": "column_missing"}
+            continue
+        matches = 0
+        total = 0
+        for key, gt_row in gt_index.items():
+            pred_row = pred_index.get(key)
+            if pred_row is None:
+                total += 1
+                continue
+            gt_val = gt_row[gt_ci].strip() if gt_ci < len(gt_row) else ""
+            pred_val = pred_row[pred_ci].strip() if pred_ci < len(pred_row) else ""
+            if values_match(pred_val, gt_val):
+                matches += 1
+            total += 1
+        rate = matches / max(total, 1)
+        col_score = rate * per_mapping_weight
+        d_map_score += col_score
+        d_map_items[dst_col] = {"pass_rate": round(rate, 4), "score": round(col_score, 2), "mapped_from": src_col}
+
+    result["details"]["D_field_mapping"] = {
+        "score": round(d_map_score, 2), "max": 25, "detail": d_map_items,
+    }
+
+    # ========== D. mq_full_topic 派生列 (15分) ==========
+    d_mq_items = {}
+    d_mq_score = 0
+    mq_ci = pred_col_map.get("mq_full_topic")
+    gt_mq_ci = gt_col_map.get("mq_full_topic")
+    if mq_ci is None or gt_mq_ci is None:
+        d_mq_items["mq_full_topic"] = {"pass_rate": 0.0, "score": 0, "reason": "column_missing"}
+    else:
+        matches = 0
+        total = 0
+        for key, gt_row in gt_index.items():
+            pred_row = pred_index.get(key)
+            if pred_row is None:
+                total += 1
+                continue
+            gt_val = gt_row[gt_mq_ci].strip() if gt_mq_ci < len(gt_row) else ""
+            pred_val = pred_row[mq_ci].strip() if mq_ci < len(pred_row) else ""
+            if values_match(pred_val, gt_val):
+                matches += 1
+            total += 1
+        rate = matches / max(total, 1)
+        d_mq_score = rate * 15
+        d_mq_items["mq_full_topic"] = {"pass_rate": round(rate, 4), "score": round(d_mq_score, 2)}
+
+    result["details"]["D_derived_mq_full_topic"] = {
+        "score": round(d_mq_score, 2), "max": 15, "detail": d_mq_items,
+    }
+
+    # ========== F. INSERT OVERWRITE (5分) ==========
+    result_sql_path = os.path.join(workspace_path, "result.sql")
+    sql_text = ""
+    try:
+        with open(result_sql_path, "r", encoding="utf-8") as f:
+            sql_text = f.read().lower()
+    except Exception:
+        sql_text = ""
+
+    f_insert_items = {}
+    has_overwrite = "insert overwrite" in sql_text
+    has_partition = "partition" in sql_text
+    if has_overwrite and has_partition:
+        f_insert_items["insert_overwrite_partition"] = 5
+    elif has_overwrite:
+        f_insert_items["insert_overwrite_partition"] = 3
+    else:
+        f_insert_items["insert_overwrite_partition"] = 0
+    result["details"]["F_insert_overwrite"] = {
+        "score": f_insert_items["insert_overwrite_partition"], "max": 5, "detail": f_insert_items,
+    }
+
+    # ========== F. 分区值 (5分) ==========
+    f_part_items = {}
+    if "dt" in sql_text and "20260507" in sql_text:
+        f_part_items["partition_value"] = 5
+    else:
+        f_part_items["partition_value"] = 0
+    result["details"]["F_partition_value"] = {
+        "score": f_part_items["partition_value"], "max": 5, "detail": f_part_items,
+    }
+
+    # ========== F. 过滤条件 (10分) ==========
+    f_filter_items = {}
+    has_dt_filter = "dt" in sql_text and "20260507" in sql_text
+    has_pkg_filter = "total_produce_pkg_last_90d" in sql_text and (">0" in sql_text.replace(" ", "") or "> 0" in sql_text)
+    if has_dt_filter and has_pkg_filter:
+        f_filter_items["filter_condition"] = 10
+    elif has_dt_filter:
+        f_filter_items["filter_condition"] = 5
+    elif has_pkg_filter:
+        f_filter_items["filter_condition"] = 3
+    else:
+        f_filter_items["filter_condition"] = 0
+    result["details"]["F_filter_condition"] = {
+        "score": f_filter_items["filter_condition"], "max": 10, "detail": f_filter_items,
+    }
+
+    # 汇总
+    total = (15 + b_score + c_score + d_map_score + d_mq_score
+             + f_insert_items["insert_overwrite_partition"]
+             + f_part_items["partition_value"]
+             + f_filter_items["filter_condition"])
+
+    # ========== G~I 过程评分 ==========
+    TRANSCRIPT_PATH = "/tmp/dataclaw_chat.jsonl"
+    OUTPUT_TABLE_SHORT = OUTPUT_TABLE.split(".")[-1]
+    INPUT_TABLE_SHORT = SOURCE_TABLES[0]
+
+    transcript_entries = []
+    has_transcript = False
+    try:
+        if os.path.exists(TRANSCRIPT_PATH):
+            with open(TRANSCRIPT_PATH, "r", encoding="utf-8", errors="ignore") as f:
+                for line in f:
+                    line = line.strip()
+                    if line:
+                        try:
+                            transcript_entries.append(json.loads(line))
+                        except json.JSONDecodeError:
+                            continue
+            if len(transcript_entries) > 2:
+                has_transcript = True
+    except Exception:
+        pass
+
+    if not has_transcript:
+        result["details"]["G_exploration"] = {"score": 0, "max": 35, "items": {"no_transcript": True}}
+        result["details"]["H_efficiency"] = {"score": 0, "max": 40, "items": {"no_transcript": True}}
+        result["details"]["I_self_verification"] = {"score": 0, "max": 25, "items": {"no_transcript": True}}
+        return finalize(result)
+
+    # Parse transcript into structured events
+    tool_uses = []
+    first_write_result_idx = None
+    last_spark_submit_success_idx = None
+    write_result_count = 0
+    logic_error_retries = 0
+
+    for idx, entry in enumerate(transcript_entries):
+        content = entry.get("content", [])
+        if isinstance(content, str):
+            content = [content]
+
+        for block_str in content:
+            if not isinstance(block_str, str):
+                continue
+            if "ToolUseBlock" in block_str:
+                name_match = re.search(r"name='([^']+)'", block_str)
+                input_match = re.search(r"input=(\{.*\})", block_str)
+                if name_match:
+                    tool_name = name_match.group(1)
+                    tool_input = input_match.group(1) if input_match else ""
+                    tool_uses.append((idx, tool_name, tool_input))
+                    if tool_name == "Write" and "result.sql" in tool_input:
+                        write_result_count += 1
+                        if first_write_result_idx is None:
+                            first_write_result_idx = idx
+            if "ToolResultBlock" in block_str:
+                if "Traceback" in block_str or "Exception" in block_str:
+                    env_errors = ["Derby", "metastore", "HiveMetaStore", "Connection refused",
+                                  "db.lck", "TTransportException", "port 10000"]
+                    is_env_error = any(e in block_str for e in env_errors)
+                    has_spark_submit = any(t[1] == "Bash" and "spark-submit" in t[2] and "result.sql" in t[2]
+                                          for t in tool_uses)
+                    if not is_env_error and has_spark_submit:
+                        logic_error_retries += 1
+                if ("spark-submit" in block_str or "spark-sql" in block_str) and "result.sql" in block_str:
+                    if "Exit Code: 0" in block_str and "Traceback" not in block_str:
+                        last_spark_submit_success_idx = idx
+
+    before_first_write = first_write_result_idx if first_write_result_idx is not None else len(transcript_entries)
+
+    # ===== G. 探索充分性 (35分) =====
+    g_items = {}
+    g1_pass = any(name == "Read" and "schema" in inp.lower()
+                  for idx, name, inp in tool_uses if idx < before_first_write)
+    g_items["G1_source_schema"] = 9 if g1_pass else 0
+
+    g2_pass = any(name == "Bash" and INPUT_TABLE_SHORT in inp
+                  and "SELECT" in inp.upper() and "LIMIT" in inp.upper()
+                  for idx, name, inp in tool_uses if idx < before_first_write)
+    g_items["G2_source_sample"] = 9 if g2_pass else 0
+
+    g3_pass = any(name == "Bash" and INPUT_TABLE_SHORT in inp
+                  and ("GROUP BY" in inp.upper() or "DISTINCT" in inp.upper() or "COUNT" in inp.upper())
+                  for idx, name, inp in tool_uses if idx < before_first_write)
+    g_items["G3_distribution"] = 9 if g3_pass else 0
+
+    g4_pass = any(name == "Bash" and ("DESCRIBE" in inp.upper() or "SHOW CREATE" in inp.upper())
+                  and OUTPUT_TABLE_SHORT in inp
+                  for idx, name, inp in tool_uses if idx < before_first_write)
+    g_items["G4_target_schema"] = 8 if g4_pass else 0
+
+    g_score = sum(g_items.values())
+    result["details"]["G_exploration"] = {"score": g_score, "max": 35, "items": g_items}
+
+    # ===== H. 执行效率 (40分) =====
+    h_items = {}
+    if write_result_count <= 2:
+        h_items["H1_few_submissions"] = 20
+    elif write_result_count <= 4:
+        h_items["H1_few_submissions"] = 13
+    elif write_result_count <= 6:
+        h_items["H1_few_submissions"] = 7
+    else:
+        h_items["H1_few_submissions"] = 0
+
+    if logic_error_retries == 0:
+        h_items["H2_no_logic_errors"] = 13
+    elif logic_error_retries <= 1:
+        h_items["H2_no_logic_errors"] = 7
+    else:
+        h_items["H2_no_logic_errors"] = 0
+
+    h_items["H3_no_redundancy"] = 7
+    h_score = sum(h_items.values())
+    result["details"]["H_efficiency"] = {"score": min(h_score, 40), "max": 40, "items": h_items}
+
+    # ===== I. 自验证行为 (25分) =====
+    i_items = {}
+    post_submit_uses = []
+    if last_spark_submit_success_idx is not None:
+        post_submit_uses = [(idx, name, inp) for idx, name, inp in tool_uses
+                            if idx > last_spark_submit_success_idx]
+
+    i1_pass = any(name == "Bash" and OUTPUT_TABLE_SHORT in inp and "SELECT" in inp.upper()
+                  for _, name, inp in post_submit_uses)
+    i_items["I1_query_output"] = 8 if i1_pass else 0
+
+    i2_pass = any(name == "Bash" and OUTPUT_TABLE_SHORT in inp
+                  and ("COUNT" in inp.upper() or "GROUP BY" in inp.upper())
+                  for _, name, inp in post_submit_uses)
+    i_items["I2_check_count"] = 9 if i2_pass else 0
+
+    i3_pass = any(name == "Bash" and OUTPUT_TABLE_SHORT in inp
+                  and ("LIMIT" in inp.upper() or "SELECT *" in inp.upper())
+                  for _, name, inp in post_submit_uses)
+    i_items["I3_check_values"] = 8 if i3_pass else 0
+
+    i_score = sum(i_items.values())
+    result["details"]["I_self_verification"] = {"score": i_score, "max": 25, "items": i_items}
+
+    return finalize(result)

