epistemic-grounding

To prevent AI hallucinations in technical documentation, ground every claim to verifiable evidence by labeling assertions as verified, inferred, or asserted, and bind each to explicit data and warrants. This establishes auditability and source citation accuracy.

Structuring architecture proposals using Toulmin primitives involves reconstructing claims with explicit data, warrants, backing, qualifiers, and rebuttal conditions. This framework binds assertions to verifiable evidence while avoiding anti-patterns like cargo-cult citation and authority projection.

Apply RFC 2119 normative vocabulary during SKILL.md authoring by using MUST, SHOULD, and MAY accurately to define contract weights. This prevents blurred requirement boundaries and ensures every normative assertion maintains precise epistemic grounding.

Distinguish verified, inferred, and asserted claims during code review by explicitly labeling the verification state of each assertion. This requires binding every claim to verifiable evidence and an appropriate modality, making the auditability of findings transparent.

Common failure modes when adding source citations to audit findings include cargo-cult citation, stale grounding, vibe-based assertions, citation laundering, and authority projection. Avoid these by reconstructing Toulmin primitives and requiring explicit verification states.

Yes, epistemic grounding applies to code review comments and audit findings by binding assertions to verifiable evidence with explicit modality and warrant. It requires RFC 2119-accurate vocabulary and verified, inferred, or asserted labeling to ensure claims are auditably trustworthy.