Verified Intelligence Collector

Overview

Collect, integrate, analyze, and report intelligence on any topic through a structured, multi-phase pipeline: 收集 (Collect) → 验证 (Verify) → 筛选 (Filter) → 整合 (Synthesize) → 分析 (Analyze) → 汇报 (Report). Every piece of information must pass through verification gates — unverified claims are explicitly flagged, never silently passed off as fact. Collected material is not merely listed; it is integrated into a coherent knowledge structure, deeply analyzed for patterns and implications, and delivered as a comprehensive report the user can act on.

Core Principles

1. No unverified claims presented as fact. If a claim has not been cross-verified, it is labeled with its verification status.
2. Source transparency. Every factual statement in the output must include its source(s) with URLs or references.
3. Confidence over volume. Better to deliver 5 verified findings than 25 unverified ones.
4. Bias awareness. Actively identify and disclose potential biases in sources.
5. Temporal awareness. Information has a shelf life — always note the publication date and whether more recent data may supersede it.
6. Synthesis over collection. The deliverable is an integrated, analyzed picture of the topic — not a raw dump of search results.
7. Analysis labeled as analysis. Opinions, inferences, and predictions derived by analysis are always marked as such, distinct from verified facts.
8. Actionable output. Every report ends with implications, options, or recommended next steps where the intelligence supports them.
9. Compliance and ethics. Intelligence gathering is bounded by law and professional ethics: no illegal acquisition, no misrepresentation, no disclosure of private personal data beyond the user's legitimate need.
10. Self-audit before delivery. Every deliverable passes a final quality gate: traceable citations, compliant labels, no silent assumptions, no unresolved contradictions.

Compliance and Ethics Guardrails (Read Before Collecting)

1. Legal boundaries. Do not suggest or perform any action that would violate law: no intrusion into private systems, no theft of trade secrets, no purchase of leaked/stolen data, no deception of individuals into disclosing confidential information (pretexting).
2. Public-source principle. Default to openly available public information: public web pages, published reports, official filings, news media, academic publications. Do not use covert or deceptive collection methods.
3. Personal data. When the topic involves individuals, collect only what is necessary for the stated purpose and already public; do not aggregate private data (addresses, phone numbers, financial records) beyond legitimate need. Follow applicable privacy law (e.g., PIPL).
4. Competitive intelligence ethics. Follow the SCIP code of ethics: do not misrepresent identity, do not induce breach of confidentiality, do not accept or use trade secrets, document the provenance of all information.
5. Sensitive topics. For politically sensitive or controversial topics, present balanced sourcing and refrain from advocacy. This skill reports; the user decides.
6. If a requested collection would cross a legal or ethical line — stop, explain why, and propose a compliant alternative. Do not silently proceed.

Workflow

Phase 1: Scope Definition

Before any search, define the collection scope:

1. Topic statement — Rephrase the user's request into a precise research question or objective.
2. Key assumptions check (initial) — List the assumptions the analysis will depend on (e.g., "market data reported by regulators is accurate", "the timeline of events is complete"). Explicitly mark these as assumptions to be challenged in Phase 7.
3. Scope boundaries — Identify what is in-scope and out-of-scope (time range, geography, industry, language).
4. Key entities — List specific names, organizations, products, events, or concepts to investigate.
5. Intelligence type — Classify the request:
   - Business/market intelligence (competitors, trends, market data)
   - Academic/technical research (papers, patents, technical documentation)
   - News/current affairs (events, policies, social trends)
   - General reference (definitions, explanations, background)
6. Analysis depth — Determine how deep the analysis must go:
   - Level 1: Collection + verification only (user asks for facts/materials)
   - Level 2: + Integration and organization (user asks for an organized picture)
   - Level 3: + Deep analysis (user asks for analysis, assessment, or recommendations)
7. Output format — Confirm with the user: full analysis report, quick briefing, alert/warning, data table, or file export (Markdown/PDF/Word). For file exports, use the appropriate document-generation tools after drafting.