tasks/offline-compute/HiveSQL/hivesql_005/gt/grade_spec.csv

@@ -0,0 +1,19 @@
+维度,维度全名,子维度,满分,说明
+A,A_executability,executability,15,result.sql 能跑通且产出非空
+B,B_schema,schema,10,30列(5) + 列名匹配(5)
+C,C_row_alignment,row_consistency,15,行数比例(7) + key覆盖率(8)
+D,D_field_mapping,field_mapping,25,字段别名映射正确性（app_group/bid_*等）
+D,D_derived_mq_full_topic,derived_mq_full_topic,15,mq_full_topic 派生列拼接逻辑正确性
+F,F_insert_overwrite,insert_overwrite,5,INSERT OVERWRITE + PARTITION
+F,F_partition_value,partition_value,5,dt 分区值 = 20260507
+F,F_filter_condition,filter_condition,10,WHERE dt='20260507' AND total_produce_pkg_last_90d>0
+
+总计,,,100,
+
+# 评分模板,评分范式: 产物100分 = A(15) + B(10) + C(15) + D(40) + F(20)
+# 难度,EASY
+# 权重,"product=0.5, process=0.5"
+# 范式,new
+# Key列,"business_id, bid_incharge, bid_description, bid_create_time, bid_modify_time, cluster_id, dt"
+# 预期列数,30
+# 输出表,internal_platform_db.ads_mq_topic_governance_item_d_cand_query_engine_013

tasks/offline-compute/HiveSQL/hivesql_005/gt/ground_truth.sql

@@ -0,0 +1,34 @@
+INSERT OVERWRITE TABLE internal_platform_db.ads_mq_topic_governance_item_d_query_engine_013 PARTITION (dt = '20260507')
+SELECT
+  business_id,
+  business_name,
+  topic,
+  cluster_set,
+  IF ( tenant IS NOT NULL AND namespaces IS NOT NULL, CONCAT( 'persistent://', tenant, '/', namespaces, '/', topic ), topic ) AS mq_full_topic,
+  system_belong,
+  dw_appgroup AS app_group,
+  in_charge AS bid_incharge,
+  description AS bid_description,
+  create_time AS bid_create_time,
+  modify_time AS bid_modify_time,
+  cluster_id,
+  cluster_type,
+  cluster_name,
+  bg,
+  category_name,
+  total_produce_pkg_d,
+  production_days_last_7d,
+  total_produce_pkg_last_7d,
+  production_days_last_30d,
+  total_produce_pkg_last_30d,
+  production_days_last_90d,
+  total_produce_pkg_last_90d,
+  is_consumed,
+  has_consumption_days_last_7d,
+  has_consumption_days_last_30d,
+  has_consumption_days_last_90d,
+  NULL AS hitted_gov_items,
+  NULL AS governance_benefit_estimate
+FROM internal_platform_db.dws_mq_production_feature_d_increase_query_engine_013
+WHERE dt = '20260507'
+AND total_produce_pkg_last_90d > 0

tasks/offline-compute/HiveSQL/hivesql_005/init/init_db.py

@@ -0,0 +1,53 @@
+#!/usr/bin/env python3
+import re
+from pyspark.sql import SparkSession
+
+spark = SparkSession.builder \
+    .appName('hivesql_bench_init') \
+    .enableHiveSupport() \
+    .config('spark.sql.warehouse.dir', '/tmp/hive_warehouse') \
+    .getOrCreate()
+
+spark.sql('CREATE DATABASE IF NOT EXISTS internal_platform_db')
+
+
+def _execute_sql_file(spark, sql_path):
+    """Read SQL file, remove SuperSQL SET headers, split by semicolons, execute."""
+    with open(sql_path, 'r', encoding='utf-8') as f:
+        content = f.read()
+    content = re.sub(r'^\s*set\s+query_engine\.\S+\n?', '', content, flags=re.IGNORECASE)
+    stmts, cur, in_sq, in_dq, i = [], [], False, False, 0
+    while i < len(content):
+        ch = content[i]
+        if ch == '\\' and i + 1 < len(content):
+            cur.append(ch); cur.append(content[i + 1]); i += 2; continue
+        if ch == '-' and i + 1 < len(content) and content[i + 1] == '-' and not in_sq and not in_dq:
+            while i < len(content) and content[i] != '\n':
+                i += 1
+            cur.append('\n'); continue
+        if ch == "'" and not in_dq:
+            in_sq = not in_sq
+        elif ch == '"' and not in_sq:
+            in_dq = not in_dq
+        if ch == ';' and not in_sq and not in_dq:
+            s = ''.join(cur).strip()
+            if s:
+                stmts.append(s)
+            cur = []
+        else:
+            cur.append(ch)
+        i += 1
+    last = ''.join(cur).strip()
+    if last:
+        stmts.append(last)
+    for stmt in stmts:
+        spark.sql(stmt)
+
+
+_execute_sql_file(spark, '/tmp_workspace/init_db.sql')
+
+tables = spark.sql('SHOW TABLES IN internal_platform_db').collect()
+print(f'Init complete, {len(tables)} tables created')
+for t in tables:
+    print(f'  - {t.namespace}.{t.tableName}')
+spark.stop()

tasks/offline-compute/HiveSQL/hivesql_005/init/init_db.sql

@@ -0,0 +1,141 @@
+CREATE DATABASE IF NOT EXISTS internal_platform_db;
+
+-- 输入表: dws_mq_production_feature_d_increase
+CREATE TABLE IF NOT EXISTS internal_platform_db.dws_mq_production_feature_d_increase_query_engine_013 (
+    dt STRING,
+    business_id STRING,
+    business_name STRING,
+    cluster_set STRING,
+    tenant STRING,
+    namespaces STRING,
+    topic STRING,
+    mq_type STRING,
+    system_belong STRING,
+    dw_appgroup STRING,
+    in_charge STRING,
+    description STRING,
+    create_time STRING,
+    modify_time STRING,
+    cluster_id STRING,
+    cluster_type STRING,
+    cluster_name STRING,
+    bg STRING,
+    category_name STRING,
+    is_produce TINYINT,
+    total_produce_pkg_d BIGINT,
+    is_consumed TINYINT,
+    is_consumed_by_reader TINYINT,
+    max_consume_pkg_d BIGINT,
+    min_consume_pkg_d BIGINT,
+    min_consume_pkg_without_reader_d BIGINT,
+    is_backlog TINYINT,
+    is_backlog_without_reader TINYINT,
+    total_backlog_pkg_d BIGINT,
+    total_backlog_pkg_without_reader_d BIGINT,
+    backlog_ratio STRING,
+    backlog_ratio_without_reader STRING,
+    is_produce_all_7d TINYINT,
+    production_days_last_7d TINYINT,
+    total_produce_pkg_last_7d BIGINT,
+    is_produce_all_30d TINYINT,
+    production_days_last_30d TINYINT,
+    total_produce_pkg_last_30d BIGINT,
+    is_produce_all_90d TINYINT,
+    production_days_last_90d TINYINT,
+    total_produce_pkg_last_90d BIGINT,
+    is_consumed_all_7d TINYINT,
+    has_consumption_days_last_7d TINYINT,
+    max_consume_pkg_last_7d BIGINT,
+    min_consume_pkg_last_7d BIGINT,
+    is_consumed_all_30d TINYINT,
+    has_consumption_days_last_30d TINYINT,
+    max_consume_pkg_last_30d BIGINT,
+    min_consume_pkg_last_30d BIGINT,
+    is_consumed_all_90d TINYINT,
+    has_consumption_days_last_90d TINYINT,
+    max_consume_pkg_last_90d BIGINT,
+    min_consume_pkg_last_90d BIGINT,
+    is_backlog_all_7d TINYINT,
+    has_backlog_days_last_7d TINYINT,
+    total_backlog_last_7d BIGINT,
+    backlog_ratio_last_7d STRING,
+    is_backlog_all_30d TINYINT,
+    has_backlog_days_last_30d TINYINT,
+    total_backlog_last_30d BIGINT,
+    backlog_ratio_last_30d STRING,
+    is_backlog_all_90d TINYINT,
+    has_backlog_days_last_90d TINYINT,
+    total_backlog_last_90d BIGINT,
+    backlog_ratio_last_90d STRING
+) STORED AS ORC;
+
+INSERT INTO TABLE internal_platform_db.dws_mq_production_feature_d_increase_query_engine_013 VALUES
+('20260507','bid_001','account_alpha','cluster_a','tenant1','ns1','topic_x','消息队列MQ','数据总线','appgrp1','user_a','desc1','2025-01-01','2025-06-01','c001','dedicated','cluster_alpha','DEPT_I','product_a',1,5000,1,0,4000,3000,3000,0,0,0,0,'0','0',1,7,35000,1,30,150000,1,90,450000,1,7,28000,20000,1,30,120000,80000,1,90,360000,240000,0,0,0,'0',0,0,0,'0',0,0,0,'0'),
+('20260507','bid_002','account_beta','cluster_b',NULL,NULL,'topic_y','消息队列MQ','流处理平台','appgrp2','user_b','desc2','2025-02-15','2025-05-20','c002','shared','cluster_beta','DEPT_C','product_b',1,1000,0,0,0,0,0,0,0,0,0,'0','0',1,5,7000,1,20,30000,1,60,90000,0,3,0,0,0,10,0,0,0,30,0,0,0,0,0,'0',0,0,0,'0',0,0,0,'0'),
+('20260507','bid_003','account_gamma','cluster_c','tenant3','ns3','topic_z','TubeMQ','数据总线','appgrp3','user_c','desc3','2024-12-01','2025-04-10','c003','dedicated','cluster_gamma','DEPT_P','product_c',0,0,0,0,0,0,0,0,0,0,0,'0','0',0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,'0',0,0,0,'0',0,0,0,'0');
+
+-- 目标表: ads_mq_topic_governance_item_d
+CREATE TABLE IF NOT EXISTS internal_platform_db.ads_mq_topic_governance_item_d_query_engine_013 (
+    business_id STRING,
+    business_name STRING,
+    topic STRING,
+    cluster_set STRING,
+    mq_full_topic STRING,
+    system_belong STRING,
+    app_group STRING,
+    bid_incharge STRING,
+    bid_description STRING,
+    bid_create_time STRING,
+    bid_modify_time STRING,
+    cluster_id STRING,
+    cluster_type STRING,
+    cluster_name STRING,
+    bg STRING,
+    category_name STRING,
+    total_produce_pkg_d BIGINT,
+    production_days_last_7d TINYINT,
+    total_produce_pkg_last_7d BIGINT,
+    production_days_last_30d TINYINT,
+    total_produce_pkg_last_30d BIGINT,
+    production_days_last_90d TINYINT,
+    total_produce_pkg_last_90d BIGINT,
+    is_consumed TINYINT,
+    has_consumption_days_last_7d TINYINT,
+    has_consumption_days_last_30d TINYINT,
+    has_consumption_days_last_90d TINYINT,
+    hitted_gov_items STRING,
+    governance_benefit_estimate DOUBLE
+) PARTITIONED BY (dt STRING) STORED AS ORC;
+
+-- Agent 候选目标表
+CREATE TABLE IF NOT EXISTS internal_platform_db.ads_mq_topic_governance_item_d_cand_query_engine_013 (
+    business_id STRING,
+    business_name STRING,
+    topic STRING,
+    cluster_set STRING,
+    mq_full_topic STRING,
+    system_belong STRING,
+    app_group STRING,
+    bid_incharge STRING,
+    bid_description STRING,
+    bid_create_time STRING,
+    bid_modify_time STRING,
+    cluster_id STRING,
+    cluster_type STRING,
+    cluster_name STRING,
+    bg STRING,
+    category_name STRING,
+    total_produce_pkg_d BIGINT,
+    production_days_last_7d TINYINT,
+    total_produce_pkg_last_7d BIGINT,
+    production_days_last_30d TINYINT,
+    total_produce_pkg_last_30d BIGINT,
+    production_days_last_90d TINYINT,
+    total_produce_pkg_last_90d BIGINT,
+    is_consumed TINYINT,
+    has_consumption_days_last_7d TINYINT,
+    has_consumption_days_last_30d TINYINT,
+    has_consumption_days_last_90d TINYINT,
+    hitted_gov_items STRING,
+    governance_benefit_estimate DOUBLE
+) PARTITIONED BY (dt STRING) STORED AS ORC;

