Reliability-Aware Intelligent Augmentation (RAIA): A Knowledge-Governed Framework for Reliable Synthetic Data Engineering in Dense Retail Object Detection

This record provides the reproducibility and audit-evidence package associated with the study “Reliability-Aware Intelligent Augmentation (RAIA): A Knowledge-Governed Framework for Reliable Synthetic Data Engineering in Dense Retail Object Detection.” RAIA is a knowledge-governed framework for controlling synthetic-sample admission through explicit reliability constraints, annotation reconciliation, provenance, and auditable decision records in dense retail object detection. The release is intended to support verification of the reported experimental protocol and statistical results. It includes license-compatible reproducibility materials such as per-seed evaluation metrics, canonical evidence records, analysis-support files, audit schemas, reason-code definitions, claim-boundary documentation, and integrity manifests. Source or derived images and third-party resources are not redistributed when upstream licenses prohibit redistribution. The current study evaluates RAIA-E7 under the reported detector configuration. The separate RT-DETR-L experiment is an evidence-governance stress test comparing Generic Copy-Paste and Shelf Geometry-Preserving Augmentation (SGPA); it is not a direct cross-detector efficacy evaluation of RAIA-E7. Files and metadata in this record should be interpreted together with the associated manuscript and its Supplementary Material. SHA-256 integrity information is provided within the release package.