add 3 atoms: surgical rewrite limits + booster ROI curve + dimension saturation
Browse files- 知识库/原子库/atoms.jsonl +3 -0
知识库/原子库/atoms.jsonl
CHANGED
|
@@ -39,3 +39,6 @@
|
|
| 39 |
{"id": "2026Q2_039", "knowledge": "Jekyll layout `<title>` 加 \" | Site Name\" 后缀 = 全站每页浪费 20 字符。43 篇 title >70c 中 20 篇仅靠移除后缀就达标。Layout-level fix = 1 commit, 20+ pages improved", "source": "2026-05-07 全站审计", "date": "2026-05-07", "topics": ["seo", "ctr", "jekyll", "layout"], "skills": ["gr-blog-post", "gr-seo-patrol"], "type": "lesson", "confidence": "high"}
|
| 40 |
{"id": "2026Q2_040", "knowledge": "SEO title 黄金区间 50-60 字符；70 字符是 Google SERP 硬截断阈值。Mobile SERP 更短（55 chars）。CTR 重写时要平衡 long-tail keyword + 长度，不能只为关键词堆词", "source": "JeffLi1993/seo-audit-skill 验证", "date": "2026-05-07", "topics": ["seo", "ctr", "title-length"], "skills": ["gr-blog-post"], "type": "principle", "confidence": "high"}
|
| 41 |
{"id": "2026Q2_041", "knowledge": "Script + LLM 双层架构 (JeffLi1993/seo-audit-skill 模式): Python scripts 输出 JSON envelope {status, detail, llm_review_required}, LLM 仅处理语义判断。可降低 hallucination 风险，提高一致性。HARD RULE whitelist 强制约束 LLM 输出范围", "source": "seo-audit-skill 设计模式", "date": "2026-05-07", "topics": ["skill-design", "architecture", "anti-hallucination"], "skills": ["gr-seo-patrol", "gr-geo-cite", "gr-blog-post"], "type": "method", "confidence": "high"}
|
|
|
|
|
|
|
|
|
|
|
|
| 39 |
{"id": "2026Q2_039", "knowledge": "Jekyll layout `<title>` 加 \" | Site Name\" 后缀 = 全站每页浪费 20 字符。43 篇 title >70c 中 20 篇仅靠移除后缀就达标。Layout-level fix = 1 commit, 20+ pages improved", "source": "2026-05-07 全站审计", "date": "2026-05-07", "topics": ["seo", "ctr", "jekyll", "layout"], "skills": ["gr-blog-post", "gr-seo-patrol"], "type": "lesson", "confidence": "high"}
|
| 40 |
{"id": "2026Q2_040", "knowledge": "SEO title 黄金区间 50-60 字符；70 字符是 Google SERP 硬截断阈值。Mobile SERP 更短（55 chars）。CTR 重写时要平衡 long-tail keyword + 长度，不能只为关键词堆词", "source": "JeffLi1993/seo-audit-skill 验证", "date": "2026-05-07", "topics": ["seo", "ctr", "title-length"], "skills": ["gr-blog-post"], "type": "principle", "confidence": "high"}
|
| 41 |
{"id": "2026Q2_041", "knowledge": "Script + LLM 双层架构 (JeffLi1993/seo-audit-skill 模式): Python scripts 输出 JSON envelope {status, detail, llm_review_required}, LLM 仅处理语义判断。可降低 hallucination 风险，提高一致性。HARD RULE whitelist 强制约束 LLM 输出范围", "source": "seo-audit-skill 设计模式", "date": "2026-05-07", "topics": ["skill-design", "architecture", "anti-hallucination"], "skills": ["gr-seo-patrol", "gr-geo-cite", "gr-blog-post"], "type": "method", "confidence": "high"}
|
| 42 |
+
{"id": "2026Q2_042", "knowledge": "Citability scorer 的页面分 = top 5 passages 平均。给已经强的 passage prepend 100+ 字方法论会撑过 134-167 词 sweet spot，self_containment 反而扣分 → 净 -0.6 到 -1.6 分。教训：要么写新的独立 passage（< 167 词），要么完全重构旧 passage（不是 prepend）", "source": "2026-05-07 实测 5 篇 surgical rewrites", "date": "2026-05-07", "topics": ["geo", "citability", "passage-optimization"], "skills": ["gr-geo-cite", "gr-blog-post"], "type": "lesson", "confidence": "high"}
|
| 43 |
+
{"id": "2026Q2_043", "knowledge": "Booster 章节（\"## Our Research Methodology\" / \"## Hard Numbers Only\" / \"## TL;DR\" 等）作为全新 passage 注入页首，**对薄内容文章效果最好**（Community directory 54.4 → 62.2，+7.8 = +14%）。对已经饱和的文章（PH master、社媒监听）几乎不动。规律：baseline 越低，booster ROI 越高", "source": "2026-05-07 5 篇实测", "date": "2026-05-07", "topics": ["geo", "citability", "booster-strategy"], "skills": ["gr-geo-cite", "gr-blog-post"], "type": "principle", "confidence": "high"}
|
| 44 |
+
{"id": "2026Q2_044", "knowledge": "Citability 维度饱和度排序（哪个最难刷分）：uniqueness_signals 10pts 最敏感（\"Our X-launch dataset\" + 具体产品名 + case study）。statistical_density 15pts 次之（% / $ / 年份 / 单位）。structural_readability 20pts 容易拉（短句 + 数字列表）。self_containment 25pts 受 word count 影响最重（必须 134-167 词）", "source": "zubair-trabzada/geo-seo-claude 评分体系 + 实测", "date": "2026-05-07", "topics": ["geo", "citability", "scoring-dimensions"], "skills": ["gr-geo-cite"], "type": "method", "confidence": "high"}
|