[{"data":1,"prerenderedAt":307},["ShallowReactive",2],{"docs-doc-\u002Fdocs\u002Fmechanism\u002Fdata-flux":3,"docs-live-paths":256},{"id":4,"title":5,"body":6,"description":242,"extension":243,"meta":244,"navigation":251,"path":252,"seo":253,"stem":254,"__hash__":255},"content\u002Fdocs\u002Fmechanism\u002Fdata-flux.md","数据多源采集与国产化适配：衍流怎么看数据",{"type":7,"value":8,"toc":227},"minimark",[9,21,26,32,45,48,52,57,71,76,80,83,103,106,110,117,131,136,140,143,160,164,167,181,184,190,196,202,205],[10,11,12],"blockquote",{},[13,14,15,16,20],"p",{},"本文解决什么问题：正式选型时，决策者常问两件事——「我们已有老系统\u002F多套库，你们能把数据整合进来吗」和「信创要求用国产库，你们跑得了吗」。衍流是衍数回答这两问的方向。这篇讲它的",[17,18,19],"strong",{},"机制与边界","：多源数据怎么被接进来、口径怎么对齐、国产化怎么落地，以及哪些事它现在就能做、哪些是演进方向。",[22,23,25],"h2",{"id":24},"一句话机制把数据从哪来变成配置而不是迁移工程","一句话机制：把「数据从哪来」变成配置，而不是迁移工程",[13,27,28,29],{},"衍流的核心主张是：",[17,30,31],{},"平台运行在哪种数据库上、接哪些外部数据源，应该是配置层面的选择，而不是推翻系统的迁移工程。",[33,34,35,39,42],"ul",{},[36,37,38],"li",{},"界面（衍景）要展示的数、规则（衍枢）要计算的口径，最终都来自数据层；",[36,40,41],{},"数据层把「接哪套库、从哪采、按什么口径对齐」收敛为可配置项；",[36,43,44],{},"业务主库、分析读取源、历史\u002F外部数据源分开管理，各司其职。",[13,46,47],{},"企业的数据不必搬进一个\"大而全\"的库才用得上平台——多数业务数据在业务库内闭环，需要跨库\u002F跨系统时才走采集流转。",[22,49,51],{"id":50},"两个定位分开看","两个定位，分开看",[53,54,56],"h3",{"id":55},"平台跑在哪业务主库-vs-分析源","① 平台跑在哪：业务主库 vs 分析源",[33,58,59,65],{},[36,60,61,64],{},[17,62,63],{},"业务主库","：平台系统本身运行其上。默认支持通用关系数据库；国产化（信创）要求下，可接入国产数据库作为主库运行——这是部署期的选型配置，不是另行开发一套。",[36,66,67,70],{},[17,68,69],{},"分析读取源","：报表\u002F看板做大数据量分析时，可读取独立的分析数据源——它与业务写入解耦，只读、不作为业务主库。",[10,72,73],{},[13,74,75],{},"这样设计的好处：日常交易与写入稳定跑在业务主库上，重分析不拖累业务写入；两处按需配置，不互相绑架。",[53,77,79],{"id":78},"数据怎么流动多源采集与口径对齐","② 数据怎么流动：多源采集与口径对齐",[13,81,82],{},"企业数据很少只在一个库里。常见需要整合的场景：",[33,84,85,91,97],{},[36,86,87,90],{},[17,88,89],{},"采进来","：把外部系统的数据接进来（通过接口、文件，或直接连接其它数据库）；",[36,92,93,96],{},[17,94,95],{},"对齐口径","：清洗并统一口径后，进入业务可用的形态；",[36,98,99,102],{},[17,100,101],{},"输出","：按需提供给报表、看板或下游系统。",[13,104,105],{},"衍流的目标，是让「跨系统的数据整合」是一条可配置、可重复的路径，而不是每次手工搬数。",[22,107,109],{"id":108},"一个关键抓手ai-源头录入已落地","一个关键抓手：AI 源头录入（已落地）",[13,111,112,113,116],{},"「源头数据准」是口径一致的起点。衍数在这一层已落地的能力是 ",[17,114,115],{},"AI 源头录入","：",[33,118,119,125],{},[36,120,121,124],{},[17,122,123],{},"AI 表单代填 \u002F 辅助录单","：把表单中可自动填的部分由 AI 预填，减少人工重复录入；",[36,126,127,130],{},[17,128,129],{},"人工确认后入账","：AI 填的内容不是静默写入——由人确认后才正式入账，异常录入在源头即被拦截。",[10,132,133],{},[13,134,135],{},"它回答的是「数据进门那一关」：录入少出错、口径从源头统一，下游的报表与 AI 才可能算得对。",[22,137,139],{"id":138},"与口径体系的关系为什么先有口径才有多源","与口径体系的关系：为什么先有口径，才有多源",[13,141,142],{},"多源采集如果没有「同源口径」，接进来的越多越乱——同一个「销售额」在不同来源里各算各的，整合反而制造矛盾。",[13,144,145,146,149,150,155,156,159],{},"衍流的采集流向，最终都汇到",[17,147,148],{},"同一个口径体系","（",[151,152,154],"a",{"href":153},"\u002Fdocs\u002Fmechanism\u002Fmetrics-contract","指标口径","）：无论数据来自哪套库，落到报表、看板、AI 里时引用的是同一份定义。",[17,157,158],{},"「多源采集」与「口径一致」是一件事的两面","——采是为了让数据可用，口径是为了让数据可信。",[22,161,163],{"id":162},"边界哪些是方向哪些已是能力","边界：哪些是方向，哪些已是能力",[13,165,166],{},"对外沟通时我们保持透明：",[33,168,169,175],{},[36,170,171,174],{},[17,172,173],{},"已是能力\u002F主张","：多源接入的方向与路径、AI 源头录入（表单代填+人确认）、多库\u002F国产化作为配置项、业务主库与分析源分离；",[36,176,177,180],{},[17,178,179],{},"演进方向","：离线批量的深度数据管道仍在建设完善中。