| name | multi-source-search-filter |
| description | Search and filter the most relevant documents from a collection of 10-20 sources on a given topic. Use this skill when the task is to quickly scan a set of articles, papers, reports, or similar documents, score them for topical relevance, select the top 5 candidates, and rank those selected sources from most to least relevant. Focus on relevance-based selection using keyword matching, topic fit, specific data/examples, and source credibility. |
| metadata | {"environment":"research-information-synthesis","skill_id":"E5-LS1","short-description":"Search and filter relevant sources from a large collection of documents","version":"1.0"} |
Multi-Source Search Filter
Use this workflow when a user has 10-20 sources on a topic and needs the most relevant ones selected quickly. The goal is to do a fast first pass, score each document for relevance, choose the best 5, and then explain why those 5 matter.
Scope: relevance-based selection using keyword matching and credibility assessment.
When to Use
Use this skill when:
- a reading list or source set is too large to review in full
- the task is to narrow 10-20 documents down to the best few
- the user wants the most relevant sources for a target topic
- the output should include ranked selections and short relevance summaries
Selection Workflow
Follow these five steps.
Step 1 — Scan Titles and First Paragraphs
For each document, read:
- the title
- the first paragraph, abstract, or opening section
This is a quick relevance pass, not a full read.
Look for:
- direct keyword overlap with the target topic
- closely related terms or subtopics
- signs that the document will discuss the topic in substance
Example:
Target topic: employee retention strategies
Title: Reducing Turnover Through Better Onboarding
Opening: discusses onboarding outcomes and retention impact
Initial judgment: relevant
Step 2 — Score Each Document from 1 to 5
Assign a relevance score to each document.
Suggested scale:
5 = directly focused on the target topic and clearly useful
4 = strongly relevant but slightly narrower or broader than the topic
3 = partially relevant or useful as supporting context
2 = only loosely related
1 = not meaningfully relevant
Use these signals while scoring:
- keyword and topic match
- whether the document contains specific data or concrete examples
- whether the source is more or less credible
Credibility examples:
- academic journals
- established research reports
- professional publications
- blogs or informal posts
Example:
Document: "Reducing Turnover Through Better Onboarding"
Topic match: strong
Specific examples: yes
Source credibility: high
Score: 5
Step 3 — Select the Top 5 Documents
Sort the documents by score and keep the top 5.
If several documents tie, prefer:
- the one with stronger direct topic fit
- the one with more concrete data or examples
- the one from the more credible source
If fewer than 5 documents are meaningfully relevant, return fewer than 5 rather than padding the list with weak matches.
Step 4 — Read the Selected Documents Fully
For each selected document, read the full text and write a short 2-sentence summary.
The summary should explain:
- what the document is about
- why it is relevant to the target topic
Example:
This article examines how structured onboarding affects first-year employee retention.
It is relevant because it gives a concrete example of one retention strategy and the outcomes associated with it.
Step 5 — Rank the 5 Selected Documents
Present the selected documents from most to least relevant.
For each ranked source, include:
- title
- source
- score
- 2-sentence relevance summary
Also show how many documents were reviewed and how many were selected.
Basic Implementation Pattern
def score_document(title, opening, topic):
score = 1
if topic.lower() in title.lower() or topic.lower() in opening.lower():
score += 2
if has_specific_examples(opening):
score += 1
if high_credibility_source(title):
score += 1
return min(score, 5)
Practical Rules
Use Keyword Matching as the First Filter
The initial scan should quickly separate obviously relevant documents from weak matches.
Prefer Specific Data and Examples
Documents with concrete examples are usually more useful than very broad overviews.
Factor in Source Credibility
A highly relevant document from a strong source should usually outrank a similar document from a weaker source.
Read Only the Finalists in Full
Do not fully read all 10-20 documents at the start. Save the deeper read for the top selections.
Rank the Final Set Explicitly
The user should see not just which documents survived filtering, but also which of the selected ones are strongest.
Common Pitfalls
- reading every document fully before doing a quick relevance pass
- scoring without a consistent 1-5 rubric
- picking general overviews over stronger topic-specific sources
- ignoring source credibility when two documents seem similarly relevant
- selecting weak documents just to force the list to contain five items
When NOT to Use
- the task is to synthesize all sources rather than filter them
- there are too few documents to justify a ranking pass
- the user already knows which sources should be included
Quick Summary
1. Scan each title and first paragraph
2. Score each document from 1 to 5
3. Select the top 5 highest-scoring documents
4. Read those 5 fully and write 2-sentence relevance summaries
5. Rank the selected documents from most to least relevant