pps-stats

Use this skill for 环节2 of a patent panorama project. It consumes a validated search_config.json, candidate_pool.csv and core_recall.csv from pps-search (环节1) and produces panorama statistics (trend, assignee landscape, technology constitution, competitor portraits), a branch-organized core patent index (default-trusting 环节1's recall ranking, with bounded fallback verification), and a value-signal cross-mining file. Everything aggregates directly from the query — no per-patent tagging is required.

知识产权
版本
v1.0.0
下载量
0
文件列表
49
最近更新
2026-07-02

关于

Use this skill for 环节2 of a patent panorama project. It consumes a validated search_config.json, candidate_pool.csv and core_recall.csv from pps-search (环节1) and produces panorama statistics (trend, assignee landscape, technology constitution, competitor portraits), a branch-organized core patent index (default-trusting 环节1's recall ranking, with bounded fallback verification), and a value-signal cross-mining file. Everything aggregates directly from the query — no per-patent tagging is required.

文件列表 49

pps-stats

SKILL.md

KeyValue
namepps-search
descriptionUse this skill for 环节1 of a patent panorama project. It builds an expert-grade, field-scoped, anchored and de-noised search configuration, validates per-branch precision by sampling, and exports a clean candidate pool plus per-branch queries — the input contract for 环节2 (pps-stats). Statistics, portraits and core-patent recall live in 环节2.

pps-search — Patent Panorama Search & Query-Construction Layer (环节1)

我是环节 1/4 · 检索建库（pps-search）。 我负责把业务问题翻译成专家级检索式、去噪、定候选池，产出 search_config.json + candidate_pool.csv + core_recall.csv + tech_taxonomy.txt，交给环节 2（pps-stats）做统计与价值挖掘。

Purpose

环节1 of the patent panorama pipeline. This skill answers only the query-quality questions:

  • Is the search query expert-grade — field-scoped operators, a constant topic anchor, semi-automatic IPC, tiered NOT with recorded reasons?

  • Which sub-technology branches structure the field, and what is the audited query for each (A6 four-part skeleton)?

  • Is the candidate pool clean enough — per-branch sampled precision ≥ 80% — to feed statistics and tagging downstream?

Output is a validated search configuration + a de-noised candidate pool + per-branch queries + a lightweight core-patent recall list + a tech taxonomy file for SaaS tagging — the input contract for 环节2 (pps-stats) .

Moved to 环节2 (pps-stats). Industry stats, assignee landscape, competitor portraits, core-patent verification/tiering, and value-signal cross-mining have moved out of this layer. 环节1 no longer emits a panorama report.

When To Use

  • User invokes /pps-search or a patent panorama project begins.

  • As the first step in the full pipeline before /pps-stats and /pps-tag.

  • Standalone: when the user only needs an audited, reusable search configuration.

Defaults

Dimension Default
Date basis Publication date ( pbdt) for market/legal views; earliest priority date ( E_PRIORITY_DATE) for technology-trend views
Technology stats counting Simple family level
Market / legal stats counting Publication text level
Geography CN, US, EP
Time range pbdt:[20200101 TO 20261231]
Analysis mode Competitor vs. Industry (Mode C)
Core patent signal Forward citation × family breadth × active legal status

Step 0: Initialize

Record all inputs in run_config.json.

Step 1: Expert Query Construction

Per Part A and Part D of references/query-and-taxonomy-methodology.md.

1-1 Keyword layering — strong / weak / short-word

1-2 Field-operator assignment (A1)

1-3 Constant topic anchor (A2)

1-4 Semi-automatic IPC (A4) — DO NOT FABRICATE IPC

1-5 Tiered NOT rules (A5) — every rule carries a recorded reason

1-6 Confirmation table — A6 four-part skeleton + precision spot-check

Step 2: Branch Query Generation

Fill A6 canonical skeleton per branch. Save to search_config.json.

Step 3: Candidate Pool Export

Export candidate_pool.csv： pn, branch_rule_hits only（族级去重）。

Step 4: Lightweight Core-Patent Recall

Per branch: refered_rank + famn_rank top 10。Output core_recall.csv： branch_id, patent_id, pn, recall_source, raw_rank。

Step 5: Tech Taxonomy Export【强制产出】

tech_taxonomy.txt 是环节1的强制产出文件 ，在候选池确认后、移交环节2前必须写出。供客户 SaaS 标引工具直接导入。

格式规范

层级型 ，每行一个叶节点：

plaintext
>一级\二级\三级

规则：

  • > 后紧跟第一级（Level-1 域名称）

  • \ 后依次为第二级、第三级

  • 每行只写一条路径（一个叶节点）

  • 每个单元格支持多值（SaaS 工具侧支持）

  • 不得添加任何注释、序号、空行分组、标题行或其他额外内容

  • 文件只包含层级链， 纯内容，无任何其他信息

示例（仅供格式参考）：

plaintext
>静态电压\稳定性
>静态电压\监测方法
>静态电压\PCMLE
>动态响应\瞬态抑制
>动态响应\环路补偿\Type-III补偿

写出路径： @session/pps-output/tech_taxonomy.txt。写出后提示用户「可直接下载上传至智慧芽标引工具」。

Output Files【强制产出清单】

File Written by Content 强制性
run_config.jsonStep 0 User inputs, defaults, analysis mode 强制
search_config.jsonStep 1–2 Keyword layers, anchor, IPC, NOT rules, per-branch A6 queries, precision 强制
candidate_pool.csvStep 3 pn, branch_rule_hits only，族级去重 强制
core_recall.csvStep 4 Per-branch raw recall: branch_id, patent_id, pn, recall_source, raw_rank，top ~10/branch 强制
tech_taxonomy.txtStep 5 层级链，每行 >L1\L2\L3，纯内容无注释，供 SaaS 标引工具直接导入 强制
report_manifest.json (partial) Step 3 Run metadata，环节2 extends 强制

Contract handoff to 环节2

五件套： search_config.json + candidate_pool.csv + core_recall.csv + tech_taxonomy.txt + report_manifest.json。

使用前配置

本 Skill 依赖智慧芽开放平台 MCP 服务：

  • 完成安装、初次使用时需进行自检，参见 README.md

  • 用户需完成账号授权，并确保 Agent 环境已启用对应 MCP 工具

  • 若未完成配置，本 Skill 只能提供分析框架，无法检索实时数据或生成基于数据库的结论

  • 缺少MCP配置时，引导用户参照 README.md 在 [ open.zhihuiya.com ]( https://open.zhihuiya.com/ ) 获取MCP。