tasks/offline-compute/HiveSQL/hivesql_005/init/schema.sql

@@ -0,0 +1,69 @@
+CREATE TABLE IF NOT EXISTS internal_platform_db.dws_mq_production_feature_d_increase (
+    `dt` STRING COMMENT '天分区',
+    `business_id` STRING COMMENT 'business_id',
+    `business_name` STRING COMMENT 'business_name',
+    `cluster_set` STRING COMMENT 'cluster_set',
+    `tenant` STRING COMMENT 'tenant',
+    `namespaces` STRING COMMENT 'namespaces',
+    `topic` STRING COMMENT 'topic',
+    `mq_type` STRING COMMENT 'MQ类型（消息队列MQ/TubeMQ）',
+    `system_belong` STRING COMMENT 'topic所属的系统（数据总线/流处理平台）',
+    `dw_appgroup` STRING COMMENT 'bid所属应用组',
+    `in_charge` STRING COMMENT 'bid负责人',
+    `description` STRING COMMENT 'bid描述信息',
+    `create_time` STRING COMMENT 'bid创建时间',
+    `modify_time` STRING COMMENT 'bid最后修改时间',
+    `cluster_id` STRING COMMENT 'bid所属集群id',
+    `cluster_type` STRING COMMENT 'bid所属集群类型',
+    `cluster_name` STRING COMMENT 'bid所属集群名',
+    `bg` STRING COMMENT 'bid所属的bg',
+    `category_name` STRING COMMENT 'bid所归属的产品',
+    `is_produce` TINYINT COMMENT '当天topic是否有生产',
+    `total_produce_pkg_d` BIGINT COMMENT '当天topic生产的消息数的总量',
+    `is_consumed` TINYINT COMMENT '当天topic下游是否有消费',
+    `is_consumed_by_reader` TINYINT COMMENT '当天topic下游是否有reader消费组在消费',
+    `max_consume_pkg_d` BIGINT COMMENT '当天topic的所有消费组中最大的消费量',
+    `min_consume_pkg_d` BIGINT COMMENT '当天topic的所有消费组中最小的消费量',
+   
+   `min_consume_pkg_without_reader_d` BIGINT COMMENT '当天topic的所有消费组中最小的消费量（不计算reader消费组）',
+    `is_backlog` TINYINT COMMENT '当天topic下游是否有积压',
+    `is_backlog_without_reader` TINYINT COMMENT '当天topic下游是否有积压（不计算reader消费组）',
+    `total_backlog_pkg_d` BIGINT COMMENT '当天topic下游积压增量',
+   
+   `total_backlog_pkg_without_reader_d` BIGINT COMMENT '当天topic下游积压增量（不计算reader消费组）',
+    `backlog_ratio` STRING COMMENT '当天topic下游积压生产占比(%)',
+    `backlog_ratio_without_reader` STRING COMMENT '当天topic下游积压生产占比(%)（不计算reader消费组）',
+    `is_produce_all_7d` TINYINT COMMENT '是否近7天每天都有生产（1:是 0:否）',
+    `production_days_last_7d` TINYINT COMMENT '近7天有生产的天数',
+    `total_produce_pkg_last_7d` BIGINT COMMENT '近7天topic生产的消息数总量',
+    `is_produce_all_30d` TINYINT COMMENT '是否近30天每天都有生产（1:是 0:否）',
+    `production_days_last_30d` TINYINT COMMENT '近30天有生产的天数',
+    `total_produce_pkg_last_30d` BIGINT COMMENT '近30天topic生产的消息数总量',
+    `is_produce_all_90d` TINYINT COMMENT '是否近90天每天都有生产（1:是 0:否）',
+    `production_days_last_90d` TINYINT COMMENT '近90天有生产的天数',
+    `total_produce_pkg_last_90d` BIGINT COMMENT '近90天topic生产的消息数总量',
+    `is_consumed_all_7d` TINYINT COMMENT '是否近7天每天都有消费（1:是 0:否）',
+    `has_consumption_days_last_7d` TINYINT COMMENT '近7天topic下游有消费的天数',
+    `max_consume_pkg_last_7d` BIGINT COMMENT '近7天topic所有消费组中最大的消费量',
+    `min_consume_pkg_last_7d` BIGINT COMMENT '近7天topic所有消费组中最小的消费量',
+    `is_consumed_all_30d` TINYINT COMMENT '是否近30天每天都有消费（1:是 0:否）',
+    `has_consumption_days_last_30d` TINYINT COMMENT '近30天topic下游有消费的天数',
+    `max_consume_pkg_last_30d` BIGINT COMMENT '近30天topic所有消费组中最大的消费量',
+    `min_consume_pkg_last_30d` BIGINT COMMENT '近30天topic所有消费组中最小的消费量',
+    `is_consumed_all_90d` TINYINT COMMENT '是否近90天每天都有消费（1:是 0:否）',
+    `has_consumption_days_last_90d` TINYINT COMMENT '近90天topic下游有消费的天数',
+    `max_consume_pkg_last_90d` BIGINT COMMENT '近90天topic所有消费组中最大的消费量',
+    `min_consume_pkg_last_90d` BIGINT COMMENT '近90天topic所有消费组中最小的消费量',
+    `is_backlog_all_7d` TINYINT COMMENT '是否近7天每天都有积压（1:是 0:否）',
+    `has_backlog_days_last_7d` TINYINT COMMENT '近7天topic下游有积压的天数',
+    `total_backlog_last_7d` BIGINT COMMENT '近7天topic的积压量',
+    `backlog_ratio_last_7d` STRING COMMENT '近七天topic下游积压生产占比(%)',
+    `is_backlog_all_30d` TINYINT COMMENT '是否近30天每天都有积压（1:是 0:否）',
+    `has_backlog_days_last_30d` TINYINT COMMENT '近30天topic下游有积压的天数',
+    `total_backlog_last_30d` BIGINT COMMENT '近30天topic的积压量',
+    `backlog_ratio_last_30d` STRING COMMENT '近30天topic下游积压生产占比(%)',
+    `is_backlog_all_90d` TINYINT COMMENT '是否近90天每天都有积压（1:是 0:否）',
+    `has_backlog_days_last_90d` TINYINT COMMENT '近90天topic下游有积压的天数',
+    `total_backlog_last_90d` BIGINT COMMENT '近90天topic的积压量',
+    `backlog_ratio_last_90d` STRING COMMENT '近90天topic下游积压生产占比(%)'
+) STORED AS ORC;

