Claims, Arguments and Evidence: The Structure of an Assurance Case

Understanding the Foundation of Assurance Cases

An assurance case is a structured argument that demonstrates a system meets specified safety requirements. The fundamental building blocks of any assurance case are claims, arguments, and evidence. These elements work together to create a logical chain that supports the overall safety claim about a system.

Claims represent the statements that must be proven true to demonstrate system safety. The primary claim is usually the top-level safety assertion such as “The AI system will not cause harm to human users.” Supporting claims break this down into smaller, manageable assertions that can be verified through evidence.

For example, a supporting claim might state “The system’s decision-making process will not produce outputs exceeding acceptable risk thresholds.”

Arguments provide the logical reasoning that connects claims to evidence. They explain why the evidence supports the claim being made. Arguments must be clear, logical, and directly relevant to the claim.

They often take the form of conditional statements such as “If the system meets requirement R1, then claim C1 is satisfied.”

Building the Structure of an Assurance Case

The structure of an assurance case follows a hierarchical approach where each level builds upon the previous one. The top level contains the primary safety claim, which is then broken down into sub-claims through decomposition. Each sub-claim requires supporting arguments and evidence to validate it.

Consider a practical example of an AI-based medical diagnostic system. The primary claim might be “The diagnostic system will not misdiagnose life-threatening conditions.” This claim decomposes into sub-claims such as “The system’s accuracy rate for critical conditions exceeds 95%” and “The system’s response time meets clinical requirements.”

• Primary claim: The AI system meets safety requirements
• Supporting claim 1: The system’s accuracy meets clinical standards
• Supporting claim 2: The system’s response time meets operational requirements
• Supporting claim 3: The system’s failure modes are properly managed

Each supporting claim requires specific arguments that explain the reasoning process. The argument for accuracy might state “The system’s validation testing shows accuracy exceeding clinical benchmarks through testing with 1000 clinical cases.” The evidence would then include test results, validation reports, and clinical trial data.

Linking Evidence to Claims Effectively

Evidence must directly support the claims it is meant to validate. The relationship between evidence and claims must be clearly established through logical arguments. Evidence alone is insufficient; it must be connected to claims through reasoning that explains the connection.

Effective evidence selection involves choosing data that directly addresses the specific claim being made. For instance, if the claim is about system reliability, evidence should focus on failure rates, uptime statistics, or reliability testing results. If the claim involves data privacy, evidence must demonstrate compliance with data protection measures and privacy controls.

Organising evidence within an assurance case involves creating clear pathways from claims through arguments to supporting data. Each piece of evidence should have a designated purpose that directly contributes to proving a particular claim. The evidence must be sufficient, relevant, and credible to support the argument being made.

Practical implementation requires documenting these relationships clearly. A well-structured assurance case shows exactly what evidence exists for each claim, how that evidence supports the claim through logical arguments, and what gaps might exist in the argumentation. This approach helps identify weaknesses in the safety case and provides a clear framework for addressing potential concerns.

The structure must accommodate both technical and organisational aspects of safety. Technical claims require empirical evidence from testing and analysis. Organisational claims need documentation of processes, procedures, and governance structures that support safety outcomes.

Both types of claims must be properly linked through appropriate arguments that explain how organisational factors contribute to technical safety outcomes.