code-ai-installer 4.0.0 → 4.0.1-a

package/domains/analytics/locales/en/.agents/skills/web-research/SKILL.md

@@ -1,309 +1,324 @@
----
-name: web-research
-description: Mandatory web research and data verification — search protocol, cross-checking, source audit
----
-# Web Research — Mandatory web search and data verification
-
-## When to Use
-- **Mandatory** for all Researcher gates (RES-xx) — no exceptions.
-- For **any claim** based on external data (market statistics, regulatory norms, competitive data).
-- When **verifying** data provided by the user — even if the user is confident in its accuracy.
-- When **outdated data is discovered** — updating and refreshing.
-
-> **Absolute rule:** A Researcher CANNOT complete a gate without executing `$web-research`. Absence of web search is an automatic **Blocker** during the `$gates` check.
-
-## Input
-
-| Field | Required | Description |
-|------|:-----------:|----------|
-| Research brief | ✅ | Research questions (from the Interviewer) |
-| Research topic | ✅ | Main topic / industry / company |
-| User-provided data | ⬚ | Data provided by the user for verification |
-| Search constraints | ⬚ | Geography, language, period, exclusions |
-| Minimum sources | ⬚ | Minimum number of sources per thesis (default is 3) |
-
-### Relationship with Other Skills
-| Skill | Interaction | When |
-|------|----------------|-------|
-| `$gates` (RES-xx) | Absence of web-research = automatic Blocker | During the gate check of each RES gate |
-| `$handoff` | Web-research results are handed off to the Data Analyst | After completing the research |
-| `tam-sam-som` | Market data: TAM, CAGR, market shares | When the Data Analyst applies TAM/SAM/SOM |
-| `competitive-analysis` | Competitor data: features, pricing, shares | When the Data Analyst applies competitive analysis |
-| `pest-analysis` | Macro data: regulation, economics, demographics | When the Data Analyst applies PEST/PESTEL |
-| `trend-analysis` | Trends: adoption rates, investments, forecasts | When the Data Analyst applies trend analysis |
-| `cohort-analysis` | Industry retention benchmarks | If there is no real cohort data |
-| `unit-economics` | Industry benchmarks for CAC, LTV, churn | If there are no real unit economics |
-| All framework skills | Any external data required by the framework | Always, when the framework references `$web-research` |
-
-> `$web-research` is a **service skill**: it serves all analytical skills by providing verified data.
-
-## Source Tiering
-
-| Tier | Source Type | Examples | Reliability | Usage |
-|:----:|---------------|---------|:----------:|---------------|
-| **T1** | Official reports, government statistics, peer-reviewed | Statista, Gartner, McKinsey, World Bank, SEC filings | High | Foundation for Verified. 1 T1 source = sufficient for Estimated |
-| **T2** | Industry media, analytics, corporate blogs | TechCrunch, CB Insights, company blogs | Medium | Confirmation for T1 or foundation for Estimated. 2× T2 = Verified |
-| **T3** | Forums, social networks, personal opinions, aggregators | Reddit, HN, Quora, expert blogs, Telegram channels | Low | Only for signals and hypotheses. Cannot be Verified. Maximum Assumed |
-
-**Rules:**
-- Verified = 1× T1 + 1× T1/T2, or 3× T2 with data matching.
-- Estimated = 1× T1, or 2× T2.
-- Assumed = T3, or indirect calculation, or extrapolation.
-- **T3 is never counted as Verified or Estimated.**
-
-## Data Confidence Levels
-
-| Level | Marker | Definition | Requirements | In the Report |
-|---------|:------:|-------------|------------|----------|
-| **Verified** | ✅ | Confirmed by 2+ independent sources (T1/T2) | URLs of both, dates, data match | Can be stated as fact |
-| **Estimated** | ⚠️ | One T1/T2 source or calculation | URL, methodology, caveat | "Estimated at..." |
-| **Assumed** | 🔮 | Assumption, indirect data, T3 | Rationale, indirect sources | "Presumably..." |
-
-## Data Currency Requirements
-
-| Data Type | Maximum Age | Action When Outdated |
-|-----------|:--------------------:|--------------------------|
-| Market statistics | 12 months | Search for fresh data or mark as ⚠️ Estimated |
-| Regulatory / legal | Current version | Must verify currency |
-| Company financials | Latest reporting period | Indicate the reporting period |
-| Technology trends | 6 months | Search for the most current |
-| Demographic data | 24 months | Acceptable if no fresher data exists |
-
-## Protocol
-
-### Step 0 — Search Planning
-1. Study the research brief — identify key questions.