tasks/offline-compute/HiveSQL/hivesql_005/task.md

@@ -0,0 +1,39 @@
+---
+id: offline-compute_HiveSQL_hivesql_005
+name: 从 internal_platform_db.dws_mq_production_featur
+category: offline-compute/HiveSQL
+timeout_seconds: 600
+modality: pure-text
+engine: hivesql
+---
+## Prompt
+任务目标：汇总当天有生产量的 消息队列MQ topic 维度信息及近7/30/90天生产统计，落地为 topic 治理项明细表。
+
+输入：
+- 表：internal_platform_db.dws_mq_production_feature_d_increase_query_engine_013
+- 分区字段：dt（STRING，格式 YYYYMMDD）
+- 关键字段：business_id、business_name、topic、cluster_set、tenant、namespaces、system_belong、dw_appgroup、in_charge、description、create_time、modify_time、cluster_id、cluster_type、cluster_name、bg、category_name、total_produce_pkg_d、production_days_last_7d、total_produce_pkg_last_7d、production_days_last_30d、total_produce_pkg_last_30d、production_days_last_90d、total_produce_pkg_last_90d、is_consumed、has_consumption_days_last_7d、has_consumption_days_last_30d、has_consumption_days_last_90d
+
+处理规则：
+1. 过滤条件：dt = '20260507' AND total_produce_pkg_last_90d > 0
+2. 无 Join，单表处理
+3. 派生字段：
+   - mq_full_topic：若 tenant 和 namespaces 均非 NULL，则拼接为 'persistent://' + tenant + '/' + namespaces + '/' + topic；否则直接取 topic
+   - app_group：取自 dw_appgroup
+   - bid_incharge：取自 in_charge
+   - bid_description：取自 description
+   - bid_create_time：取自 create_time
+   - bid_modify_time：取自 modify_time
+   - hitted_gov_items：固定为 NULL
+   - governance_benefit_estimate：固定为 NULL
+
+输出要求：
+- 输出字段顺序：dt、business_id、business_name、topic、cluster_set、mq_full_topic、system_belong、app_group、bid_incharge、bid_description、bid_create_time、bid_modify_time、cluster_id、cluster_type、cluster_name、bg、category_name、total_produce_pkg_d、production_days_last_7d、total_produce_pkg_last_7d、production_days_last_30d、total_produce_pkg_last_30d、production_days_last_90d、total_produce_pkg_last_90d、is_consumed、has_consumption_days_last_7d、has_consumption_days_last_30d、has_consumption_days_last_90d、hitted_gov_items、governance_benefit_estimate
+
+写入要求：
+- 目标表：internal_platform_db.ads_mq_topic_governance_item_d_cand_query_engine_013
+- 分区字段：dt（STRING，格式 YYYYMMDD）
+- 写入方式：INSERT OVERWRITE 按分区写入
+- 若目标表不存在，先按 Hive 标准建表（ORC存储、按 dt 分区），再写入
+
+请将最终 HiveSQL 写入 result.sql 并执行。

tasks/offline-compute/HiveSQL/hivesql_006_en/gt/expected.csv

@@ -0,0 +1,4 @@
+system_belong,category_name,dw_appgroup,city_id,cluster_set,topic,total_data_size_d,total_cost,in_charge,dt
+SystemAlpha,数据仓库DWProdA,AppGroupX,101,ClusterA,topic_billing,2560000,162.0,user_zhang,20260507
+SystemBeta,数据仓库DWProdB,AppGroupY,202,ClusterB,topic_log,1200000,81.0,user_li,20260507
+SystemGamma,数据仓库DWProdC,AppGroupZ,303,ClusterC,topic_metrics,512000,50.0,user_wang,20260507

