这是数据层的 目录页:有哪些数据、从哪来、多新、怎么取。它讲注入机制——那是 grounding-layer 的事(见下"边界")。

The data layer’s catalog page: what data exists, where it’s from, how fresh, fetch order. It does not cover the injection mechanism — that’s grounding-layer’s job (see “Boundary” below).

边界:总览 vs grounding-layer / Boundary (the overlap you flagged)

中文。 两页不重叠,职责正交:

  • 本页(总览)= 数据目录 + 出处 + 取数顺序:有什么数据、从哪来、多新
  • grounding-layer = 注入机制:这些数据如何被拼成提示词、塞进场景与调查员(grounding_block / lookup_uni / 具体性契约)。
  • 另两个机制页:[investigator][yulai-integration]

一句话:总览回答 WHAT/WHERE,grounding-layer 回答 HOW

English. The two pages don’t overlap — orthogonal roles:

  • This page (Overview) = data catalog + provenance + fetch order: what data, where from, how fresh.
  • grounding-layer = the injection mechanism: how that data is assembled into prompts and fed to scenes/investigator (grounding_block / lookup_uni / the concreteness contract).
  • Two more mechanism pages: investigator (how the web tier is implemented), yulai-integration (how 雨来 connects).

In one line: Overview answers WHAT/WHERE, grounding-layer answers HOW.

数据目录 / The catalog

中文。

数据集内容规模来源页面
uni_employment.json就业率/升学率/出国率/起薪 + 行业/雇主2347 校(起薪 2140 / 就业率 1547)就业质量报告(权威受限蒸馏)data-employment
major_grounding.json专业深度知识(课程/去向/岗位/经验)856 专业(859 卡)蒸馏 + 事实/观点标注data-schools-majors
major_catalogue.json本科专业目录(按门类)全目录教育部专业目录data-schools-majors
school_index_sample.json985/211/双一流 + 录取批次2347 校公开层次/批次data-schools-majors
雨来 API录取分/等位分/专业判别分省分年雨来平台(权威,接入中)data-admissions

English.

datasetcontentsizesourcepage
uni_employment.json就业率/升学率/出国率/salary + industries/employers2347 schools (salary 2140 / employ 1547)就业质量报告 (authoritative-restricted distill)data-employment
major_grounding.jsonper-major deep knowledge (courses/paths/jobs/experience)859 majorsdistilled + fact/opinion labeleddata-schools-majors
major_catalogue.json本科专业目录 (by 门类)fullMoE major cataloguedata-schools-majors
school_index_sample.json985/211/双一流 + admission batch2347 schoolspublic tier/batchdata-schools-majors
雨来 API录取分/等位分/专业判别by province/year雨来 platform (authoritative, integrating)data-admissions

取数顺序与诚信规则 / Source tier & integrity rules

中文。 取数顺序:雨来(权威) → 本地 KB(grounding/索引) → 网络检索 → 定性。诚信规则:① 事实/观点分别标注;② 数字只采信明确覆盖该校该专业的权威来源;③ 缺失即留空(honest-null),绝不编造;④ 不把一个机构的数据安到另一个头上;⑤ 每条带 provenance/时间。

English. Fetch order: 雨来 (authoritative) → local KB (grounding/index) → web search → qualitative. Integrity rules: ① fact vs opinion labeled; ② numbers only from a source explicitly covering that school+major; ③ missing → null (honest-null), never fabricated; ④ never attribute one institution’s data to another; ⑤ each carries provenance/timestamp.

data-employment · data-schools-majors · data-admissions · grounding-layer · investigator · yulai-integration