Citation Discipline in Spec-Driven Development: A Cross-Model Empirical Study of Output Determinism and Automated Hallucination Detection in LLM-Generated Code
arXiv:2606.30689v1 Announce Type: cross Abstract: Spec-Driven Development (SDD) frameworks guide Large Language Model (LLM)-powered code generation through formal specifications, yet they differ fundamentally in how they enforce traceability between requirements and generated code. This paper presents two controlled empirical studies comparing three SDD frameworks: $traceSDD$, which enforces mandatory per-line requirement citations using hierarchical REQ-XXX.Y.Z identifiers; $Spec Kit$, which uses artifact-level traceability through user stories and acceptance criteria; and $OpenSpec$, which r
The proliferation of LLM code generation necessitates robust methods for quality control, traceability, and determinism, especially as these tools move into critical applications.
Ensuring the reliability and verifiability of LLM-generated code is crucial for its adoption in enterprise and mission-critical systems, directly impacting development costs, security, and trust.
New methodologies are emerging to impose greater discipline and accountability on generative AI, moving beyond raw output to integrated, verifiable development practices.
- · Software Development Lifecycle (SDLC) tool providers
- · Enterprises adopting AI code generation
- · Companies focused on AI safety and explainability
- · Unstructured, ad-hoc AI code generation practices
- · Developers neglecting formal specification in AI integration
Improved trust and accelerated adoption of LLM-powered code generation in regulated and high-assurance domains.
Increased demand for developers skilled in formal specification and verification techniques to guide and audit AI outputs.
The potential for AI agents to write and self-verify complex software systems with unprecedented speed, potentially accelerating technological progress across sectors.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.AI