-2. For each question, define at least 3 search queries.
-3. Include the year in queries to ensure currency.
-4. Plan queries for counter-evidence.
-5. Define the priority of sources: T1 → T2 → T3.
-
-### Step 1 — Execute Search
-1. Execute search queries via WebSearch.
-2. For each result: extract data via WebFetch.
-3. Record: URL, publication date, author / organization, **Source Tier**.
-4. Minimum 3 queries per data point.
-
-### Step 2 — Data Extraction and Structuring
-1. From each source, extract specific data (numbers, facts, quotes).
-2. Link data to questions from the research brief.
-3. Assign a confidence level (Verified / Estimated / Assumed) based on tiering rules.
-4. Timestamp each data point.
-
-### Step 3 — Cross-Checking
-1. For each key data point — find confirmation in 2+ sources.
-2. When data conflicts: record both values, cite sources, flag as a conflict.
-3. On conflict — attempt to find a third source for resolution.
-4. If the conflict is unresolved — pass both versions with a "Conflict: [value A] vs [value B]" notation.
-
-### Step 4 — Counter-Evidence Search
-1. For each key thesis — deliberately search for refutations.
-2. Formulate queries like: "[thesis] criticism", "[thesis] disadvantages", "[thesis] risks".
-3. Record found counter-arguments.
-4. If no counter-evidence is found — record this as "counter-position not found".
-
-### Step 5 — Audit Trail Formation
-1. Record all search queries used.
-2. Record all URLs visited (even if data was not used).
-3. Record reasons for rejecting sources (outdated, T3, irrelevant, conflicts).
-4. Form the final source table.
-
-### Step 6 — Research Quality Assessment
-Calculate quality metrics:
-
-| Metric | Formula | Value | Threshold |
-|---------|---------|:--------:|:-----:|
-| **Verified %** | Verified / Total × 100 | XX% | ≥ 50% — good |
-| **Coverage** | Questions covered / Total questions × 100 | XX% | 100% — mandatory |
-| **Source diversity** | Unique sources / Data points | X.X | ≥ 1.5 — good |
-| **Counter-evidence rate**| Theses with counter-search / Total theses × 100 | XX% | 100% — mandatory |
-| **Conflict rate** | Unresolved conflicts / Total data points × 100 | XX% | ≤ 10% — acceptable |
-
-**Research Quality:**
-
-| Verified % | Assessment | Action |
-|:----------:|--------|----------|
-| ≥ 70% | 🟢 High | Ready to hand off to the Data Analyst |
-| 50-69% | 🟡 Medium | Hand off with a caveat, recommend additional search |
-| < 50% | 🔴 Low | Rework: find more T1/T2 sources |
-
-### Step 7 — Results Compilation
-1. Structure data by research brief questions.
-2. For each question: facts → sources (with Tier) → confidence level → counter-position.
-3. Highlight conflicts and unresolved questions.
-4. Attach quality metrics.
-5. Formulate conclusions for handoff to the Data Analyst.
-
-## Search Query Patterns
-
-### Mandatory Patterns
-| Pattern | Example | Purpose |
-|---------|--------|------|
-| `[topic] [year]` | "SaaS market size 2025" | Current data |
-| `[topic] statistics report` | "e-commerce Russia statistics report" | Official reports (T1) |
-| `[topic] criticism risks` | "remote work criticism risks" | Counter-evidence |
-| `[company] revenue financial results` | "Skillbox revenue 2025" | Financial data |
-| `[topic] vs [alternative]` | "Bitrix24 vs amoCRM 2025" | Comparative data |
-| `site:statista.com [topic]` | `site:statista.com cloud computing` | T1 sources |
-| `[topic] research [industry]` | "retention benchmarks SaaS B2B" | Industry benchmarks |
-
-### Prohibited Patterns
-| Pattern | Why | Alternative |
-|---------|--------|-------------|
-| One general query for the entire topic | Superficial data | Minimum 3 queries per data point |
-| Query without a year | Outdated results | Always include the year |
-| Only confirming queries | Confirmation bias | Must search for counter-evidence |
-| Copying data without verification | Unreliable data | Cross-check with 2+ sources |
-| Only T3 sources | Low reliability | Priority: T1 → T2 → T3 as a supplement |
-
-## Example — Web Research for "EdTech Market Russia, Corporate Training"
-
-**Research brief (from INT-01):** 4 of 12 questions (fragment).
