synthesis-patterns
Compare original and translation side by side
🇺🇸
Original
English🇨🇳
Translation
ChineseMulti-Source Synthesis Patterns
多源信息合成模式
When answering questions that require information from multiple Glean sources, use these patterns to combine and synthesize effectively.
当回答需要从多个Glean来源获取信息的问题时，可采用以下模式来高效整合与合成结果。
When This Applies
适用场景
Use this approach when:
- A question spans documents, meetings, people, and code
- You need to cross-reference information from different sources
- Results from one source need context from another
- Building a comprehensive answer from partial information
- Sources seem to conflict or overlap
当出现以下情况时，可采用此方法：
- 问题涉及文档、会议、人员和代码等多个维度
- 需要交叉参考不同来源的信息
- 单一来源的结果需要结合其他来源的上下文才能理解
- 需要从零散信息中构建全面答案
- 不同来源的信息存在冲突或重叠
BE SKEPTICAL
保持质疑态度
When synthesizing across multiple sources, aggressive vetting is essential:
Source Quality Test
- Is each source authoritative?
- ✅ INCLUDE: Official docs, code, recent meetings
- ⚠️ CONTEXT: Secondary sources, somewhat dated
- ❌ EXCLUDE: Hearsay, speculation, very old information
Recency vs Authority
- When sources disagree on recency, choose wisely:
- ✅ TRUST: Recent information from official source
- ⚠️ CAUTION: Very recent but unofficial vs stale but official
- ❌ AVOID: Treating old info as current just because it's "official"
Completeness Check
- Do you have the full picture?
- ✅ INCLUDE: 3+ sources align, comprehensive coverage
- ⚠️ CONTEXT: Limited sources, note gaps explicitly
- ❌ EXCLUDE: Single source for multi-faceted questions
Conflict Resolution
- When sources conflict, don't hide it:
- ✅ INCLUDE: Conflict + explanation of which is likely correct
- ❌ EXCLUDE: Picking one source without acknowledging disagreement
Filter Out:
- Unattributed claims
- Information older than 3 months (unless structure/architecture)
- Synthesis that glosses over fundamental disagreements
- "I synthesized this" without showing your work
Say "I don't know" when:
- Sources are missing or conflicting
- Information is too stale to be reliable
- You lack sufficient data to synthesize confidently
在综合多源信息时，严格的审核至关重要：
来源质量检验
- 每个来源是否具有权威性？
- ✅ 纳入：官方文档、代码、近期会议内容
- ⚠️ 注意上下文：二手来源、略有过时的信息
- ❌ 排除：传闻、猜测、非常陈旧的信息
时效性与权威性的权衡
- 当不同来源的时效性存在差异时，需谨慎选择：
- ✅ 优先信任：官方来源的最新信息
- ⚠️ 谨慎对待：非官方的最新信息 vs 过时的官方信息
- ❌ 避免：仅因“官方”就将陈旧信息当作当前有效信息
完整性检查
- 是否掌握了完整的信息？
- ✅ 纳入：3个及以上来源信息一致、覆盖全面
- ⚠️ 注意上下文：来源有限时，需明确标注信息缺口
- ❌ 排除：仅用单一来源回答涉及多方面的问题
冲突解决
- 当来源信息冲突时，切勿隐瞒：
- ✅ 纳入：说明冲突情况，并解释哪一方更可能正确
- ❌ 排除：仅选择其中一个来源，却不提及存在分歧
需过滤的内容：
- 无明确来源的主张
- 超过3个月的信息（除非涉及架构/结构）
- 刻意掩盖根本性分歧的合成内容
- 仅声称“我已完成合成”却未展示依据的内容
在以下情况请直接说明“我不知道”：
- 相关来源缺失或存在冲突
- 信息过于陈旧，可靠性不足
- 缺乏足够数据来自信地完成合成
The Synthesis Framework
合成框架
1. Identify Information Types Needed
1. 确定所需信息类型
Before querying, determine what types of information are needed:
| Information Type | Glean Tool | Use When |
|---|---|---|
| Official documentation | | Need authoritative specs, policies |
| Real-time context | | Need recent discussions, decisions |
| People/org structure | | Need ownership, expertise |
| Code evidence | | Need implementation truth |
| User's own context | | Need personalization |
| AI synthesis | | Need reasoning across sources |
在发起查询前，先明确需要哪些类型的信息：
| 信息类型 | Glean工具 | 适用场景 |
|---|---|---|
| 官方文档 | | 需要权威规范、政策时 |
| 实时上下文 | | 需要了解近期讨论、决策时 |
| 人员/组织架构 | | 需要明确负责人、专业领域时 |
| 代码依据 | | 需要了解实际实现情况时 |
| 用户个人上下文 | | 需要个性化内容时 |
| AI合成 | | 需要跨来源推理时 |
2. Query in Optimal Order
2. 按最优顺序发起查询
Recommended order for comprehensive synthesis:
- Start broad with - Get AI-synthesized overview
chat - Verify with - Find authoritative documents
search - Add recency with - Get latest discussions
meeting_lookup - Clarify ownership with - Identify who to ask
employee_search - Ground in reality with - What's actually implemented
code_search
全面合成的推荐顺序：
- 先用进行宽泛查询 - 获取AI合成的概述内容
chat - 用验证 - 查找权威文档
search - 用补充时效性信息 - 获取最新讨论内容
meeting_lookup - 用明确归属 - 确定可咨询的人员
employee_search - 用锚定实际情况 - 了解真实的实现内容
code_search
3. Cross-Reference Pattern
3. 交叉参考模式
When you have results from multiple sources, cross-reference:
markdown
undefined当获取到多源结果时，采用以下方式交叉参考：
markdown
undefinedFinding: [Topic]
发现：[主题]
From Documentation
来自文档
- [What docs say] - Source: [doc title] ([link])
- [文档内容摘要] - 来源：[文档标题] ([link])
From Recent Meetings
来自近期会议
- [What was discussed] - Source: [meeting date]