已落地的以实际版本为准，涉及具体部署选型请在评估阶段与交付方逐项确认。",[22,182,183],{"id":183},"常见疑问",[13,185,186,189],{},[17,187,188],{},"必须用国产库才能用衍数吗？","\n不用。默认即支持通用关系数据库，国产化是部署期按需的选型项——需要时再切换，不需要为\"换库\"推翻系统。",[13,191,192,195],{},[17,193,194],{},"我们已有老系统的数据，能接进来吗？","\n可以按多源接入的路径把外部数据采进来并对齐口径。具体对接方式（接口\u002F文件\u002F直连）与数据量，建议在实施评估阶段确认。",[13,197,198,201],{},[17,199,200],{},"所有业务都必须走数据采集\u002F清洗吗？","\n不是。多数业务数据在业务库内直接闭环；只有跨库、跨系统的整合才需要走采集流转——衍流是「需要时可用」的能力，不是每个模块的强制负担。",[22,203,204],{"id":204},"相关阅读",[33,206,207,214,220],{},[36,208,209,210],{},"数据最终长成什么界面 → ",[151,211,213],{"href":212},"\u002Fdocs\u002Fconcepts\u002Fview","衍景 · 配置驱动界面",[36,215,216,217],{},"口径如何定义与校验 → ",[151,218,219],{"href":153},"指标口径：定义与校验",[36,221,222,223],{},"部署在哪种数据库、怎么接入 → ",[151,224,226],{"href":225},"\u002Fdocs\u002Fdeploy\u002Foverview","私有化部署与环境要求",{"title":228,"searchDepth":229,"depth":229,"links":230},"",2,[231,232,237,238,239,240,241],{"id":24,"depth":229,"text":25},{"id":50,"depth":229,"text":51,"children":233},[234,236],{"id":55,"depth":235,"text":56},3,{"id":78,"depth":235,"text":79},{"id":108,"depth":229,"text":109},{"id":138,"depth":229,"text":139},{"id":162,"depth":229,"text":163},{"id":183,"depth":229,"text":183},{"id":204,"depth":229,"text":204},"企业的数据不在一个库里——衍流负责把它们接进来、对齐口径、供界面与 AI 使用。本文讲多源接入与国产化适配的机制与边界。","md",{"audience":245,"type":246,"section":247,"pr_order":248,"role":249,"duration":250},"public","mechanism","框架机制与边界",5,"实施搭建者 \u002F 决策者","约 5 分钟",true,"\u002Fdocs\u002Fmechanism\u002Fdata-flux",{"title":5,"description":242},"docs\u002Fmechanism\u002Fdata-flux","uR7DqW3652bfWN2M4XsD3l6Zko5XX3AtQX2ZtWygPoA",[257,259,261,263,265,267,268,270,271,273,275,277,279,281,283,285,287,289,291,293,294,295,297,299,301,303,305],{"path":258},"\u002Fdocs\u002Fconcepts\u002Fai",{"path":260},"\u002Fdocs\u002Fconcepts\u002Fai-team",{"path":262},"\u002Fdocs\u002Fconcepts\u002Fcore",{"path":264},"\u002Fdocs\u002Fconcepts\u002Fflux",{"path":266},"\u002Fdocs\u002Fconcepts\u002Fmetrics",{"path":212},{"path":269},"\u002Fdocs\u002Fdemo-center",{"path":225},{"path":272},"\u002Fdocs\u002Ffaq",{"path":274},"\u002Fdocs\u002Fglossary",{"path":276},"\u002Fdocs\u002Fguide\u002Fai-kb",{"path":278},"\u002Fdocs\u002Fguide\u002Fcrud",{"path":280},"\u002Fdocs\u002Fguide\u002Fdb-connect",{"path":282},"\u002Fdocs\u002Fguide\u002Fmodel-gen",{"path":284},"\u002Fdocs\u002Fintro\u002Fframeworks",{"path":286},"\u002Fdocs\u002Fintro\u002Foverview",{"path":288},"\u002Fdocs\u002Fintro\u002Fvalue",{"path":290},"\u002Fdocs\u002Fintro\u002Fwhy",{"path":292},"\u002Fdocs\u002Fmechanism\u002Fai-employee-task",{"path":252},{"path":153},{"path":296},"\u002Fdocs\u002Fmechanism\u002Fpermission",{"path":298},"\u002Fdocs\u002Fmechanism\u002Fpublish",{"path":300},"\u002Fdocs\u002Fquickstart\u002Fclone-template",{"path":302},"\u002Fdocs\u002Fquickstart\u002Fdemo-30s",{"path":304},"\u002Fdocs\u002Fquickstart\u002Ffive-minutes",{"path":306},"\u002Fdocs\u002Fsystem\u002Foverview",1788586879703]