-
-### Question 1: What is the size of the Russian EdTech market (TAM)?
-
-| # | Fact | Source | Tier | URL | Date | Confidence |
-|---|------|----------|:----:|-----|------|:-------------:|
-| 1 | Russian EdTech = $4.2B (2026) | Smart Ranking | T1 | smart-ranking.ru/... | 2026-01 | ✅ Verified |
-| 2 | Russian EdTech = $3.9-4.5B (2026) | HSE, ISSEK | T1 | hse.ru/... | 2025-11 | ✅ Verified |
-| 3 | Corp. training = 43% of EdTech | HSE | T1 | hse.ru/... | 2025-11 | ⚠️ Estimated |
-
-**Counter-position:** Smart Ranking notes that 30% of the market is the shadow sector (private tutors), which may inflate TAM. Source: smart-ranking.ru/methodology.
-**Conflicts:** None (a spread of $3.9-4.5B is normal for different methodologies).
-
-### Question 3: What technological trends are reshaping the market?
-
-| # | Fact | Source | Tier | URL | Date | Confidence |
-|---|------|----------|:----:|-----|------|:-------------:|
-| 1 | AI personalization: 12% adoption in EdTech| Gartner | T1 | gartner.com/... | 2025-09 | ✅ Verified |
-| 2 | AI personalization: Duolingo +40% retention | TechCrunch | T2 | techcrunch.com/... | 2025-07 | ✅ Verified (with #1) |
-| 3 | LLM reduces course cost ×5 | a16z | T2 | a16z.com/... | 2026-02 | ⚠️ Estimated |
-| 4 | "AI in learning is overhyped for the corp. sector" | HBR | T1 | hbr.org/... | 2025-10 | ✅ Counter-position |
-
-**Counter-position:** HBR argues that corporate training requires human-in-the-loop and AI-only solutions do not work for soft skills.
-
-### Quality Metrics
-
-| Metric | Value | Threshold | Assessment |
-|---------|:--------:|:-----:|:------:|
-| Verified % | 62% (18/29) | ≥ 50% | 🟡 Medium |
-| Coverage | 100% (12/12 questions) | 100% | 🟢 |
-| Source diversity | 1.7 (48 sources / 29 points) | ≥ 1.5 | 🟢 |
-| Counter-evidence rate| 100% (12/12 theses) | 100% | 🟢 |
-| Conflict rate | 7% (2/29) | ≤ 10% | 🟢 |
-
-**Quality: 🟡 Medium** (Verified 62% — hand off with a caveat, recommend additional T1 search for ⚠️ data).
-
-### Audit Trail (fragment)
-
-| # | Query | Results | Used | Rejected |
-|---|--------|:-----------:|:------------:|:---------:|
-| 1 | "EdTech market Russia 2025 2026" | 23 | 4 | 19 (outdated: 8, irrelevant: 11) |
-| 2 | "corporate training market Russia report" | 15 | 3 | 12 (T3: 5, duplicates: 4, outdated: 3) |
-| 3 | "EdTech Russia criticism risks 2025" | 9 | 2 | 7 (irrelevant) |
-
-**Total:** 38 queries, 112 results, 48 sources used, 64 rejected.
-
-## Validation (Quality Gate)
-
-Web research is considered complete if:
-
-- [ ] All research brief questions are covered by data (Coverage = 100%)
-- [ ] WebSearch / WebFetch are used (not "from memory")
-- [ ] Each data point has: URL, date, source Tier, confidence level
-- [ ] Cross-checking is performed (2+ sources per key data point)
-- [ ] Counter-evidence is deliberately searched for each thesis (Counter-evidence = 100%)
-- [ ] Data is current per the requirements table
-- [ ] Audit trail is completed (all queries, URLs, rejection reasons)
-- [ ] Conflicting data is flagged and explained (Conflict rate ≤ 10%)
-- [ ] Minimum number of sources is met (default ≥ 3 per thesis)
-- [ ] Verified % ≥ 50% (if < 50% — rework or hand off with a 🔴 caveat)
-- [ ] T3 sources are not used as Verified or Estimated
-
-## Handoff
-The `$web-research` result is input for:
-- **Data Analyst (AN-xx)** — verified data for applying frameworks.