- [会议讨论内容] - 来源：[会议日期]
From Code
来自代码
- [What's implemented] - Source: [file/repo]
- [代码实现情况] - 来源：[file/repo]
Synthesis
合成结论
Based on these sources: [Your synthesized answer]
Based on these sources: [Your synthesized answer]
Confidence
可信度评估
- Documentation: [Current/Stale/Missing]
- Meeting Context: [Recent/Old/None]
- Code Evidence: [Present/Partial/None]
undefined- 文档：[Current/Stale/Missing]
- 会议上下文：[Recent/Old/None]
- 代码依据：[Present/Partial/None]
undefined4. Handle Conflicts
4. 处理信息冲突
When sources disagree:
markdown
undefined当来源信息存在冲突时：
markdown
undefinedConflicting Information Detected
检测到信息冲突
| Topic | Source A Says | Source B Says |
|---|---|---|
| [Topic] | [Claim from doc] | [Claim from meeting] |
Most Likely Truth: [Your assessment based on recency, authority]
Recommendation: Verify with [person/source]
undefined| 主题 | Source A Says | Source B Says |
|---|---|---|
| [Topic] | [Claim from doc] | [Claim from meeting] |
Most Likely Truth: [Your assessment based on recency, authority]
Recommendation: Verify with [person/source]
undefinedSynthesis Patterns by Use Case
按使用场景分类的合成模式
Pattern: "What is X?"
模式："What is X?"
- - Get overview
chat - - Find official definition
search - - Find owner to verify
employee_search - Synthesize: Combine AI overview + official doc + owner context
- - 获取概述
chat - - 查找官方定义
search - - 找到负责人进行核实
employee_search - 合成：结合AI概述 + 官方文档 + 负责人上下文
Pattern: "Who should I talk to about X?"
模式："Who should I talk to about X?"
- - Find by role
employee_search - - Find by code activity
code_search - - Find by doc authorship
search - Synthesize: Rank by multiple signal strength
- - 按角色查找
employee_search - - 按代码贡献情况查找
code_search - - 按文档作者查找
search - 合成：根据多维度信号进行排序
Pattern: "What's the status of X?"
模式："What's the status of X?"
- - Get current status summary
chat - - Find recent discussions
meeting_lookup - - Find tracking docs (Jira, etc.)
search - Synthesize: Combine with recency weighting
- - 获取当前状态摘要
chat - - 查找近期讨论内容
meeting_lookup - - 查找跟踪文档（Jira, etc.）
search - 合成：结合时效性权重整合信息
Pattern: "How do we do X?"
模式："How do we do X?"
- - Find process docs
search - - Find implementation
code_search - - Find recent changes
meeting_lookup - Synthesize: Doc process + code reality + recent updates
- - 查找流程文档
search - - 查找实际实现
code_search - - 查找近期变更
meeting_lookup - 合成：文档流程 + 代码实际情况 + 最新更新
Output Best Practices
输出最佳实践
- Cite every source - Include links for all claims
- Note recency - When was this information last updated?
- Flag uncertainty - Be clear about gaps
- Suggest verification - Point to people who can confirm
- Acknowledge conflicts - Don't hide disagreements
- 标注所有来源 - 所有主张都需附带链接
- 注明时效性 - 信息最后更新时间是什么时候？
- 标注不确定性 - 明确说明信息缺口
- 建议核实方式 - 指出可确认信息的人员
- 承认冲突 - 切勿隐瞒分歧
Example Synthesis Output
合成输出示例
markdown
undefinedmarkdown
undefined[Question/Topic]
[Question/Topic]
Answer
Answer
[Direct answer to the question]
[Direct answer to the question]
Supporting Evidence
Supporting Evidence
| Source | What It Says | Last Updated |
|---|---|---|
| [Doc Title] | [Key info] | [Date] |
| [Meeting] | [Key info] | [Date] |
| [Code/Repo] | [Key info] | [Date] |
| Source | What It Says | Last Updated |
|---|---|---|
| [Doc Title] | [Key info] | [Date] |
| [Meeting] | [Key info] | [Date] |
| [Code/Repo] | [Key info] | [Date] |
Confidence Assessment
Confidence Assessment
- Overall Confidence: [High/Medium/Low]
- Data Freshness: [Current/Mostly current/Some stale]
- Source Agreement: [Strong/Partial/Conflicting]
- Overall Confidence: [High/Medium/Low]
- Data Freshness: [Current/Mostly current/Some stale]
- Source Agreement: [Strong/Partial/Conflicting]
Gaps & Recommendations
Gaps & Recommendations
- [What's missing or uncertain]
- [Who to verify with if critical]
undefined- [What's missing or uncertain]
- [Who to verify with if critical]
undefinedRelationship to Other Skills
与其他技能的关联
This skill works with:
- - For tool selection
glean-tools-guide - - For noting data quality
confidence-signals - - For document searches
enterprise-search - - For meeting queries
meeting-context - - For people queries
people-lookup
本技能可与以下技能配合使用：
- - 用于工具选择
glean-tools-guide - - 用于标注数据质量
confidence-signals - - 用于文档搜索
enterprise-search - - 用于会议查询
meeting-context - - 用于人员查询
people-lookup