Skip to content

Latest commit

 

History

History
76 lines (59 loc) · 3.03 KB

File metadata and controls

76 lines (59 loc) · 3.03 KB
status accepted
date 2025-08-26

Use "Decision Table" Instead of "Decision Tree"

Context and Problem Statement

SSVC uses a tabular model to map combinations of decision point values to recommended actions. Historically, this model was referred to as a "decision tree" or "decision policy" with both terms being used interchangeably in documentation and tooling. While our use of "decision tree" was consistent with usage in the context of Operations Research (Wikipedia: Decision tree), it caused confusion with the related-yet-different concept from Machine Learning (Wikipedia: Decision tree learning). Furthermore, the term "decision policy" has overloaded meanings beyond our intended usage. We want to avoid confusion on both fronts, therefore we need a new term.

Decision Drivers

  • Avoid ambiguity with the machine learning and operations research concept of "decision tree"
  • Use a less contentious term than "policy" as we are not dictating "capital-P Policy"
  • Accurately describe the underlying SSVC data structure, which is a table of decision point value combinations and outcomes
  • Improve clarity in documentation, tooling, and communication with practitioners
  • Align terminology with the actual data representations used (CSV and JSON tables)

Considered Options

  • Switch to decision framework
  • Switch to decision table
  • Switch to decision model

Decision Outcome

Chosen option: "Switch to decision table"

Rationale

"Decision table" accurately describes the tabular structure used by SSVC and avoids the well-established conflicting meaning of "decision tree" in machine learning and statistics.

We may use decision model for the more generic class of things to which a decision table belongs, but decision table is specific enough to make it clear exactly what our intended meaning is.

Consequences

  • Good, because the terminology now accurately reflects the underlying data model (a table mapping decision point values to outcomes)
  • Good, because it reduces confusion for practitioners familiar with machine learning
  • Bad, because some existing documentation, external references, and tooling may still use "decision tree" — these will be updated incrementally
  • Bad, because the change breaks consistency with prior SSVC publications and external references that use "decision tree"

Confirmation

Documentation, code, and tooling use "decision table" as the primary term for an SSVC decision model. References to "decision tree" in the visual/hierarchical sense (i.e., tree diagrams used to illustrate a decision table) may still appear where appropriate but should be clearly distinguished from the decision table itself.

More Information