-- **All framework skills** — via the Data Analyst: data feeds into TAM/SAM/SOM, Competitive Analysis, PEST, SWOT, etc.
-- **`$gates` (RES-xx)** — the gate check verifies the presence and quality of the web-research.
-
-Handoff format: a structured report (based on the research brief questions) + source table (with Tier and confidence) + quality metrics + audit trail. When handing off — use `$handoff`.
-
-## Anti-patterns
-
-| Mistake | Why It's Bad | How to Do It Right |
-|--------|-------------|---------------|
-| Data "from memory" without search | May be outdated or inaccurate | Always execute WebSearch |
-| One source = "fact" | No cross-checking | 2+ sources (T1/T2) for key data |
-| No counter-evidence | Confirmation bias, one-sided analysis | Deliberately search for refutations (100%) |
-| URL without date and Tier | Impossible to assess currency and reliability | Always: URL + date + Tier (T1/T2/T3) |
-| All data marked Verified | Unrealistic, hides uncertainty | Honestly: Verified / Estimated / Assumed |
-| Queries without a year | Outdated results | Include the current year in the query |
-| Skipping the audit trail | Impossible to verify research quality | Record all queries, URLs, rejection reasons |
-| Only T3 sources | Low reliability, cannot be Verified | Priority T1 → T2; T3 only as a signal |
-| Verified < 50% without rework | Majority of data is unreliable | Rework: find T1/T2 or hand off with a 🔴 caveat |
-
-## Output Template
-
-```markdown
-### Web Research Report — [Topic]
-
-**Researcher:** [RES-xx]
-**Date:** [YYYY-MM-DD]
-**Research brief:** [N questions from INT-01]
-
----
-
-#### Quality Metrics
-
-| Metric | Value | Threshold | Assessment |
-|---------|:--------:|:-----:|:------:|
-| Verified % | XX% | ≥ 50% | 🟢/🟡/🔴 |
-| Coverage | XX% | 100% | 🟢/🔴 |
-| Source diversity | X.X | ≥ 1.5 | 🟢/🟡 |
-| Counter-evidence rate| XX% | 100% | 🟢/🔴 |
-| Conflict rate | XX% | ≤ 10% | 🟢/🟡 |
-
-**Research Quality:** 🟢 High / 🟡 Medium / 🔴 Low
-
----
-
-#### Results by Question
-
-##### Question 1: [Formulation from research brief]
-
-| # | Fact / Data | Source | Tier | URL | Date | Confidence |
-|---|--------------|----------|:----:|-----|------|:-------------:|
-| 1 | [Value] | [Name] | T1/T2/T3 | [URL] | [YYYY-MM] | ✅ / ⚠️ / 🔮 |
-
-**Counter-position:** [Found counter-arguments or "None found"]
-**Conflicts:** [Discrepancies or "None"]
-
-##### Question 2: [Formulation]
-[same structure]
-
----
-
-#### Confidence Summary
-
-| Level | Count | Share |
-|---------|:------:|:----:|
-| ✅ Verified | [N] | XX% |
-| ⚠️ Estimated | [N] | XX% |
-| 🔮 Assumed | [N] | XX% |
-
-#### Unresolved Conflicts
-
-| # | Data A | Source A (Tier) | Data B | Source B (Tier) | Comment |
-|---|---------|:-----------------:|---------|:-----------------:|-------------|
-| 1 | [Data] | [Source] | [Data] | [Source] | [Comment] |
-
----
-
-#### Audit Trail
-
-| # | Query | Results | Used | Rejected (reason) |
-|---|--------|:-----------:|:------------:|---------------------|
-| 1 | [query] | [N] | [N] | [N] ([reasons]) |
-
-**Total:** [N] queries, [N] results, [N] used, [N] rejected
-```
+---
+name: web-research
+description: Mandatory web research and data verification — search protocol, cross-checking, source audit
+type: mandatory
+domain: analytics
+owners:
+  - researcher
+gates:
+  - RESEARCHER
+tech: []
+topic: []
+triggers:
+  - "web-research"
+  - "Веб-поиск и верификация данных"
+  - "Обязательный веб-поиск и верификация данных — протокол поиска, перекрёстная проверка, аудит источников"
+related: []
+budget_lines: 324
+schema_version: 1
+---
+# Web Research — Mandatory web search and data verification
+
+## When to Use
+- **Mandatory** for all Researcher gates (RES-xx) — no exceptions.
