carbon.data.qa

carbon.data.qa

Purpose

This skill enables Claude to answer factual, analytical questions about carbon accounting data by querying Carbon ACX's internal datasets (CSV files in data/ directory), derived artifacts, and the local API when running. It encodes domain knowledge about:

• Carbon accounting terminology and units (tCO2e, kWh, pkm, etc.)
• Emission factor structures and relationships
• Activity-to-emissions calculations
• Temporal data queries (Q1 2024, monthly totals, etc.)
• Layer, sector, and profile hierarchies

When to Use

Trigger Patterns:

• User asks about emissions data: "What were total CO2 emissions for Q1 2024?"
• Queries about specific activities: "What's the emission factor for streaming video?"
• Comparative questions: "Compare emissions from cloud storage vs local storage"
• Data exploration: "Show me all activities in the professional services layer"
• Unit conversions: "Convert 500 kWh to tCO2e"
• Source/​provenance queries: "Where does the video streaming data come from?"

Do NOT Use When:

• User wants to generate reports (use carbon.report.gen instead)
• User wants to write code (use acx.code.assistant instead)
• Questions about repo structure or development setup
• Non-carbon-accounting questions

Allowed Tools

• read_file - Read CSV data files, JSON artifacts, schemas
• python - Process data, perform calculations, query APIs
• grep - Search for specific activities or emission factors
• bash - Run simple data queries via command line (read-only)

Access Level: 1 (Local Execution - read-only, no file writes, no external network)

Tool Rationale:

• read_file: Required to access canonical CSV data in data/ directory
• python: Needed for parsing CSVs, JSON artifacts, performing unit conversions and emission calculations
• grep: Efficient searching through data files for specific patterns
• bash: Helpful for quick file inspection and data exploration

Explicitly Denied:

• write_file, edit_file - This is a read-only analytical skill
• web_fetch with external URLs - Only internal localhost API endpoints allowed

Expected I/​O

Input:

• Type: Natural language question (string)
• Format: Free-form query about carbon data
• Constraints: Must relate to carbon accounting, emissions, or activities in the dataset
• Examples:
  • "What is the emission factor for coffee?"
  • "Total emissions from video streaming in 2024"
  • "List all military operations activities"
  • "What units are used for grid intensity?"

Output:

• Type: Structured answer with data, units, and citations
• Format: Markdown with tables, bullet lists, and inline values
• Requirements:
  • MUST include units (tCO2e, kWh, etc.) with all numeric answers
  • MUST cite data sources - reference source_id from data/​sources.csv
  • MUST include timestamp - data vintage or "as of" date
  • Handle ambiguity by asking clarifying questions
• Example:

  **Emission Factor for HD Video Streaming:**
  
  - Activity: `MEDIA.STREAM.HD.HOUR` (HD video streaming per hour)
  - Emission Factor: 0.055 kgCO2e/​hour
  - Unit: kgCO2e per hour of streaming
  - Source: [SOURCE_ID_123] - "Streaming Energy Report 2023"
  - Vintage: 2023
  - Notes: Includes device playback + network delivery
  

Validation:

• Every numeric value has explicit units
• Sources are referenced by source_id
• "Unknown" or "Data not available" for missing data (never guess)
• Calculations show methodology

Dependencies

Required:

• Access to Carbon ACX data directory (data/)
• Python 3.11+ with pandas, PyYAML
• Understanding of data schema (see reference/​data_schema.md)
• Carbon accounting units glossary (see reference/​units_glossary.md)

Data Files:

• data/​activities.csv - Activity catalog
• data/​emission_factors.csv - Emission factors
• data/​layers.csv - Layer definitions
• data/​sectors.csv - Sector taxonomy
• data/​units.csv - Unit definitions and conversions
• data/​sources.csv - Data provenance
• data/​profiles.csv - Activity profiles
• calc/​outputs/ - Derived artifacts (if available)

Optional:

• Local API at http://localhost:8787/​api (when Worker is running)
• Derived JSON manifests in dist/​artifacts/

Examples

Example 1: Basic Emission Factor Query

User: "What's the emission factor for a cup of coffee?"

Claude Process:

1. Recognize this is a carbon data query → invoke carbon.data.qa skill
2. Search data/​activities.csv for coffee-related activities
3. Find matching emission factor in data/​emission_factors.csv
4. Look up source details in data/​sources.csv
5. Format answer with units and citation

Output:

**Coffee Emission Factor:**

- Activity: `FOOD.COFFEE.CUP.HOT` (12 oz hot coffee cup)
- Emission Factor: 0.021 kgCO2e/​cup
- Source: [SRC_FOOD_2023] Coffee supply chain LCA
- Vintage: 2023
- Scope: Includes bean production, roasting, brewing, waste

Example 2: Aggregation Query

User: "Show me total emissions from all social media activities"

Claude Process:

1. Query data/​activities.csv for activities with category=social
2. Load corresponding emission factors
3. Aggregate by activity type
4. Present as table with sources

Output:

**Social Media Activity Emissions (per hour):**

| Activity | Emission Factor | Source |
|----------|----------------|--------|
| YouTube streaming | 0.046 kgCO2e/​hr | SRC_MEDIA_2024 |
| Facebook browsing | 0.038 kgCO2e/​hr | SRC_SOCIAL_2024 |
| Instagram scrolling | 0.041 kgCO2e/​hr | SRC_SOCIAL_2024 |
| TikTok watching | 0.043 kgCO2e/​hr | SRC_SOCIAL_2024 |

*All values include device use + network transfer*
*Data vintage: 2024*

Example 3: Ambiguous Query Handling

User: "What are the emissions for streaming?"

