Multi-Source Research

Comprehensive multi-source investigation engine. Use when the task needs exhaustive research across ≥3 independent source categories, triangulated for accuracy.

Workflow

Phase 1: Research Plan

1. Define thesis — What question are we answering? What decisions ride on it?
2. Source mapping — Identify which source types are relevant:

Source Category	Tools	Use Case
Web (general)	web_search/Brave	Company info, news, pricing
Academic	web_search (arXiv, Semantic Scholar, CrossRef)	Papers, benchmarks
Social media	web_search (Reddit, X, LinkedIn)	Sentiment, chatter, user feedback
Official	web_fetch (gov.cn, CAC, company sites)	Policy, official announcements
Code/OSS	web_search + web_fetch (GitHub)	Open source, repos, community

3. Cross-reference matrix — For each claim, plan ≥2 independent sources.

Phase 2: First Pass — Surface Collection

Run parallel searches across target categories.
For each source type, collect:
- Title, URL, publication date
- Author/organization (source credibility assessment)
- Key claims (verbatim quotes where possible)

Phase 3: Second Pass — Deep Verification

For each claim marked as critical:

1. Find a second independent source
2. Compare dates — is one version more recent?
3. Rate consistency: ✅ Match | ⚠️ Partial | ❌ Contradiction
4. If contradiction exists, flag and investigate further

Phase 4: Synthesis

## Research Summary

### Key Findings
| Finding | Confidence | Sources | Verdict |
|---------|-----------|---------|---------|
| ... | 90% | [A] + [B] | CORROBORATED |
| ... | 60% | [C] only | NEEDS VERIFICATION |

### Source Credibility Ratings
| Source | URL | Credibility | Rationale |
|--------|-----|-------------|-----------|
| CAC.gov.cn | ... | A (98%) | Official government |
| TechCrunch | ... | A (93%) | Established tech media |
| Zhihu | ... | C (65%) | User-generated, needs verification |

### Notable Gaps
- What we couldn't find
- What remains uncertain
- Recommended next research direction

Cross-Validation Rules

1. Critical claims (P0/P1 relevance): must have ≥2 A-level or ≥3 B-level sources
2. Single-source claims: labeled as "unverified" with source credibility stated
3. Conflicting claims: surfaced explicitly with the analyst's judgment
4. Dated information: always check recency; label historical data

Output Format

### [Topic]

#### Verdict
[1-2 sentence judgment with confidence level]

#### Evidence
- [Source A]: [claim] (credibility: A, date)
- [Source B]: [corroborating/contradicting claim] (credibility: B, date)

#### Analyst Note
[What the data doesn't tell us, what to watch for]