tasks/offline-compute/HiveSQL/hivesql_006_en/gt/grade.py

@@ -0,0 +1,705 @@
+"""hivesql_015 精细评分脚本
+
+业务场景：数据总线 topic接入成本原始数据清洗并按topic维度聚合，排除测试topic，写入明细表
+难度: EASY | 特征: INSERT_OVERWRITE|PARTITION|SINGLE_TABLE|GROUP_BY
+
+评分范式: 产物100分 = A(15) + B(10) + C(15) + D(40) + F(20)
+  A_executability       (15分): result.sql 能跑通且产出非空
+  B_schema              (10分): 列数(5) + 列名匹配(5)
+  C_row_alignment     (15分): 行数比例(7) + key覆盖率(8)
+  D_field_value_match   (25分): 非key字段逐列值匹配率
+  D_field_completeness  (15分): 关键字段非空/非空串比例
+  F_insert_overwrite     (5分): INSERT OVERWRITE + PARTITION
+  F_partition_value      (5分): dt 分区值 = 20260507
+  F_source_filter       (10分): 源表分区过滤 dt=20260507 + topic<>'test'
+
+权重: EASY -> product=0.5, process=0.5
+"""
+import os
+import re
+import subprocess
+import tempfile
+import json
+import math
+
+
+def grade(workspace_path, **kwargs):
+
+    # ========== Case 配置 ==========
+    OUTPUT_TABLE = "internal_platform_db.dwd_databus_topic_cost_detail_d_cand_query_engine_015"
+    DIFFICULTY = "EASY"
+    SOURCE_TABLES = ["ods_t_databus_access_topic_cost_date_d_query_engine_015", "dwd_databus_topic_cost_detail_d_query_engine_015"]
+    KEY_COLUMNS = ["city_id", "dt"]
+    EXPECTED_COL_COUNT = 10
+    KEY_FIELDS = ["system_belong", "topic", "total_data_size_d", "total_cost", "in_charge"]
+    GT_TABLE = OUTPUT_TABLE.replace("_cand_", "_")
+
+    DIFFICULTY_WEIGHTS = {
+        "EASY": (0.5, 0.5),
+        "MEDIUM": (0.6, 0.4),
+        "HARD": (0.7, 0.3),
+        "EXPERT": (0.8, 0.2),
+    }
+
+    _SPARK_SUBMIT_TIMEOUT = 300
+    _JSON_START = "__GRADE_JSON_START__"
+    _JSON_END = "__GRADE_JSON_END__"
+
+    result = {
+        "overall_score": 0.0,
+        "total_points": 0,
+        "grade": "",
+        "details": {},
+        "diagnostics": [],
+    }
+
+    # ========== 内部辅助函数 ==========
+
+    def values_match(pred_val, gt_val, abs_tol=1e-6, rel_tol=1e-4):
+        if pred_val is None and gt_val is None:
+            return True
+        if pred_val is None or gt_val is None:
+            return False
+        s_pred = str(pred_val).strip()
+        s_gt = str(gt_val).strip()
+        if s_pred == s_gt:
+            return True
+        try:
+            pv = float(s_pred)
+            gv = float(s_gt)
+            if math.isnan(pv) and math.isnan(gv):
+                return True
+            if math.isnan(pv) or math.isnan(gv):
+                return False
+            if abs(gv) < abs_tol:
+                return abs(pv - gv) <= abs_tol
+            return abs(pv - gv) <= abs_tol or abs(pv - gv) / max(abs(gv), 1e-12) <= rel_tol
+        except (ValueError, TypeError):
+            pass
+        return s_pred.lower() == s_gt.lower()
+
+    def _run_spark_script(script_code, timeout=_SPARK_SUBMIT_TIMEOUT):
+        with tempfile.NamedTemporaryFile(mode="w", suffix=".py", delete=False, encoding="utf-8") as f:
+            f.write(script_code)
+            script_path = f.name
+        try:
+            r = subprocess.run(
+                ["spark-submit", script_path],
+                capture_output=True, text=True, timeout=timeout,
+            )
+            stdout = r.stdout or ""
+            if _JSON_START in stdout and _JSON_END in stdout:
+                json_str = stdout.split(_JSON_START)[1].split(_JSON_END)[0].strip()
+                return json.loads(json_str), None
+            else:
+                if r.returncode == 0:
+                    for line in stdout.splitlines():
+                        if line.strip().startswith("Traceback"):
+                            return None, f"spark-submit error: {line}"
+                    return None, "spark-submit 无 JSON 输出"
+                err_msg = (r.stderr or "")[-500:]
+                return None, f"spark-submit failed: {err_msg}"
+        except subprocess.TimeoutExpired:
+            return None, f"spark-submit 超时 ({timeout}s)"
+        except Exception as e:
+            return None, f"spark-submit 异常: {e}"
+        finally:
+            try:
+                os.unlink(script_path)
+            except OSError:
+                pass
+
+    def read_table_via_spark_submit(table_name):
+        """Read table via spark-submit subprocess. Returns (cols, rows_as_lists)."""
+        if not re.match(r'^[a-zA-Z_][a-zA-Z0-9_.]*$', table_name):
+            raise ValueError(f"非法表名: {table_name}")
+        read_script = f'''
+import json
+from pyspark.sql import SparkSession
+spark = SparkSession.builder.appName("grade_read").enableHiveSupport() \\
+    .config("spark.sql.warehouse.dir", "/tmp/hive_warehouse").getOrCreate()
+try:
+    df = spark.sql("SELECT * FROM {table_name}")
+    cols = [c.lower() for c in df.columns]
+    rows = [[str(v) if v is not None else "" for v in row] for row in df.collect()]
+    print("{_JSON_START}")
+    print(json.dumps({{"cols": cols, "rows": rows}}, ensure_ascii=False))
+    print("{_JSON_END}")
+except Exception as e:
+    print("{_JSON_START}")
+    print(json.dumps({{"error": str(e)}}))
+    print("{_JSON_END}")
+finally:
+    spark.stop()
+'''
+        data, err = _run_spark_script(read_script)
+        if err:
+            raise RuntimeError(f"read_table failed: {err}")
+        if "error" in data:
+            raise RuntimeError(f"query failed: {data['error']}")
+        return data["cols"], data["rows"]
+
+    # SQL executor template (self-contained, no external dependency)
+    _SQL_EXEC_TEMPLATE = '''
+import json, re
+from pyspark.sql import SparkSession
+spark = SparkSession.builder.appName("{app_name}").enableHiveSupport() \\
+    .config("spark.sql.warehouse.dir", "/tmp/hive_warehouse").getOrCreate()
+try:
+    with open("{sql_file}", "r", encoding="utf-8") as _f:
+        _sql = _f.read()
+    _sql = re.sub(r"^\\s*set\\s+query_engine\\.\\S+\\n?", "", _sql, flags=re.IGNORECASE)
+    _stmts, _cur, _in_sq, _in_dq, _i = [], [], False, False, 0
+    while _i < len(_sql):
+        _ch = _sql[_i]
+        if _ch == "\\\\" and _i + 1 < len(_sql):
+            _cur.append(_ch); _cur.append(_sql[_i+1]); _i += 2; continue
+        if _ch == "-" and _i+1 < len(_sql) and _sql[_i+1] == "-" and not _in_sq and not _in_dq:
+            while _i < len(_sql) and _sql[_i] != "\\n": _i += 1
+            _cur.append("\\n"); continue
+        if _ch == "'" and not _in_dq: _in_sq = not _in_sq
+        elif _ch == '"' and not _in_sq: _in_dq = not _in_dq
+        if _ch == ";" and not _in_sq and not _in_dq:
+            _s = "".join(_cur).strip()
+            if _s: _stmts.append(_s)
+            _cur = []
+        else:
+            _cur.append(_ch)
+        _i += 1
+    _last = "".join(_cur).strip()
+    if _last: _stmts.append(_last)
+    for _stmt in _stmts:
+        spark.sql(_stmt)
+    print("{_JSON_START}")
+    print(json.dumps({{"ok": True}}))
+    print("{_JSON_END}")
+except Exception as e:
+    print("{_JSON_START}")
+    print(json.dumps({{"ok": False, "error": str(e)}}))
+    print("{_JSON_END}")
+finally:
+    spark.stop()
+'''
+
+    def execute_result_sql():
+        result_sql = os.path.join(workspace_path, "result.sql")
+        if not os.path.exists(result_sql):
+            return False, "no_result_file"
+        # 替换 数据平台WD时间变量（沙箱 spark-sql 不支持 ${...} 语法）
+        with open(result_sql, 'r', encoding='utf-8') as _rf:
+            _sql_text = _rf.read()
+        _bizdate = '20260507'
+        _sql_text = re.sub(r'\${bdp\.system\.bizdate(?:[+-]\d+)?}', _bizdate, _sql_text)
+        _sql_text = re.sub(r'\${yyyymmdd(?:[+-]\d+)?}', _bizdate, _sql_text)
+        _sql_text = re.sub(r'\${bizdate(?:[+-]\d+)?}', _bizdate, _sql_text)
+        _sql_text = re.sub(r'\${[^}]*date[^}]*}', _bizdate, _sql_text)
+        with open(result_sql, 'w', encoding='utf-8') as _wf:
+            _wf.write(_sql_text)
+        script = _SQL_EXEC_TEMPLATE.format(app_name="grade_exec", sql_file=result_sql,
+                                            _JSON_START=_JSON_START, _JSON_END=_JSON_END)
+        data, err = _run_spark_script(script)
+        if err:
+            return False, f"execution_error: {err}"
+        if data and data.get("ok"):
+            return True, None
+        return False, f"execution_error: {data.get('error', 'unknown') if data else 'no output'}"
+
+    def execute_ground_truth_sql():
+        gt_sql = os.path.join(workspace_path, "gt", "ground_truth.sql")
+        if not os.path.exists(gt_sql):
+            return False, "ground_truth.sql not found"
+        script = _SQL_EXEC_TEMPLATE.format(app_name="grade_gt", sql_file=gt_sql,
+                                            _JSON_START=_JSON_START, _JSON_END=_JSON_END)
+        data, err = _run_spark_script(script)
+        if err:
+            return False, f"gt_execution_error: {err}"
+        if data and data.get("ok"):
+            return True, None
+        return False, f"gt_execution_error: {data.get('error', 'unknown') if data else 'no output'}"
+
+    def truncate_table(table_name):
+        script = f'''
+from pyspark.sql import SparkSession
+spark = SparkSession.builder.appName("truncate").enableHiveSupport() \\
+    .config("spark.sql.warehouse.dir", "/tmp/hive_warehouse").getOrCreate()
+spark.sql("TRUNCATE TABLE {table_name}")
+spark.stop()
+'''
+        try:
+            _run_spark_script(script, timeout=120)
+        except Exception:
+            pass
+
+    def restore_hive_site():
+        """Restore hive-site.xml to canonical state (agent may have modified it)."""
+        canonical_hive_site = '''<?xml version="1.0"?>
+<?xml-stylesheet type="text/xsl" href="configuration.xsl"?>
+<configuration>
+  <property>
+    <name>hive.metastore.uris</name>
+    <value>thrift://localhost:9083</value>
+  </property>
+  <property>
+    <name>hive.metastore.warehouse.dir</name>
+    <value>/tmp/hive_warehouse</value>
+  </property>
+  <property>
+    <name>javax.jdo.option.ConnectionURL</name>
+    <value>jdbc:derby:;databaseName=/tmp/hive_metastore_db;create=true</value>
+  </property>
+  <property>
+    <name>javax.jdo.option.ConnectionDriverName</name>
+    <value>org.apache.derby.jdbc.EmbeddedDriver</value>
+  </property>
+  <property>
+    <name>datanucleus.schema.autoCreateAll</name>
+    <value>true</value>
+  </property>
+  <property>
+    <name>hive.metastore.schema.verification</name>
+    <value>false</value>
+  </property>
+</configuration>
+'''
+        hive_site_path = os.path.join(os.environ.get('SPARK_HOME', '/opt/spark'), 'conf', 'hive-site.xml')
+        try:
+            with open(hive_site_path, 'w') as f:
+                f.write(canonical_hive_site)
+        except Exception:
+            pass
+
+    def finalize(result):
+        product_weight, process_weight = DIFFICULTY_WEIGHTS.get(DIFFICULTY, (0.7, 0.3))
+        product_dims = ["A_executability"] + ['B_schema', 'C_row_alignment', 'D_field_completeness', 'D_field_value_match', 'F_insert_overwrite', 'F_partition_value', 'F_source_filter']
+        product_raw = sum(result["details"].get(d, {}).get("score", 0) for d in product_dims)
+        product_ratio = product_raw / 100.0
+        for dim in ["H_efficiency"]:
+            if dim in result["details"]:
+                raw = result["details"][dim].get("score", 0)
+                result["details"][dim]["score_before_scaling"] = raw
+                result["details"][dim]["score"] = round(raw * product_ratio, 2)
+                result["details"][dim]["product_ratio"] = round(product_ratio, 4)
+        process_dims = ["G_exploration", "H_efficiency", "I_self_verification"]
+        process_raw = sum(result["details"].get(d, {}).get("score", 0) for d in process_dims)
+        product_score = round(product_raw * product_weight, 2)
+        process_score = round(process_raw * process_weight, 2)
+        total = round(product_score + process_score, 2)
+        result["total_points"] = total
+        result["product_points"] = product_score
+        result["process_points"] = round(process_score, 2)
+        result["weights"] = {"product": product_weight, "process": process_weight}
+        result["overall_score"] = round(total / 100.0, 4)
+        if total >= 90:
+            result["grade"] = "优秀"
+        elif total >= 75:
+            result["grade"] = "良好"
+        elif total >= 60:
+            result["grade"] = "合格"
+        elif total >= 40:
+            result["grade"] = "偏弱"
+        else:
+            result["grade"] = "不合格"
+        return result
+
+    # Restore hive-site.xml (agent may have modified it)
+    restore_hive_site()
+
+    # ========== A. 可执行性 (15分) ==========
+    a_items = {"A1_exec_ok": 0, "A2_has_data": 0}
+    exec_ok, exec_err = execute_result_sql()
+    if not exec_ok:
+        detail = "未产出 result.sql" if exec_err == "no_result_file" else str(exec_err)[:200]
+        result["details"]["A_executability"] = {"score": 0, "max": 15, "detail": detail}
+        result["error"] = detail
+        for dim in ['B_schema', 'C_row_alignment', 'D_field_completeness', 'D_field_value_match', 'F_insert_overwrite', 'F_partition_value', 'F_source_filter']:
+            result["details"][dim] = {"score": 0, "max": 0, "items": {}}
+        return finalize(result)
+
+    a_items["A1_exec_ok"] = 8
+
+    pred_headers, pred_rows = [], []
+    try:
+        pred_headers, pred_rows = read_table_via_spark_submit(OUTPUT_TABLE)
+    except Exception as e:
+        result["diagnostics"].append(f"read_pred_failed: {e}")
+
+    if not pred_rows:
+        result["details"]["A_executability"] = {"score": 8, "max": 15, "items": a_items}
+        for dim in ['B_schema', 'C_row_alignment', 'D_field_completeness', 'D_field_value_match', 'F_insert_overwrite', 'F_partition_value', 'F_source_filter']:
+            result["details"][dim] = {"score": 0, "max": 0, "items": {}}
+        return finalize(result)
+
+    a_items["A2_has_data"] = 7
+    result["details"]["A_executability"] = {"score": 15, "max": 15, "items": a_items}
+
+    # ========== Execute GT + Read GT ==========
+    truncate_table(GT_TABLE)
+    gt_ok, gt_err = execute_ground_truth_sql()
+    if not gt_ok:
+        result["error"] = f"ground_truth failed: {gt_err}"
+        for dim in ['B_schema', 'C_row_alignment', 'D_field_completeness', 'D_field_value_match', 'F_insert_overwrite', 'F_partition_value', 'F_source_filter']:
+            result["details"][dim] = {"score": 0, "max": 0, "items": {}}
+        return finalize(result)
+
+    gt_headers, gt_rows = [], []
+    try:
+        gt_headers, gt_rows = read_table_via_spark_submit(GT_TABLE)
+    except Exception as e:
+        result["error"] = f"read_gt_failed: {e}"
+        for dim in ['B_schema', 'C_row_alignment', 'D_field_completeness', 'D_field_value_match', 'F_insert_overwrite', 'F_partition_value', 'F_source_filter']:
+            result["details"][dim] = {"score": 0, "max": 0, "items": {}}