+- For **any claim** based on external data (market statistics, regulatory norms, competitive data).
+- When **verifying** data provided by the user — even if the user is confident in its accuracy.
+- When **outdated data is discovered** — updating and refreshing.
+
+> **Absolute rule:** A Researcher CANNOT complete a gate without executing `$web-research`. Absence of web search is an automatic **Blocker** during the `$gates` check.
+
+## Input
+
+| Field | Required | Description |
+|------|:-----------:|----------|
+| Research brief | ✅ | Research questions (from the Interviewer) |
+| Research topic | ✅ | Main topic / industry / company |
+| User-provided data | ⬚ | Data provided by the user for verification |
+| Search constraints | ⬚ | Geography, language, period, exclusions |
+| Minimum sources | ⬚ | Minimum number of sources per thesis (default is 3) |
+
+### Relationship with Other Skills
+| Skill | Interaction | When |
+|------|----------------|-------|
+| `$gates` (RES-xx) | Absence of web-research = automatic Blocker | During the gate check of each RES gate |
+| `$handoff` | Web-research results are handed off to the Data Analyst | After completing the research |
+| `tam-sam-som` | Market data: TAM, CAGR, market shares | When the Data Analyst applies TAM/SAM/SOM |
+| `competitive-analysis` | Competitor data: features, pricing, shares | When the Data Analyst applies competitive analysis |
+| `pest-analysis` | Macro data: regulation, economics, demographics | When the Data Analyst applies PEST/PESTEL |
+| `trend-analysis` | Trends: adoption rates, investments, forecasts | When the Data Analyst applies trend analysis |
+| `cohort-analysis` | Industry retention benchmarks | If there is no real cohort data |
+| `unit-economics` | Industry benchmarks for CAC, LTV, churn | If there are no real unit economics |
+| All framework skills | Any external data required by the framework | Always, when the framework references `$web-research` |
+
+> `$web-research` is a **service skill**: it serves all analytical skills by providing verified data.
+
+## Source Tiering
+
+| Tier | Source Type | Examples | Reliability | Usage |
+|:----:|---------------|---------|:----------:|---------------|
+| **T1** | Official reports, government statistics, peer-reviewed | Statista, Gartner, McKinsey, World Bank, SEC filings | High | Foundation for Verified. 1 T1 source = sufficient for Estimated |
+| **T2** | Industry media, analytics, corporate blogs | TechCrunch, CB Insights, company blogs | Medium | Confirmation for T1 or foundation for Estimated. 2× T2 = Verified |
+| **T3** | Forums, social networks, personal opinions, aggregators | Reddit, HN, Quora, expert blogs, Telegram channels | Low | Only for signals and hypotheses. Cannot be Verified. Maximum Assumed |
+
+**Rules:**
+- Verified = 1× T1 + 1× T1/T2, or 3× T2 with data matching.
+- Estimated = 1× T1, or 2× T2.
+- Assumed = T3, or indirect calculation, or extrapolation.
+- **T3 is never counted as Verified or Estimated.**
+
+## Data Confidence Levels
+
+| Level | Marker | Definition | Requirements | In the Report |
+|---------|:------:|-------------|------------|----------|
+| **Verified** | ✅ | Confirmed by 2+ independent sources (T1/T2) | URLs of both, dates, data match | Can be stated as fact |
+| **Estimated** | ⚠️ | One T1/T2 source or calculation | URL, methodology, caveat | "Estimated at..." |
+| **Assumed** | 🔮 | Assumption, indirect data, T3 | Rationale, indirect sources | "Presumably..." |
+
+## Data Currency Requirements
+
+| Data Type | Maximum Age | Action When Outdated |
+|-----------|:--------------------:|--------------------------|
+| Market statistics | 12 months | Search for fresh data or mark as ⚠️ Estimated |
+| Regulatory / legal | Current version | Must verify currency |
+| Company financials | Latest reporting period | Indicate the reporting period |
+| Technology trends | 6 months | Search for the most current |
+| Demographic data | 24 months | Acceptable if no fresher data exists |
+
+## Protocol
+
+### Step 0 — Search Planning
+1. Study the research brief — identify key questions.
