Citation Discipline in Spec-Driven Development: A Cross-Model Empirical Study of Output Determinism and Automated Hallucination Detection in LLM-Generated Code
2026-07-01T08:51:54Z•288f7137920c18c0c9148b64c285d941d555c78b8b0d1798d0fa4f2321256afc
LLM-code-generationLLVM-optimizationautomated-repaircontrastive-unlearningdeprecated-APIsfalse-trustgit-tag-mutationhallucination-detectionlicense-comparisonlocalizationmemory-optimizationmodel-unlearningoutput-determinismoverconfidencepackage-managementrepository-grounded-repairreproducible-buildssecurity-calibrationspec-driven-developmentsupply-chain-securitytraceability
What happened
This collection of papers (arXiv 2606.*) examines LLM-driven software engineering and supply-chain integrity with concrete security implications. Key findings: (1) Spec-driven citation (traceSDD) enables automated hallucination detection but reduces output determinism, exposing a trade-off between verifiability and reproducibility. (2) Security calibration of code LLMs is weak—models are frequently overconfident (False Trust) in vulnerable code; architectural gating helps on controlled tasks but calibration degrades in real repository contexts, and calibration-guided repair yields limited, oft
Why it matters
A reviewed impact interpretation has not been published for this record.
Evidence and limitations
- Source ID
- arxiv_cs_se
- Record identifier
- 288f7137920c18c0c9148b64c285d941d555c78b8b0d1798d0fa4f2321256afc
- Enrichment time
- 2026-07-01T08:51:54Z
- AI-assisted enrichment
- Yes
This record may overlap with other records. Its enrichment can be incomplete or wrong, and machine assistance was used. Validate consequential decisions against the linked source and your own environment.