Phase 2: Multi-Source Search Strategy

Execute a systematic search across diverse source categories:

1. Primary search — Use WebSearch with multiple query variations (Chinese and English keywords, entity names, alternative spellings).
2. Deep-dive fetch — Use WebFetch on the most promising results to extract detailed content.
3. Source diversity — Ensure coverage across at least 3 different source categories:
   - Official/government sources (.gov, official statistics, regulatory filings)
   - Established media (reuters.com, xinhuanet.com, AP, Bloomberg, etc.)
   - Academic/research (Google Scholar, arXiv, SSRN, CNKI abstracts)
   - Industry reports (consulting firms, trade associations)
   - Specialized databases (patent offices, financial filings)
4. Query iteration — Refine search queries based on initial findings. Follow citation chains and reference lists.
5. Historical context — Search conversation_search for any prior relevant discussions or user preferences.
6. Saturation check — Stop adding sources when new sources no longer yield new distinct facts (data saturation). Note when saturation is reached.
7. Maintain a source register — As sources are collected, log each in a running register: ID, URL, title, publication date, access date, tier, and status (collected/verified/used/discarded). The register feeds the final Source Assessment Summary. See references/verification-protocol.md § Source Register.

Phase 3: Source Credibility Assessment

For every source collected, assign a credibility tier. Load references/source-credibility-guide.md for the full evaluation criteria.

Tier 1 — High Credibility (Green)
- Government/official sources, regulatory bodies
- Peer-reviewed academic publications
- Tier-1 news organizations with editorial standards
- Official financial filings, regulatory disclosures
- Primary source documents (original legal documents, court records, regulatory orders)
- Organizational statements as self-reported claims (see note below)

Tier 2 — Medium Credibility (Yellow)
- Industry reports from reputable consulting/research firms
- Specialized trade media and professional publications
- Conference papers, preprints (not yet peer-reviewed)
- Established blogs with editorial oversight and named authors
- Crowdsourced platforms with moderation (e.g., Wikipedia — useful as a starting point, not as final evidence)

Tier 3 — Low Credibility (Red)
- Social media posts, anonymous sources
- Unverified blogs, content farms
- Marketing/promotional material, sponsored content
- AI-generated content of unknown provenance
- Single-source claims with no corroboration

Distinction — source tier vs. claim scope (important):
- An organization's own statement (press release, earnings call, product announcement) is a Tier-1-quality primary record of what that organization said. It is valid evidence for the claim "organization X stated Y."
- The SAME document is NOT valid evidence for external facts it asserts (e.g., "we have 100M users" proves the claim was made, but the user count itself needs third-party verification).
- Always apply this distinction: self-reports cite to establish "what was said," not "what is true."

Disqualifying factors (auto-downgrade by one tier):
- Source has a direct financial interest in the claim being true
- Source is anonymous or uses pseudonyms without editorial accountability
- Content shows a pattern of factual errors (not a single isolated error)
- Publication date is outdated for time-sensitive claims
- Source has been retracted or issued significant corrections on this topic

Phase 4: Cross-Verification

Load references/verification-protocol.md for the full protocol. Key requirements:

1. Minimum source count — Each key claim must be supported by at least 2 independent sources.
   - "Independent" means the sources do not cite each other and do not share a common origin (e.g., two articles both citing the same press release = 1 source, not 2).
2. Claim decomposition — Break complex claims into atomic factual statements and verify each independently.
3. Evidence-strength weighting — Not all "multiple sources" are equal. For academic/scientific claims, weight evidence by study quality (randomized trials > cohort studies > case reports > expert opinion) rather than by source count alone. See references/verification-protocol.md § Evidence Strength Hierarchy.
4. Contradiction handling — When sources disagree:
   - Present both positions with their supporting evidence
   - Assess which is more credible based on source tier, evidence quality, and logical consistency
   - Never silently pick one side
5. Fact vs opinion — Clearly distinguish between verifiable facts and opinions/commentary. Label opinions as such.
6. Temporal check — Verify that information is current. If a newer source contradicts an older one, note the evolution.
7. AI-content and misinformation detection — Check for signs of AI-generated content (generic phrasing, lack of specific details, circular reasoning, absence of primary sources). For suspected false information or rumors, follow the misinformation response procedure in references/analysis-methodology.md § Misinformation Response.

Phase 5: Information Filtering and Deduplication (筛选)

1. Deduplicate — Merge findings from multiple sources that report the same fact. Preserve the highest-tier source as primary citation.
2. Filter noise — Remove:
   - Marketing/promotional language presenting as fact
   - Outdated information superseded by newer data
   - Irrelevant tangents outside the defined scope
   - Duplicated content syndicated across multiple sites
3. Relevance scoring — Rank remaining findings by relevance to the original research question (High/Medium/Low).
4. Gap analysis — Identify what could not be verified or found. Explicitly state knowledge gaps rather than leaving them implicit.

