Skip to main content

Taxonomy

As the OSCAL Foundation's FedRAMP Technology Focus Group (TFG) reaches agreements as to the path forward, they are captured here. This is a work in progress. Once the group reaches an appropriate milestone, this will be consolidated and re-organized into more consumable guidance.

DRAFT - Please review and provide feedback. Self-register via Login to leave comments.

See Taxonomy

Guiding Principles

We defined the following DRAFT guiding principles.

  • Data Analysis: The Data Analysis must clearly define the intended result of data collection, either:
    • a pass/fail determination; or
    • a consistently quantifiable degree of health.
  • Measurement Basis: Data Source and Data Analysis must distinguish Base, Derived, Interpretive and Probabilistic Measures, including those measures produced by Agentic AI.
  • Agent Basis: Data Source and Data Analysis must identify the involvement of Automation, Statistical Model, Human, and/or Agentic AI; Telemetry and Findings inherit this classification from the Provenance Agent and Analysis Agent that produced them

Data Analysis

The Data Analysis must clearly define the intended result of data collection, either:

  • a Pass/Fail determination; or
  • a consistently quantifiable Degree of Health.

All KSI telemetry is gathered in support of a clear and specific result.

Pass/Fail

Most compliance frameworks seek to define clear pass/fail criteria against specific controls or requirements.

While Pass/Fail is still a preferred target for FedRAMP KSIs, it is not mandatory.

Degree of Health

FedRAMP KSIs open the door to other quantifiable metrics that can be analyzed to indicate a system's overall cybersecurity "health". Telemetry reflecting Degree of Health should include an unambiguous analysis mechanism whenever practical, such as thresholds.

In some instances, measurements may require organization-specific or system-specific "baselining" before thresholds can be defined. This should be conducted within a defined period of time and adjusted periodically as necessary.

Measurement Basis

Data Source and Data Analysis must distinguish Base, Derived, Interpretive and Probabilistic Measures, including those measures produced by Agentic AI.

Categories

The Measurement Basis Categories are:

  • Concrete Measures: Are quantitative without subjectivity.
    • Base Measure: A measure obtained by direct observation of a single attribute, requiring no computation or inference from other measures. (ISO 15939 term)
    • Derived Measure: A measure computed as a function of two or more Base and/or Derived Measures. (ISO 15939 term)
  • Inferential Measures: May be quantitative or qualitative and have some element of subjectivity.
    • Interpretive Measure: A measure produced through human or Agentic AI judegment rather than deterministic computation or statistical inference — a qualitative assessment that cannot be fully reduced to a formula.
    • Probabilistic Measure: A measure produced by applying a statistical model, threshold, or algorithmic inference to underlying data, yielding a likelihood, score, or classification rather than a directly observed value.

Application

Measurement Basis applies to both Acquisition Basis and Analysis Basis.

  • Acquisition Basis: The Measurement Basis classification of the data as received from the Data Source, before the KSI's own Data Analysis is applied.
  • Analysis Basis: The Measurement Basis classification of the logic the KSI itself applies to produce the Finding.

A Finding's Acquisition Basis and Analysis Basis may differ from one another. For example, A Finding built on a Concrete Measure is still, in effect, an Inferential Finding if the underlying Data Source itself returned a Probabilistic value.

Both Acquisition Basis and Analysis Basis must be provided such that an Agency can filter based on their risk tolerance for Evidence based on Inferential Measures.

Agent Basis

Data Source and Data Analysis must identify the involvement of Automation, Statistical Model, Human, and/or Agentic AI; Telemetry and Findings inherit this classification from the Provenance Agent and Analysis Agent that produced them

Categories

The Agent Basis Categories are:

  • Automation: Deterministic, rule-based, automated mechanism.
  • Statistical Model: Machine Learning, probabilistic, non-agentic mechanism.
  • Human: Manual assembly, subjective analysis, judgement calls or similar activities performed by humans.
  • Agentic AI: An autonomous or semi-autonomous AI system capable of independent action, tool use, or decision-making without direct human execution of each step

Application

Agent Basis applies to both Provenance Agent and Analysis Agent.

  • Provenance Agent: Categorizes the actor that makes the data available ahead of its acquisition by the Collection Mechanism.
  • Analysis Agent: Categorizes the mechanism that performs the analysis. See Agent Basis below for details.

Agencies need a way to filter Evidence based on strictly deterministic automation or less deterministic influences from humans, machine-learning and Agentic AI mechanisms.

KSI Approach

  • KSI Statement Analysis

    • Identify KSI goal(s)

  • KSI Data Requirements:

    • Intent: A simple, unambiguous status. Typically Pass/Fail
    • Decision Point: Sampling or full coverage?
    • Identify evidence to collect in support of KSI goal(s)
    • Define evidence interpretation
      • Evidence type (count, true/false, setting)
      • Collection frequency
      • Evidence fidelity
      • Correlation
      • Thresholds
        • Could be more granular than just pass/fail
        • Example: Satisfactory, Degraded, Critical

  • Organizational Considerations:

    • Action Triggers: Define the triggers for that system/org
      • Define required action(s) when triggered

  • KSI Techical Collection Approach

    • Identify all evidence source/component
      • Identify the source format(s)
    • Define centralized evidence collection target
    • Define the automation required to acquire evidence from each source/component and deliver to centralied collection target
      • Decision Point: Evidence delivered raw or normalized?

  • KSI Technical Interpretation Approach

    • Apply data requirements/thresholds to produce Findings
      • Findings are continuously updating as new data is received and analyzed
    • Raise action triggers when appropriate
Back to top