Research into the JAMES framework identifies a systemic failure in standard Retrieval-Augmented Generation (RAG) architectures regarding auditability and temporal integrity. Through the Replayable Audit Benchmark (RAB) and Lifecycle Retrieval Benchmark (LRB), the study demonstrates that vanilla RAG systems suffer from "temporal decay" and zero replay fidelity (RF 0.000), rendering them non-compliant with EU AI Act mandates for record-keeping and transparency. The JAMES framework utilizes an audit-native Graph-RAG architecture to enable "time-travel retrieval," achieving a Replay Fidelity of 1.000 and an R@1 of 0.845. This transition from retrieval-centric to audit-centric design is critical for meeting the August 2026 enforcement deadlines for high-risk AI systems.

• The Vulnerability of Standard RAG Architectures

  • Temporal Decay Problem: Evolving or superseded factual data renders static RAG retrieval states inaccurate or impossible to reconstruct.
  • Auditability Gap: Standard architectures prioritize retrieval accuracy (finding chunks) over decision defensibility (reconstructing why a specific answer was given).
  • Regulatory Non-Compliance: Inability to reconstruct historical decision states fails to meet mandatory logging and transparency requirements for high-risk AI.

• Technical Framework: The JAMES Architecture

  • Audit-Native Graph-RAG: A local-first implementation designed to maintain a deterministic link between retrieved data and decision context.
  • Technical Stack: Integration of Ollama (gemma4:e4b) for inference, BAAI/bge-m3 for high-dimensional embeddings, and ChromaDB for vector storage.
  • Time-Travel Retrieval: Implementation of temporal logic to allow users to query the knowledge state as it existed at specific historical timestamps.

• Benchmark Methodology: RAB and LRB

  • Replayable Audit Benchmark (RAB): Measures Audit Completeness (AC), Replay Fidelity (RF), and Provenance Coverage (PC) to ensure all decision-relevant events are logged.
  • Lifecycle Retrieval Benchmark (LRB): Evaluates temporal accuracy by comparing Vanilla RAG, Naive-supersede (newest-wins), and Validity-window retrieval methods.
  • Deterministic Verification: Benchmarks are pre-registered to prevent post-hoc metric manipulation and ensure research integrity.

• Performance Analysis and Regulatory Impact

  • Auditability Metrics: JAMES achieved perfect Replay Fidelity (RF 1.000), whereas baseline RAG systems demonstrated zero ability to replay decisions (RF 0.000).
  • Retrieval Efficiency: At publication scale (S3), JAMES achieved an R@1 of 0.845, significantly outperforming the 0.502 baseline of vanilla RAG.
  • EU AI Act Alignment: Architecture directly satisfies Articles 10 (Record-keeping), 12 (Logging), and 19 (Transparency) ahead of the August 2026 enforcement deadline.