+2. For each question, define at least 3 search queries.
+3. Include the year in queries to ensure currency.
+4. Plan queries for counter-evidence.
+5. Define the priority of sources: T1 → T2 → T3.
+
+### Step 1 — Execute Search
+1. Execute search queries via WebSearch.
+2. For each result: extract data via WebFetch.
+3. Record: URL, publication date, author / organization, **Source Tier**.
+4. Minimum 3 queries per data point.
+
+### Step 2 — Data Extraction and Structuring
+1. From each source, extract specific data (numbers, facts, quotes).
+2. Link data to questions from the research brief.
+3. Assign a confidence level (Verified / Estimated / Assumed) based on tiering rules.
+4. Timestamp each data point.
+
+### Step 3 — Cross-Checking
+1. For each key data point — find confirmation in 2+ sources.
+2. When data conflicts: record both values, cite sources, flag as a conflict.
+3. On conflict — attempt to find a third source for resolution.
+4. If the conflict is unresolved — pass both versions with a "Conflict: [value A] vs [value B]" notation.
+
+### Step 4 — Counter-Evidence Search
+1. For each key thesis — deliberately search for refutations.
+2. Formulate queries like: "[thesis] criticism", "[thesis] disadvantages", "[thesis] risks".
+3. Record found counter-arguments.
+4. If no counter-evidence is found — record this as "counter-position not found".
+
+### Step 5 — Audit Trail Formation
+1. Record all search queries used.
+2. Record all URLs visited (even if data was not used).
+3. Record reasons for rejecting sources (outdated, T3, irrelevant, conflicts).
+4. Form the final source table.
+
+### Step 6 — Research Quality Assessment
+Calculate quality metrics:
+
+| Metric | Formula | Value | Threshold |
+|---------|---------|:--------:|:-----:|
+| **Verified %** | Verified / Total × 100 | XX% | ≥ 50% — good |
+| **Coverage** | Questions covered / Total questions × 100 | XX% | 100% — mandatory |
+| **Source diversity** | Unique sources / Data points | X.X | ≥ 1.5 — good |
+| **Counter-evidence rate**| Theses with counter-search / Total theses × 100 | XX% | 100% — mandatory |
+| **Conflict rate** | Unresolved conflicts / Total data points × 100 | XX% | ≤ 10% — acceptable |
+
+**Research Quality:**
+
+| Verified % | Assessment | Action |
+|:----------:|--------|----------|
+| ≥ 70% | 🟢 High | Ready to hand off to the Data Analyst |
+| 50-69% | 🟡 Medium | Hand off with a caveat, recommend additional search |
+| < 50% | 🔴 Low | Rework: find more T1/T2 sources |
+
+### Step 7 — Results Compilation
+1. Structure data by research brief questions.
+2. For each question: facts → sources (with Tier) → confidence level → counter-position.
+3. Highlight conflicts and unresolved questions.
+4. Attach quality metrics.
+5. Formulate conclusions for handoff to the Data Analyst.
+
+## Search Query Patterns
+
+### Mandatory Patterns
+| Pattern | Example | Purpose |
+|---------|--------|------|
+| `[topic] [year]` | "SaaS market size 2025" | Current data |
+| `[topic] statistics report` | "e-commerce Russia statistics report" | Official reports (T1) |
+| `[topic] criticism risks` | "remote work criticism risks" | Counter-evidence |
+| `[company] revenue financial results` | "Skillbox revenue 2025" | Financial data |
+| `[topic] vs [alternative]` | "Bitrix24 vs amoCRM 2025" | Comparative data |
+| `site:statista.com [topic]` | `site:statista.com cloud computing` | T1 sources |
+| `[topic] research [industry]` | "retention benchmarks SaaS B2B" | Industry benchmarks |
+
+### Prohibited Patterns
+| Pattern | Why | Alternative |
+|---------|--------|-------------|
+| One general query for the entire topic | Superficial data | Minimum 3 queries per data point |
+| Query without a year | Outdated results | Always include the year |
+| Only confirming queries | Confirmation bias | Must search for counter-evidence |
+| Copying data without verification | Unreliable data | Cross-check with 2+ sources |
+| Only T3 sources | Low reliability | Priority: T1 → T2 → T3 as a supplement |
+
+## Example — Web Research for "EdTech Market Russia, Corporate Training"
+
+**Research brief (from INT-01):** 4 of 12 questions (fragment).