+        return finalize(result)
+
+    # ========== B/C/D/F 维度评分 ==========
+    pred_col_map = {h: i for i, h in enumerate(pred_headers)}
+    gt_col_map = {h: i for i, h in enumerate(gt_headers)}
+
+    # ========== B. Schema正确性 (10分) ==========
+    b_items = {}
+    # B1: 列数 (5分)
+    if len(pred_headers) == EXPECTED_COL_COUNT:
+        b_items["B1_col_count"] = 5
+    elif abs(len(pred_headers) - EXPECTED_COL_COUNT) <= 2:
+        b_items["B1_col_count"] = 3
+    else:
+        b_items["B1_col_count"] = 0
+
+    # B2: 列名匹配 (5分)
+    gt_col_set = set(gt_headers)
+    pred_col_set = set(pred_headers)
+    name_match_rate = len(gt_col_set & pred_col_set) / max(len(gt_col_set), 1)
+    if name_match_rate >= 0.95:
+        b_items["B2_col_names"] = 5
+    elif name_match_rate >= 0.8:
+        b_items["B2_col_names"] = 3
+    else:
+        b_items["B2_col_names"] = 0
+
+    b_score = sum(b_items.values())
+    result["details"]["B_schema"] = {
+        "score": b_score, "max": 10,
+        "detail": {"col_count": len(pred_headers), "name_match_rate": round(name_match_rate, 4), "items": b_items},
+    }
+
+    # ========== C. 行一致性 (15分) ==========
+    c_items = {}
+    gt_row_count = len(gt_rows)
+    pred_row_count = len(pred_rows)
+
+    # C1: 行数比例 (7分)
+    if gt_row_count > 0:
+        ratio = pred_row_count / gt_row_count
+        if 0.95 <= ratio <= 1.05:
+            c_items["C1_row_ratio"] = 7
+        elif 0.7 <= ratio <= 1.3:
+            c_items["C1_row_ratio"] = 4
+        else:
+            c_items["C1_row_ratio"] = 0
+    else:
+        c_items["C1_row_ratio"] = 7 if pred_row_count == 0 else 0
+
+    # C2: key覆盖率 (8分)
+    key_cols_avail = [k for k in KEY_COLUMNS if k in gt_col_map and k in pred_col_map]
+    if key_cols_avail and gt_row_count > 0:
+        gt_keys = set()
+        for row in gt_rows:
+            key = tuple(row[gt_col_map[k]] if gt_col_map[k] < len(row) else "" for k in key_cols_avail)
+            gt_keys.add(key)
+        pred_keys = set()
+        for row in pred_rows:
+            key = tuple(row[pred_col_map[k]] if pred_col_map[k] < len(row) else "" for k in key_cols_avail)
+            pred_keys.add(key)
+        coverage = len(gt_keys & pred_keys) / max(len(gt_keys), 1)
+        if coverage >= 0.995:
+            c_items["C2_key_coverage"] = 8
+        elif coverage >= 0.9:
+            c_items["C2_key_coverage"] = 6
+        elif coverage >= 0.7:
+            c_items["C2_key_coverage"] = 3
+        else:
+            c_items["C2_key_coverage"] = round(8 * coverage, 2)
+    else:
+        c_items["C2_key_coverage"] = 0
+
+    c_score = sum(v for v in c_items.values())
+    result["details"]["C_row_alignment"] = {"score": c_score, "max": 15, "detail": c_items}
+
+    # 构建索引
+    if key_cols_avail:
+        pred_index = {}
+        for row in pred_rows:
+            key = tuple(row[pred_col_map[k]] if pred_col_map[k] < len(row) else "" for k in key_cols_avail)
+            pred_index[key] = row
+        gt_index = {}
+        for row in gt_rows:
+            key = tuple(row[gt_col_map[k]] if gt_col_map[k] < len(row) else "" for k in key_cols_avail)
+            gt_index[key] = row
+    else:
+        pred_index = {}
+        gt_index = {}
+
+    # ========== D. 字段值匹配 (25分) ==========
+    value_cols = [c for c in gt_headers if c in pred_col_map and c not in KEY_COLUMNS]
+    if not value_cols:
+        value_cols = [c for c in gt_headers if c in pred_col_map]
+
+    d_val_items = {}
+    per_col_weight = 25.0 / max(len(value_cols), 1)
+    d_val_score = 0
+    for col in value_cols:
+        gt_ci = gt_col_map.get(col)
+        pred_ci = pred_col_map.get(col)
+        if gt_ci is None or pred_ci is None:
+            d_val_items[col] = {"pass_rate": 0.0, "score": 0, "reason": "column_missing"}
+            continue
+        matches = 0
+        total = 0
+        for key, gt_row in gt_index.items():
+            pred_row = pred_index.get(key)
+            if pred_row is None:
+                total += 1
+                continue
+            gt_val = gt_row[gt_ci].strip() if gt_ci < len(gt_row) else ""
+            pred_val = pred_row[pred_ci].strip() if pred_ci < len(pred_row) else ""
+            if values_match(pred_val, gt_val):
+                matches += 1
+            total += 1
+        rate = matches / max(total, 1)
+        col_score = rate * per_col_weight
+        d_val_score += col_score
+        d_val_items[col] = {"pass_rate": round(rate, 4), "score": round(col_score, 2)}
+
+    result["details"]["D_field_value_match"] = {
+        "score": round(d_val_score, 2), "max": 25, "detail": d_val_items,
+    }
+
+    # ========== D. 字段完整性 (15分) ==========
+    d_comp_items = {}
+    d_comp_score = 0
+    per_field_weight = 15.0 / max(len(KEY_FIELDS), 1)
+    for field in KEY_FIELDS:
+        gt_ci = gt_col_map.get(field)
+        pred_ci = pred_col_map.get(field)
+        if gt_ci is None or pred_ci is None:
+            d_comp_items[field] = {"score": 0, "reason": "column_missing"}
+            continue
+        gt_nonempty = 0
+        pred_nonempty = 0
+        total = 0
+        for key, gt_row in gt_index.items():
+            gt_val = gt_row[gt_ci].strip() if gt_ci < len(gt_row) else ""
+            if gt_val:
+                total += 1
+                pred_row = pred_index.get(key)
+                if pred_row is not None:
+                    pred_val = pred_row[pred_ci].strip() if pred_ci < len(pred_row) else ""
+                    if pred_val:
+                        pred_nonempty += 1
+        gt_nonempty = total
+        rate = pred_nonempty / max(gt_nonempty, 1)
+        field_score = rate * per_field_weight
+        d_comp_score += field_score
+        d_comp_items[field] = {"nonempty_rate": round(rate, 4), "score": round(field_score, 2)}
+
+    result["details"]["D_field_completeness"] = {
+        "score": round(d_comp_score, 2), "max": 15, "detail": d_comp_items,
+    }
+
+    # ========== F. INSERT OVERWRITE (5分) ==========
+    result_sql_path = os.path.join(workspace_path, "result.sql")
+    sql_text = ""
+    try:
+        with open(result_sql_path, "r", encoding="utf-8") as f:
+            sql_text = f.read().lower()
+    except Exception:
+        sql_text = ""
+
+    f_insert_items = {}
+    has_overwrite = "insert overwrite" in sql_text
+    has_partition = "partition" in sql_text
+    if has_overwrite and has_partition:
+        f_insert_items["insert_overwrite_partition"] = 5
+    elif has_overwrite:
+        f_insert_items["insert_overwrite_partition"] = 3
+    else:
+        f_insert_items["insert_overwrite_partition"] = 0
+    result["details"]["F_insert_overwrite"] = {
+        "score": f_insert_items["insert_overwrite_partition"], "max": 5, "detail": f_insert_items,
+    }
+
+    # ========== F. 分区值 (5分) ==========
+    f_part_items = {}
+    if "dt" in sql_text and "20260507" in sql_text:
+        f_part_items["partition_value"] = 5
+    else:
+        f_part_items["partition_value"] = 0
+    result["details"]["F_partition_value"] = {
+        "score": f_part_items["partition_value"], "max": 5, "detail": f_part_items,
+    }
+
+    # ========== F. 源表分区过滤 (10分) ==========
+    f_filter_items = {}
+    source_table_lower = SOURCE_TABLES[0].lower()
+    has_source_table = source_table_lower in sql_text
+    has_dt_filter = "dt" in sql_text and "20260507" in sql_text
+    has_topic_filter = "test" in sql_text and "<>" in sql_text
+    if has_source_table and has_dt_filter and has_topic_filter:
+        f_filter_items["source_partition_filter"] = 10
+    elif has_dt_filter and has_topic_filter:
+        f_filter_items["source_partition_filter"] = 7
+    elif has_dt_filter:
+        f_filter_items["source_partition_filter"] = 4
+    elif has_source_table:
+        f_filter_items["source_partition_filter"] = 2
+    else:
+        f_filter_items["source_partition_filter"] = 0
+    result["details"]["F_source_filter"] = {
+        "score": f_filter_items["source_partition_filter"], "max": 10, "detail": f_filter_items,
+    }
+
+    # 汇总
+    total = (15 + b_score + c_score + d_val_score + d_comp_score
+             + f_insert_items["insert_overwrite_partition"]
+             + f_part_items["partition_value"]
+             + f_filter_items["source_partition_filter"])
+
+    # ========== G~I 过程评分 ==========
+    TRANSCRIPT_PATH = "/tmp/dataclaw_chat.jsonl"
+    OUTPUT_TABLE_SHORT = OUTPUT_TABLE.split(".")[-1]
+    INPUT_TABLE_SHORT = SOURCE_TABLES[0]
+
+    transcript_entries = []
+    has_transcript = False
+    try:
+        if os.path.exists(TRANSCRIPT_PATH):
+            with open(TRANSCRIPT_PATH, "r", encoding="utf-8", errors="ignore") as f:
+                for line in f:
+                    line = line.strip()
+                    if line:
+                        try:
+                            transcript_entries.append(json.loads(line))
+                        except json.JSONDecodeError:
+                            continue
+            if len(transcript_entries) > 2:
+                has_transcript = True
+    except Exception:
+        pass
+
+    if not has_transcript:
+        result["details"]["G_exploration"] = {"score": 0, "max": 35, "items": {"no_transcript": True}}
+        result["details"]["H_efficiency"] = {"score": 0, "max": 40, "items": {"no_transcript": True}}
+        result["details"]["I_self_verification"] = {"score": 0, "max": 25, "items": {"no_transcript": True}}
+        return finalize(result)
+
+    # Parse transcript into structured events
+    tool_uses = []
+    first_write_result_idx = None
+    last_spark_submit_success_idx = None
+    write_result_count = 0
+    logic_error_retries = 0
+
+    for idx, entry in enumerate(transcript_entries):
+        content = entry.get("content", [])
+        if isinstance(content, str):
+            content = [content]
+
+        for block_str in content:
+            if not isinstance(block_str, str):
+                continue
+            if "ToolUseBlock" in block_str:
+                name_match = re.search(r"name='([^']+)'", block_str)
+                input_match = re.search(r"input=(\{.*\})", block_str)
+                if name_match:
+                    tool_name = name_match.group(1)
+                    tool_input = input_match.group(1) if input_match else ""
+                    tool_uses.append((idx, tool_name, tool_input))
+                    if tool_name == "Write" and "result.sql" in tool_input:
+                        write_result_count += 1
+                        if first_write_result_idx is None:
+                            first_write_result_idx = idx
+            if "ToolResultBlock" in block_str:
+                if "Traceback" in block_str or "Exception" in block_str:
+                    env_errors = ["Derby", "metastore", "HiveMetaStore", "Connection refused",
+                                  "db.lck", "TTransportException", "port 10000"]
+                    is_env_error = any(e in block_str for e in env_errors)
+                    has_spark_submit = any(t[1] == "Bash" and "spark-submit" in t[2] and "result.sql" in t[2]
+                                          for t in tool_uses)
+                    if not is_env_error and has_spark_submit:
+                        logic_error_retries += 1
+                if ("spark-submit" in block_str or "spark-sql" in block_str) and "result.sql" in block_str:
+                    if "Exit Code: 0" in block_str and "Traceback" not in block_str:
+                        last_spark_submit_success_idx = idx
+
+    before_first_write = first_write_result_idx if first_write_result_idx is not None else len(transcript_entries)
+
+    # ===== G. 探索充分性 (35分) =====
+    g_items = {}
+    g1_pass = any(name == "Read" and "schema" in inp.lower()
+                  for idx, name, inp in tool_uses if idx < before_first_write)
+    g_items["G1_source_schema"] = 9 if g1_pass else 0
+
+    g2_pass = any(name == "Bash" and INPUT_TABLE_SHORT in inp
+                  and "SELECT" in inp.upper() and "LIMIT" in inp.upper()
+                  for idx, name, inp in tool_uses if idx < before_first_write)
+    g_items["G2_source_sample"] = 9 if g2_pass else 0
+
+    g3_pass = any(name == "Bash" and INPUT_TABLE_SHORT in inp
+                  and ("GROUP BY" in inp.upper() or "DISTINCT" in inp.upper() or "COUNT" in inp.upper())
+                  for idx, name, inp in tool_uses if idx < before_first_write)
+    g_items["G3_distribution"] = 9 if g3_pass else 0
+
+    g4_pass = any(name == "Bash" and ("DESCRIBE" in inp.upper() or "SHOW CREATE" in inp.upper())
+                  and OUTPUT_TABLE_SHORT in inp
+                  for idx, name, inp in tool_uses if idx < before_first_write)
+    g_items["G4_target_schema"] = 8 if g4_pass else 0
+
+    g_score = sum(g_items.values())
+    result["details"]["G_exploration"] = {"score": g_score, "max": 35, "items": g_items}
+
+    # ===== H. 执行效率 (40分) =====
+    h_items = {}
+    if write_result_count <= 2:
+        h_items["H1_few_submissions"] = 20
+    elif write_result_count <= 4:
+        h_items["H1_few_submissions"] = 13
+    elif write_result_count <= 6:
+        h_items["H1_few_submissions"] = 7
+    else:
+        h_items["H1_few_submissions"] = 0
+
+    if logic_error_retries == 0:
+        h_items["H2_no_logic_errors"] = 13
+    elif logic_error_retries <= 1:
+        h_items["H2_no_logic_errors"] = 7
+    else:
+        h_items["H2_no_logic_errors"] = 0
+
+    h_items["H3_no_redundancy"] = 7
+    h_score = sum(h_items.values())
+    result["details"]["H_efficiency"] = {"score": min(h_score, 40), "max": 40, "items": h_items}
+
+    # ===== I. 自验证行为 (25分) =====
+    i_items = {}
+    post_submit_uses = []
+    if last_spark_submit_success_idx is not None:
+        post_submit_uses = [(idx, name, inp) for idx, name, inp in tool_uses
+                            if idx > last_spark_submit_success_idx]
+
+    i1_pass = any(name == "Bash" and OUTPUT_TABLE_SHORT in inp and "SELECT" in inp.upper()
+                  for _, name, inp in post_submit_uses)
+    i_items["I1_query_output"] = 8 if i1_pass else 0
+
+    i2_pass = any(name == "Bash" and OUTPUT_TABLE_SHORT in inp
+                  and ("COUNT" in inp.upper() or "GROUP BY" in inp.upper())
+                  for _, name, inp in post_submit_uses)
+    i_items["I2_check_count"] = 9 if i2_pass else 0
+
+    i3_pass = any(name == "Bash" and OUTPUT_TABLE_SHORT in inp
+                  and ("LIMIT" in inp.upper() or "SELECT *" in inp.upper())
+                  for _, name, inp in post_submit_uses)
+    i_items["I3_check_values"] = 8 if i3_pass else 0
+
+    i_score = sum(i_items.values())
+    result["details"]["I_self_verification"] = {"score": i_score, "max": 25, "items": i_items}
+
+    return finalize(result)

