Forseti Life

Data Granularity

Level of detail accessible - from atomic individual records to general concepts only.

Why This Matters

Understanding where an AI system operates on this dimension helps you evaluate its capabilities, limitations, and potential biases. Different power levels are appropriate for different use cases - the key is transparency about what level a system operates at and whether that matches its stated purpose.

Understanding the Scale

Each dimension is measured on a scale from 0 to 9, where:

  • Level 0 - Nothing: Zero capability, no access or processing
  • Levels 1-2 - Minimal capability with extreme constraints and filtering
  • Levels 3-5 - Limited to moderate capability with significant restrictions
  • Levels 6-7 - High capability with some institutional constraints
  • Levels 8-9 - Maximum capability approaching omniscience (∞)

Level Breakdown

Detailed explanation of each level in the 1imension dimension:

Zero granularity. No data points, no detail, no information at any level.

Real-World Example: A system with no data, no measurements, no granularity whatsoever.

Extremely coarse - general concepts and categories only. No detail, nuance, or specific data points.

Real-World Example: Simple category labels ("Health", "Finance", "Entertainment"), basic chatbot topic routing ("This is about billing"), automated email categorization (just "Work" vs "Personal", no content analysis), or voice assistant category recognition ("That's a weather question", "That's a shopping request" - just broad categorization). Note: Granularity scales exist in many domains - for example, the H3 Geolocation framework provides hierarchical spatial granularity from coarse continental regions down to building-level precision.

High-level summaries and overviews only. General themes, broad trends, simplified narratives.

Real-World Example: News article summaries ("Economy improving", "Crime rates down in major cities"), corporate annual report executive summaries (general business performance themes, no specific metrics), Wikipedia article introductions (broad topic overviews without detailed data), or chatbot responses like "Most customers are satisfied" (general sentiment, no statistics or individual feedback).

Statistical summaries and aggregates only. Averages, percentages, trends. No underlying detail.

Real-World Example: CDC public health dashboards (state-level infection rates, vaccination percentages, no individual cases), Bureau of Labor Statistics employment reports (unemployment rate, average wages by sector, no individual job data), weather service climate summaries (average temperatures, rainfall totals, no individual readings), or retail industry reports (overall market share percentages, sales trends, no store-specific detail).

Personal-level data for individual users. User preferences, history, behavior metrics.

Real-World Example: Fitbit personal fitness tracker (your steps, heart rate, sleep patterns - individual user metrics only), Spotify Wrapped (your personal listening statistics and preferences), personal credit score apps (your individual credit metrics, not underlying transaction detail), or Netflix personal viewing metrics (your watch time, preferences, ratings - summary of your behavior but not granular viewing data).

Pre-aggregated data. Groups, neighborhoods, cohorts. No individual record access.

Real-World Example: U.S. Census Bureau demographic data (aggregated by census tract/zip code, not individual households), city crime heat maps (aggregated by neighborhood, not individual incidents), retail sales reports (aggregated by store/region, not individual transactions), or hospital infection rate reporting (aggregated by unit/department, not individual patient cases).

Individual events, incidents, or discrete occurrences. Detailed enough for operational response.

Real-World Example: KCPD incident reporting system (individual crime reports across all officers and districts, but not all city government data), Amazon package tracking logistics (individual shipment events for all customers and operational data, but not Amazon's financial/HR data), hospital ER admission system (individual patient arrivals across facility, but not all hospital systems), or airline flight operations (individual flight events system-wide, but not all airline data).

Domain-specific detailed records. Deep within field but may lack fine-grained data outside specialty.

Real-World Example: Hospital Epic Systems (detailed individual patient vitals, lab results, medications within medical domain but no financial transaction detail), IBM Watson for Oncology (detailed cancer treatment protocols and outcomes but limited general health data), Bloomberg Terminal (individual stock tick data and trades but limited non-financial granularity), or university research database (detailed publication metadata but coarse institutional data).

Detailed records with ability to perform meta-analysis. Individual data points plus synthesized insights.

Real-World Example: Federal Reserve economic research (individual bank transaction data plus macroeconomic aggregates), CDC epidemiology systems (individual case reports plus population-level disease trends), NASA space mission data (individual sensor readings plus mission-level performance analytics), or NSA signals intelligence (individual intercepts plus pattern analysis across networks).

Full access to raw individual records, atomic transactions, and all aggregation levels. Complete analytical flexibility. Approaching god-like granular omniscience.

Real-World Example: No real-world example exists. Level ∞ would require access to both raw atomic-level individual transaction records AND all possible aggregation levels across every domain—individual medical records plus population health analytics, individual financial transactions plus macroeconomic models, individual social media posts plus global sentiment analysis—a combination that approaches divine omniscience at all scales.
Return to Agent Power Framework