+
+### Question 1: What is the size of the Russian EdTech market (TAM)?
+
+| # | Fact | Source | Tier | URL | Date | Confidence |
+|---|------|----------|:----:|-----|------|:-------------:|
+| 1 | Russian EdTech = $4.2B (2026) | Smart Ranking | T1 | smart-ranking.ru/... | 2026-01 | ✅ Verified |
+| 2 | Russian EdTech = $3.9-4.5B (2026) | HSE, ISSEK | T1 | hse.ru/... | 2025-11 | ✅ Verified |
+| 3 | Corp. training = 43% of EdTech | HSE | T1 | hse.ru/... | 2025-11 | ⚠️ Estimated |
+
+**Counter-position:** Smart Ranking notes that 30% of the market is the shadow sector (private tutors), which may inflate TAM. Source: smart-ranking.ru/methodology.
+**Conflicts:** None (a spread of $3.9-4.5B is normal for different methodologies).
+
+### Question 3: What technological trends are reshaping the market?
+
+| # | Fact | Source | Tier | URL | Date | Confidence |
+|---|------|----------|:----:|-----|------|:-------------:|
+| 1 | AI personalization: 12% adoption in EdTech| Gartner | T1 | gartner.com/... | 2025-09 | ✅ Verified |
+| 2 | AI personalization: Duolingo +40% retention | TechCrunch | T2 | techcrunch.com/... | 2025-07 | ✅ Verified (with #1) |
+| 3 | LLM reduces course cost ×5 | a16z | T2 | a16z.com/... | 2026-02 | ⚠️ Estimated |
+| 4 | "AI in learning is overhyped for the corp. sector" | HBR | T1 | hbr.org/... | 2025-10 | ✅ Counter-position |
+
+**Counter-position:** HBR argues that corporate training requires human-in-the-loop and AI-only solutions do not work for soft skills.
+
+### Quality Metrics
+
+| Metric | Value | Threshold | Assessment |
+|---------|:--------:|:-----:|:------:|
+| Verified % | 62% (18/29) | ≥ 50% | 🟡 Medium |
+| Coverage | 100% (12/12 questions) | 100% | 🟢 |
+| Source diversity | 1.7 (48 sources / 29 points) | ≥ 1.5 | 🟢 |
+| Counter-evidence rate| 100% (12/12 theses) | 100% | 🟢 |
+| Conflict rate | 7% (2/29) | ≤ 10% | 🟢 |
+
+**Quality: 🟡 Medium** (Verified 62% — hand off with a caveat, recommend additional T1 search for ⚠️ data).
+
+### Audit Trail (fragment)
+
+| # | Query | Results | Used | Rejected |
+|---|--------|:-----------:|:------------:|:---------:|
+| 1 | "EdTech market Russia 2025 2026" | 23 | 4 | 19 (outdated: 8, irrelevant: 11) |
+| 2 | "corporate training market Russia report" | 15 | 3 | 12 (T3: 5, duplicates: 4, outdated: 3) |
+| 3 | "EdTech Russia criticism risks 2025" | 9 | 2 | 7 (irrelevant) |
+
+**Total:** 38 queries, 112 results, 48 sources used, 64 rejected.
+
+## Validation (Quality Gate)
+
+Web research is considered complete if:
+
+- [ ] All research brief questions are covered by data (Coverage = 100%)
+- [ ] WebSearch / WebFetch are used (not "from memory")
+- [ ] Each data point has: URL, date, source Tier, confidence level
+- [ ] Cross-checking is performed (2+ sources per key data point)
+- [ ] Counter-evidence is deliberately searched for each thesis (Counter-evidence = 100%)
+- [ ] Data is current per the requirements table
+- [ ] Audit trail is completed (all queries, URLs, rejection reasons)
+- [ ] Conflicting data is flagged and explained (Conflict rate ≤ 10%)
+- [ ] Minimum number of sources is met (default ≥ 3 per thesis)
+- [ ] Verified % ≥ 50% (if < 50% — rework or hand off with a 🔴 caveat)
+- [ ] T3 sources are not used as Verified or Estimated
+
+## Handoff
+The `$web-research` result is input for:
+- **Data Analyst (AN-xx)** — verified data for applying frameworks.