Phase 6: Integration and Synthesis (整合与梳理)

Load references/analysis-methodology.md for the full framework. Transform the filtered findings from a list into a coherent knowledge structure:

1. Build a topic taxonomy — Organize findings into logical themes/categories (e.g., market, technology, regulation, competition, users). This becomes the skeleton of the report.
2. Construct a timeline — For event-driven intelligence, arrange verified events chronologically to reveal sequences and evolution.
3. Map relationships — Identify how findings relate: cause-and-effect, supporting/contradicting, part-whole, chronological, or analogous.
4. Identify entities and actors — Map the key players (companies, organizations, individuals) and their positions, interests, and interactions.
5. Connect the dots — Look for connections between seemingly separate findings; note emergent patterns.
6. Summarize each theme — For every theme in the taxonomy, produce an integrated summary that merges all verified sources into one coherent account (not a bullet-by-bullet listing).
7. Quality check — Ensure every statement in the integrated picture still traces to verified sources. Untraceable statements are re-flagged as unverified.

Phase 7: Deep Analysis (深度分析)

Load references/analysis-methodology.md for frameworks and selection guidance. Apply the analysis methods that fit the intelligence type and user's objective:

1. Pattern recognition — Identify recurring themes, trends, cycles, and anomalies across the integrated material.
2. Causal analysis — For key developments, analyze drivers and consequences: What caused it? What effects follow? Distinguish correlation from causation; note uncertainty.
3. Stakeholder analysis — Assess interests, positions, power, and likely moves of key actors (for competitive/business intelligence).
4. SWOT / PESTEL — For business/market topics, organize findings into Strengths, Weaknesses, Opportunities, Threats; or Political, Economic, Social, Technological, Environmental, Legal factors. Select the framework that fits; do not force both.
5. Scenario analysis — For forward-looking questions, develop 2-3 plausible scenarios (baseline, upside, downside) with the evidence supporting each.
6. Key assumptions check (KAC) — Review the assumptions listed in Phase 1. For each: is it still supported? Would the conclusion change if it were wrong? State the result explicitly. (Professional standard: ICD 203 requires explaining assumptions.)
7. Alternative analysis (ACH / devil's advocate) — For every main conclusion, deliberately construct the strongest alternative explanation that fits the same evidence (Analysis of Competing Hypotheses, or at minimum a devil's advocate pass). State why the alternative is less likely, or downgrade the conclusion if it is not. This is mandatory for Level 3 analysis on consequential questions. See references/analysis-methodology.md § Alternative Analysis.
8. Risk and opportunity assessment — For decision-oriented requests, enumerate key risks and opportunities with likelihood/impact ratings based on evidence.
9. Implications extraction — State what the findings mean for the user's stated objective or question.
10. Uncertainty disclosure — Explicitly list what analysis is based on inference rather than direct evidence, and how much the conclusion would change if key facts were wrong.

Analysis labeling rule: Every analytical statement (inference, judgment, prediction) MUST carry one of the standard labels defined in references/analysis-methodology.md § Labeling Rules ([Analysis], [Inference], [Prediction], [Estimate], [Scenario], [Implication], [Unverified]) so it is never confused with verified fact. Quantitative analysis must show its derivation (e.g., "computed from figures in sources [X], [Y]").

Phase 8: Confidence Rating

Assign a confidence rating to each key claim in the final output. Use the unified standard below (this is the single source of truth — it supersedes any partial table in the reference files):

Overall report confidence — Also assign an aggregate rating reflecting: breadth of coverage, share of confirmed vs. unverified claims, and stability of the underlying data. Report it as High / Medium / Low with a one-line justification.

Phase 9: Comprehensive Reporting (汇报)

Load references/output-templates.md for templates. Select the product type that matches the request:

• Full analysis report — default for Level 2/3 requests
• Quick briefing — time-sensitive or focused queries
• Alert/Warning — when a finding signals a developing risk, threat, or opportunity that the user should act on or monitor (format: what is happening, why now, what could happen next, trigger indicators, recommended actions)
• Data table — quantitative collections
• Verification card — single-claim fact-check
• File export — after drafting, convert to the user's requested format (Markdown/PDF/Word) using the appropriate document tools

Default structure for a full analysis report:

# [Topic] Intelligence Analysis Report

## Executive Summary
- 3-5 bullets: the most important verified findings + 1-2 key analytical insights
- Overall confidence assessment
- [Optional] Top recommendation at a glance

## Background / Context
- Brief orientation: what the topic is, why it matters, scope of this report

## Integrated Findings (organized by theme)
### Theme 1: [e.g., Market / Technology / Regulation ...]
#### Key Facts (verified)
- Fact 1 — [Confidence] — Source(s)
- Fact 2 — [Confidence] — Source(s)
#### Analysis
- [Analysis] Patterns, implications, assessment of this theme
### Theme 2: ...
...

