Explainability

Making Sense of Insights

Explainability in Blankstate refers to the ability for users to understand *why* certain operational insights, Metric values, or OEI scores are generated. It's about providing transparency into the analytical process, even though the underlying AI models can be complex.

Pillars of Explainability

• The Blueprint: The Blueprint itself is the primary source of explainability. It's a human-readable, user-defined framework that explicitly states *what* aspects of operations are being measured and *how* (via Protocols and Metrics).
• Protocols and Nuances: Protocols define the specific criteria and nuances (labels with associated scores) that the AI models use for analysis. By reviewing a Protocol's definition, users can understand the specific conditions or patterns the system is looking for.
• Markers and Rationale (Replay): In Replay, the "State of Entities" visualization highlights "Markers" – segments in documents that triggered a Protocol score. Optional AI-generated "Rationale" can further explain *why* a particular Marker received a certain score based on the Protocol definition. This provides a direct link between the insight and the source data.
• Protocol Quality Indicators: The automated Quality Indicators (QIs) calculated for Protocols (like Specificity and Completeness) provide insight into how well-defined and unambiguous a Protocol is, contributing to transparency about its expected analytical behavior.
• Metric Definitions: Metric definitions clearly state *how* Protocol scores are aggregated and calculated into quantitative measurements, providing transparency into how the OEI and other performance indicators are derived.