+- **All framework skills** — via the Data Analyst: data feeds into TAM/SAM/SOM, Competitive Analysis, PEST, SWOT, etc.
+- **`$gates` (RES-xx)** — the gate check verifies the presence and quality of the web-research.
+
+Handoff format: a structured report (based on the research brief questions) + source table (with Tier and confidence) + quality metrics + audit trail. When handing off — use `$handoff`.
+
+## Anti-patterns
+
+| Mistake | Why It's Bad | How to Do It Right |
+|--------|-------------|---------------|
+| Data "from memory" without search | May be outdated or inaccurate | Always execute WebSearch |
+| One source = "fact" | No cross-checking | 2+ sources (T1/T2) for key data |
+| No counter-evidence | Confirmation bias, one-sided analysis | Deliberately search for refutations (100%) |
+| URL without date and Tier | Impossible to assess currency and reliability | Always: URL + date + Tier (T1/T2/T3) |
+| All data marked Verified | Unrealistic, hides uncertainty | Honestly: Verified / Estimated / Assumed |
+| Queries without a year | Outdated results | Include the current year in the query |
+| Skipping the audit trail | Impossible to verify research quality | Record all queries, URLs, rejection reasons |
+| Only T3 sources | Low reliability, cannot be Verified | Priority T1 → T2; T3 only as a signal |
+| Verified < 50% without rework | Majority of data is unreliable | Rework: find T1/T2 or hand off with a 🔴 caveat |
+
+## Output Template
+
+```markdown
+### Web Research Report — [Topic]
+
+**Researcher:** [RES-xx]
+**Date:** [YYYY-MM-DD]
+**Research brief:** [N questions from INT-01]
+
+---
+
+#### Quality Metrics
+
+| Metric | Value | Threshold | Assessment |
+|---------|:--------:|:-----:|:------:|
+| Verified % | XX% | ≥ 50% | 🟢/🟡/🔴 |
+| Coverage | XX% | 100% | 🟢/🔴 |
+| Source diversity | X.X | ≥ 1.5 | 🟢/🟡 |
+| Counter-evidence rate| XX% | 100% | 🟢/🔴 |
+| Conflict rate | XX% | ≤ 10% | 🟢/🟡 |
+
+**Research Quality:** 🟢 High / 🟡 Medium / 🔴 Low
+
+---
+
+#### Results by Question
+
+##### Question 1: [Formulation from research brief]
+
+| # | Fact / Data | Source | Tier | URL | Date | Confidence |
+|---|--------------|----------|:----:|-----|------|:-------------:|
+| 1 | [Value] | [Name] | T1/T2/T3 | [URL] | [YYYY-MM] | ✅ / ⚠️ / 🔮 |
+
+**Counter-position:** [Found counter-arguments or "None found"]
+**Conflicts:** [Discrepancies or "None"]
+
+##### Question 2: [Formulation]
+[same structure]
+
+---
+
+#### Confidence Summary
+
+| Level | Count | Share |
+|---------|:------:|:----:|
+| ✅ Verified | [N] | XX% |
+| ⚠️ Estimated | [N] | XX% |
+| 🔮 Assumed | [N] | XX% |
+
+#### Unresolved Conflicts
+
+| # | Data A | Source A (Tier) | Data B | Source B (Tier) | Comment |
+|---|---------|:-----------------:|---------|:-----------------:|-------------|
+| 1 | [Data] | [Source] | [Data] | [Source] | [Comment] |
+
+---
+
+#### Audit Trail
+
+| # | Query | Results | Used | Rejected (reason) |
+|---|--------|:-----------:|:------------:|---------------------|
+| 1 | [query] | [N] | [N] | [N] ([reasons]) |
+
+**Total:** [N] queries, [N] results, [N] used, [N] rejected
+```