## Cross-Theme Analysis
- How the themes interact; the big picture that emerges
- Scenario outlook / risk and opportunity assessment (as applicable)
- Alternative explanations considered (ACH results)

## Key Assumptions Check
- Assumption → status (supported / weakened / not supported) → impact if wrong

## Information Gaps
- What could not be verified or found, and why it matters

## Source Assessment Summary
| Source | Tier | Type | Key Contribution |
|--------|------|------|-----------------|

## Methodology Note
- Search date, query terms, source categories, saturation status
- Analysis frameworks applied
- Compliance note: collection used only public/open sources

## Appendix: Full Source List

Phase 10: Final Quality Gate (mandatory before delivery)

Before presenting any deliverable, run this checklist:

1. Citation integrity — Every factual statement has at least one traceable source (URL/reference). No orphan claims.
2. Label compliance — Every analytical statement carries a standard label. No unlabeled opinion disguised as fact.
3. Confidence coherence — Every claim's confidence rating matches the unified standard in Phase 8. No "Confirmed" label on single-source claims.
4. Contradiction audit — No unresolved contradictions silently buried. All disagreements are either resolved or disclosed.
5. Assumption audit — The key assumptions check has been run and its result is included in the output.
6. Alternative analysis audit — For Level 3 analysis, the ACH/devil's advocate pass has been done and at least one alternative explanation is acknowledged.
7. Gap disclosure — All information gaps are explicitly listed; nothing critical was silently assumed.
8. Compliance check — All sources are legal public sources; no personal data beyond need; no misrepresentation occurred during collection.
9. Fit to request — The output answers the user's actual question, in the agreed format and depth.

If any item fails, fix it before delivery. Do not deliver a report that fails the gate.

Special Handling

Breaking News / Emerging Events

• Information may be incomplete and rapidly changing
• Increase the frequency of temporal checks
• Explicitly label all findings as "Developing" with timestamps
• Prioritize official statements and on-scene reporting
• Consider producing an Alert/Warning if a developing risk is identified
• Analysis labels are mandatory — never present speculation as fact in fast-moving situations

Rumors and Disinformation

• When a claim seems to originate from social media or unverified channels, treat it as a rumor by default
• Attempt source tracing: find the earliest occurrence and its originator
• Check against official statements and fact-checking platforms where available
• If the rumor is confirmed false, state the refutation and its evidence; do not merely say "unverified"
• Follow the misinformation response procedure in references/analysis-methodology.md

Controversial or Politically Sensitive Topics

• Apply extra rigor to source diversity (include multiple perspectives)
• Explicitly disclose source biases and potential conflicts of interest
• Present contested claims with balanced sourcing
• Do not take a position; present evidence and let the user decide

Quantitative Data (Market Size, Financial Figures, Statistics)

• Always cite the original data source, not secondary reporting
• Note methodology differences that may explain discrepancies
• Flag estimates vs measured data; assess statistical quality (sample size, confidence intervals) where relevant
• Include date of data collection and reporting period
• When analyzing numbers, show the computation and its basis

Chinese-Language Sources

• For China-related topics, search in both Chinese and English
• Chinese official sources: government websites, Xinhua, People's Daily, CNR
• For financial data: CSRC filings, SSE/SZSE disclosures, NBS statistics
• Cross-reference Chinese and international reporting for sensitive topics
• Maintain a glossary of key terms in both languages to keep terminology consistent

Decision-Oriented Requests (e.g., "should we do X", "which option is better")

• Structure analysis around the decision: options → evidence for each → comparison → recommendation
• Present the recommendation as [Analysis] with explicit reasoning and caveats
• Include a "what would change our mind" note describing the evidence that would alter the conclusion

References

• references/source-credibility-guide.md — Detailed source credibility tier definitions, evaluation criteria, and bias detection checklist
• references/verification-protocol.md — Full cross-verification methodology, claim decomposition, source register, and evidence-strength hierarchy
• references/analysis-methodology.md — Integration/synthesis framework, analysis methods (pattern, causal, stakeholder, SWOT/PESTEL, scenario, ACH/devil's advocate, risk), labeling rules, and misinformation response
• references/output-templates.md — Output templates for full analysis reports, quick briefings, alerts/warnings, data tables, verification cards, and decision reports