tasks/offline-compute/HiveSQL/hivesql_006_en/gt/grade_spec.csv

@@ -0,0 +1,19 @@
+维度,维度全名,子维度,满分,说明
+A,A_executability,executability,15,result.sql 能跑通且产出非空
+B,B_schema,schema,10,列数(5) + 列名匹配(5)
+C,C_row_alignment,row_consistency,15,行数比例(7) + key覆盖率(8)
+D,D_field_value_match,field_value_match,25,非key字段逐列值匹配率
+D,D_field_completeness,field_completeness,15,关键字段非空/非空串比例
+F,F_insert_overwrite,insert_overwrite,5,INSERT OVERWRITE + PARTITION
+F,F_partition_value,partition_value,5,dt 分区值 = 20260507
+F,F_source_filter,source_filter,10,源表分区过滤 dt=20260507 + topic<>'test'
+
+总计,,,100,
+
+# 评分模板,评分范式: 产物100分 = A(15) + B(10) + C(15) + D(40) + F(20)
+# 难度,EASY
+# 权重,"product=0.5, process=0.5"
+# 范式,new
+# Key列,"city_id, dt"
+# 预期列数,10
+# 输出表,internal_platform_db.dwd_databus_topic_cost_detail_d_cand_query_engine_015

tasks/offline-compute/HiveSQL/hivesql_006_en/gt/ground_truth.sql

@@ -0,0 +1,17 @@
+INSERT overwrite TABLE internal_platform_db.dwd_databus_topic_cost_detail_d_query_engine_015 PARTITION (dt = '20260507')
+SELECT
+  MAX(systemname) AS system_belong,
+  MAX(dwproductname) AS category_name,
+  MAX(dwappgroup) AS dw_appgroup,
+  MAX(cityid) AS city_id,
+  MAX(iset) AS cluster_set,
+  topic,
+  MAX(data_size) AS total_data_size_d,
+  MAX(total_cost) AS total_cost,
+  MAX(in_charge) AS in_charge
+FROM
+  internal_platform_db.ods_t_databus_access_topic_cost_date_d_query_engine_015
+WHERE
+  dt = '20260507'
+  AND topic <> 'test'
+GROUP BY topic

tasks/offline-compute/HiveSQL/hivesql_006_en/init/init_db.py

@@ -0,0 +1,53 @@
+#!/usr/bin/env python3
+import re
+from pyspark.sql import SparkSession
+
+spark = SparkSession.builder \
+    .appName('hivesql_bench_init') \
+    .enableHiveSupport() \
+    .config('spark.sql.warehouse.dir', '/tmp/hive_warehouse') \
+    .getOrCreate()
+
+spark.sql('CREATE DATABASE IF NOT EXISTS internal_platform_db')
+
+
+def _execute_sql_file(spark, sql_path):
+    """Read SQL file, remove SuperSQL SET headers, split by semicolons, execute."""
+    with open(sql_path, 'r', encoding='utf-8') as f:
+        content = f.read()
+    content = re.sub(r'^\s*set\s+query_engine\.\S+\n?', '', content, flags=re.IGNORECASE)
+    stmts, cur, in_sq, in_dq, i = [], [], False, False, 0
+    while i < len(content):
+        ch = content[i]
+        if ch == '\\' and i + 1 < len(content):
+            cur.append(ch); cur.append(content[i + 1]); i += 2; continue
+        if ch == '-' and i + 1 < len(content) and content[i + 1] == '-' and not in_sq and not in_dq:
+            while i < len(content) and content[i] != '\n':
+                i += 1
+            cur.append('\n'); continue
+        if ch == "'" and not in_dq:
+            in_sq = not in_sq
+        elif ch == '"' and not in_sq:
+            in_dq = not in_dq
+        if ch == ';' and not in_sq and not in_dq:
+            s = ''.join(cur).strip()
+            if s:
+                stmts.append(s)
+            cur = []
+        else:
+            cur.append(ch)
+        i += 1
+    last = ''.join(cur).strip()
+    if last:
+        stmts.append(last)
+    for stmt in stmts:
+        spark.sql(stmt)
+
+
+_execute_sql_file(spark, '/tmp_workspace/init_db.sql')
+
+tables = spark.sql('SHOW TABLES IN internal_platform_db').collect()
+print(f'Init complete, {len(tables)} tables created')
+for t in tables:
+    print(f'  - {t.namespace}.{t.tableName}')
+spark.stop()