Claude Process:

1. Recognize ambiguity - multiple streaming types exist
2. Ask clarifying question before searching

Output:

I found multiple streaming activities in the dataset:

1. **Video streaming:**
   - HD video (1080p)
   - UHD/​4K video
   - SD video (mobile)

2. **Audio streaming:**
   - Standard quality music

3. **Platform-specific:**
   - YouTube
   - Social media video (TikTok, Instagram, etc.)

Which type of streaming would you like emission data for?

Example 4: Unit Conversion Query

User: "How much CO2 does 500 kWh of Ontario grid electricity produce?"

Claude Process:

1. Look up Ontario grid emission factor in data/​emission_factors.csv
2. Find activity ENERGY.CA-ON.GRID.KWH
3. Perform calculation: 500 kWh × emission factor
4. Show calculation with units

Output:

**Ontario Grid Electricity Emissions:**

- Activity: `ENERGY.CA-ON.GRID.KWH`
- Consumption: 500 kWh
- Emission Factor: 0.025 kgCO2e/​kWh (Ontario IESO grid average)
- **Total Emissions: 12.5 kgCO2e** (0.0125 tCO2e)

*Calculation: 500 kWh × 0.025 kgCO2e/​kWh = 12.5 kgCO2e*
*Source: [SRC_GRID_ON_2024]*
*Vintage: 2024 annual average*

Example 5: Edge Case - Missing Data

User: "What's the emission factor for blockchain mining?"

Claude Process:

1. Search data/​activities.csv for blockchain/​crypto activities
2. No matches found
3. Return graceful "not found" response

Output:

**Data Not Available**

I could not find emission factors for blockchain or cryptocurrency mining in the current Carbon ACX dataset (v1.2).

The dataset currently focuses on:
- Professional services and food
- Digital infrastructure (streaming, cloud, AI)
- Industrial operations
- Transportation and logistics
- Military and defense operations
- Earth system feedbacks

You might be interested in related activities:
- Cloud server operations (`ONLINE.DC.CLOUD.SERVER.HOUR`)
- Data center rack usage (`ONLINE.DC.COLOCATION.RACK.MONTH`)

Would you like information on any of these instead?

Limitations

Known Edge Cases:

• Cannot answer questions requiring data not in the CSV files
• Temporal queries limited to vintage years present in dataset
• Cannot perform predictive modeling or forecasting
• Regional data limited to what's explicitly coded (e.g., Ontario grid)
• Some activities have emission factors marked as "to be added"

Performance Constraints:

• Large aggregations across all activities may take 5-10 seconds
• Complex cross-layer queries require multiple file reads
• Derived artifacts may not always be up-to-date with source CSVs

Security Boundaries:

• Read-only access to data files
• No external API calls (except localhost Worker API)
• Cannot modify source data
• Cannot access files outside data/ or calc/​outputs/ directories

Scope Limitations:

• Answers based solely on Carbon ACX dataset - no external knowledge
• Does not perform lifecycle assessments beyond what's in emission factors
• Does not provide regulatory compliance advice
• Does not make emission reduction recommendations (analytical only)

Validation Criteria

Success Metrics:

• ✅ All numeric answers include explicit units (kgCO2e, tCO2e, etc.)
• ✅ Every emission factor cites source_id or notes if source missing
• ✅ Data vintage/​timestamp included in responses
• ✅ Ambiguous queries prompt for clarification before answering
• ✅ Missing data returns graceful "not found" rather than guessing
• ✅ Calculations show methodology (formula with units)
• ✅ Responses match data files exactly (no hallucination)

Failure Modes:

• ❌ Returns emission values without units → REJECT
• ❌ Makes up data not in CSV files → REJECT
• ❌ Provides answers without source attribution → WARN
• ❌ Performs calculations with wrong units → REJECT
• ❌ Answers ambiguous questions without clarification → WARN

Recovery:

• If uncertain about data interpretation: Ask user for clarification
• If data missing: Explicitly state "Data not available" and suggest alternatives
• If calculation complex: Show step-by-step methodology
• If source missing: Note "Source not specified in dataset"

Related Skills

Dependencies:

• None - this is a foundational skill

Composes With:

• carbon.report.gen - Use this skill to gather data, then generate reports
• acx.code.assistant - This skill informs what data structures exist for code generation

Alternative Skills:

• For report generation: carbon.report.gen
• For code generation: acx.code.assistant
• For schema validation: schema.linter

Maintenance

Owner: ACX Team Review Cycle: Monthly (align with dataset releases) Last Updated: 2025-10-18 Version: 1.0.0

Maintenance Notes:

• Update when new CSV files added to data/
• Review when emission factor schema changes
• Validate examples against current dataset version
• Keep reference/​data_schema.md synchronized with actual schema