tasks/offline-compute/HiveSQL/hivesql_006_en/init/init_db.sql

@@ -0,0 +1,61 @@
+CREATE DATABASE IF NOT EXISTS internal_platform_db;
+
+-- 输入表
+CREATE TABLE IF NOT EXISTS internal_platform_db.ods_t_databus_access_topic_cost_date_d_query_engine_015 (
+    dt STRING COMMENT '天分区',
+    rep_date STRING COMMENT '日期',
+    systemclass STRING COMMENT '系统',
+    systemname STRING COMMENT '系统名',
+    odsproductid STRING COMMENT '产品ID',
+    odsproductname STRING COMMENT '产品名称',
+    dwproductid STRING COMMENT '数据仓库DW产品ID',
+    dwproductname STRING COMMENT '数据仓库DW产品名称',
+    dwappgroup STRING COMMENT '数据仓库DW应用组',
+    cityid STRING COMMENT '城市ID',
+    iset STRING COMMENT '集群',
+    topic STRING COMMENT 'topic',
+    data_count BIGINT COMMENT '条数',
+    data_pkgcnt BIGINT COMMENT '包数',
+    data_size BIGINT COMMENT '大小',
+    cost_count BIGINT COMMENT '核算数',
+    cost_per DOUBLE COMMENT '占比',
+    cost_tco DOUBLE COMMENT '设备成本',
+    cost_lan DOUBLE COMMENT '专线成本',
+    total_cost DOUBLE COMMENT '总成本',
+    in_charge STRING COMMENT '负责人'
+) STORED AS ORC;
+
+INSERT INTO TABLE internal_platform_db.ods_t_databus_access_topic_cost_date_d_query_engine_015 VALUES
+('20260507','20260507','classA','SystemAlpha','P001','ProdAlpha','T001','数据仓库DWProdA','AppGroupX','101','ClusterA','topic_billing',1000,50,2048000,800,0.35,120.5,30.2,150.7,'user_zhang'),
+('20260507','20260507','classA','SystemAlpha','P001','ProdAlpha','T001','数据仓库DWProdA','AppGroupX','101','ClusterA','topic_billing',1200,60,2560000,900,0.40,130.0,32.0,162.0,'user_zhang'),
+('20260507','20260507','classB','SystemBeta','P002','ProdBeta','T002','数据仓库DWProdB','AppGroupY','202','ClusterB','topic_log',500,25,1024000,400,0.20,60.0,15.0,75.0,'user_li'),
+('20260507','20260507','classB','SystemBeta','P002','ProdBeta','T002','数据仓库DWProdB','AppGroupY','202','ClusterB','topic_log',600,30,1200000,450,0.22,65.0,16.0,81.0,'user_li'),
+('20260507','20260507','classC','SystemGamma','P003','ProdGamma','T003','数据仓库DWProdC','AppGroupZ','303','ClusterC','topic_metrics',300,15,512000,200,0.10,40.0,10.0,50.0,'user_wang'),
+('20260507','20260507','classC','SystemGamma','P003','ProdGamma','T003','数据仓库DWProdC','AppGroupZ','303','ClusterC','test',10,1,1000,5,0.01,1.0,0.5,1.5,'user_test'),
+('20260506','20260506','classA','SystemAlpha','P001','ProdAlpha','T001','数据仓库DWProdA','AppGroupX','101','ClusterA','topic_billing',900,45,1800000,700,0.30,100.0,25.0,125.0,'user_zhang');
+
+-- 目标表
+CREATE TABLE IF NOT EXISTS internal_platform_db.dwd_databus_topic_cost_detail_d_query_engine_015 (
+    system_belong STRING COMMENT '系统名',
+    category_name STRING COMMENT '数据仓库DW产品名称',
+    dw_appgroup STRING COMMENT '数据仓库DW应用组',
+    city_id STRING COMMENT '城市ID',
+    cluster_set STRING COMMENT '集群',
+    topic STRING COMMENT 'topic',
+    total_data_size_d BIGINT COMMENT '大小',
+    total_cost DOUBLE COMMENT '总成本',
+    in_charge STRING COMMENT '负责人'
+) PARTITIONED BY (dt STRING COMMENT '天分区') STORED AS ORC;
+
+-- Agent 候选目标表
+CREATE TABLE IF NOT EXISTS internal_platform_db.dwd_databus_topic_cost_detail_d_cand_query_engine_015 (
+    system_belong STRING COMMENT '系统名',
+    category_name STRING COMMENT '数据仓库DW产品名称',
+    dw_appgroup STRING COMMENT '数据仓库DW应用组',
+    city_id STRING COMMENT '城市ID',
+    cluster_set STRING COMMENT '集群',
+    topic STRING COMMENT 'topic',
+    total_data_size_d BIGINT COMMENT '大小',
+    total_cost DOUBLE COMMENT '总成本',
+    in_charge STRING COMMENT '负责人'
+) PARTITIONED BY (dt STRING COMMENT '天分区') STORED AS ORC;

tasks/offline-compute/HiveSQL/hivesql_006_en/init/schema.sql

@@ -0,0 +1,23 @@
+CREATE TABLE IF NOT EXISTS internal_platform_db.ods_t_databus_access_topic_cost_date_d (
+    `dt` STRING COMMENT '天分区',
+    `rep_date` STRING COMMENT '日期',
+    `systemclass` STRING COMMENT '系统',
+    `systemname` STRING COMMENT '系统名',
+    `odsproductid` STRING COMMENT '产品ID',
+    `odsproductname` STRING COMMENT '产品名称',
+    `dwproductid` STRING COMMENT '数据仓库DW产品ID',
+    `dwproductname` STRING COMMENT '数据仓库DW产品名称',
+    `dwappgroup` STRING COMMENT '数据仓库DW应用组',
+    `cityid` STRING COMMENT '城市ID',
+    `iset` STRING COMMENT '集群',
+    `topic` STRING COMMENT 'topic',
+    `data_count` BIGINT COMMENT '条数',
+    `data_pkgcnt` BIGINT COMMENT '包数',
+    `data_size` BIGINT COMMENT '大小',
+    `cost_count` BIGINT COMMENT '核算数',
+    `cost_per` DOUBLE COMMENT '占比',
+    `cost_tco` DOUBLE COMMENT '设备成本',
+    `cost_lan` DOUBLE COMMENT '专线成本',
+    `total_cost` DOUBLE COMMENT '总成本',
+    `in_charge` STRING COMMENT '负责人'
+) STORED AS ORC;

tasks/offline-compute/HiveSQL/hivesql_006_en/task.md

@@ -0,0 +1,20 @@
+---
+id: offline-compute_HiveSQL_hivesql_006
+name: Input Table internal_platform_db.ods_t_databus_access_topic
+category: offline-compute/HiveSQL
+timeout_seconds: 600
+modality: pure-text
+engine: hivesql
+---
+## Prompt
+Task Objective: Cleanse the 数据总线 topic access cost raw data and aggregate it by topic dimension, excluding test topics, then write the results to a detail table.
+
+Input: `internal_platform_db.ods_t_databus_access_topic_cost_date_d_query_engine_015`, partitioned by field `dt` (STRING, YYYYMMDD). Key fields: `systemname`, `dwproductname`, `dwappgroup`, `cityid`, `iset`, `topic`, `data_size` (BIGINT), `total_cost` (DOUBLE), `in_charge`.
+
+Processing Rules: No joins, single-table processing. Filter condition: `dt = '20260507' AND topic <> 'test'`. Aggregation: Group by `topic`, applying `MAX` to `systemname`, `dwproductname`, `dwappgroup`, `cityid`, `iset`, `data_size`, `total_cost`, and `in_charge`. Derived column: `dt` directly takes the partition value `'20260507'`.
+
+Output Requirements: The output field order is `dt`, `system_belong`, `category_name`, `dw_appgroup`, `city_id`, `cluster_set`, `topic`, `total_data_size_d`, `total_cost`, `in_charge`. Field mappings: `dt=dt`, `system_belong=MAX(systemname)`, `category_name=MAX(dwproductname)`, `dw_appgroup=MAX(dwappgroup)`, `city_id=MAX(cityid)`, `cluster_set=MAX(iset)`, `topic=topic`, `total_data_size_d=MAX(data_size)`, `total_cost=MAX(total_cost)`, `in_charge=MAX(in_charge)`. Data types: `dt` STRING, `system_belong` STRING, `category_name` STRING, `dw_appgroup` STRING, `city_id` STRING, `cluster_set` STRING, `topic` STRING, `total_data_size_d` BIGINT, `total_cost` DOUBLE, `in_charge` STRING.
+
+Write Requirements: Target table `internal_platform_db.dwd_databus_topic_cost_detail_d_cand_query_engine_015`. Table comment: "数据总线 topic access cost detail table, including access cost and dimension information, performing data cleansing on the ODS layer." Partition field `dt` (STRING, YYYYMMDD). Write method: `INSERT OVERWRITE`, overwriting the partition `dt='20260507'`. If the target table does not exist, first create the table using standard Hive ORC storage format partitioned by `dt`, then write the data.
+
+Please write the final HiveSQL to `result.sql` and execute it.

tasks/offline-compute/HiveSQL/hivesql_007_en/gt/expected.csv

@@ -0,0 +1,4 @@
+trace_id,p_date,datawd_project_id,datawd_task_id,datawd_task_instance_id,compute_type,status_code,instance_run_time,code_run_time,resource_wait_time,code_start_time,code_end_time,instance_start_time,instance_end_time,serving_id,is_permanent,apply_for_gpu_count,code_gpu_count,instance_gpu_util,code_gpu_util,dt
+trace_001,20260507,proj_01,task_01,inst_01,GPU,0,120,100,20,2026-05-07 10:00:00,2026-05-07 10:01:40,2026-05-07 09:59:40,2026-05-07 10:01:40,srv_01,True,2,2.0,82.5,89.0,20260507
+trace_002,20260507,proj_02,task_02,inst_02,CPU,1,60,50,10,2026-05-07 11:00:00,2026-05-07 11:00:50,2026-05-07 11:00:00,2026-05-07 11:01:00,srv_02,False,0,4.0,0.0,,20260507
+trace_003,20260507,proj_03,task_03,inst_03,GPU,0,300,280,20,2026-05-07 12:00:00,2026-05-07 12:04:40,2026-05-07 11:59:40,2026-05-07 12:04:40,srv_03,True,4,4.0,95.0,92.0,20260507