Published July 3, 2026
| Version v1.0.5
Software
Open
Kharisara/mer-jit-llm-fgcs: The corresponding figure remains available in the reproducibility artifact package.
Authors/Creators
Description
This release updates the ReplayBench-PG reproducibility package accompanying the final manuscript.
Main changes:
- Renamed the framework to ReplayBench-PG (Replay Benchmark for Policy-Gated AI Pipelines).
- Replaced the proxy diagnostic policy with the deterministic risk_proxy policy.
- Added compact RQ7 controlled fault-validation workflows.
- Added validation outputs for action-flip, unauthorized invocation, action-trace corruption, dropped-row, and duplicated-row fault scenarios.
- Updated local benchmark outputs covering 360 deterministic replay conditions.
- Updated cross-region Google Cloud Run validation outputs for asia-southeast1 and us-central1.
- Updated manuscript-ready tables, figures, benchmark summaries, and reproducibility artifacts.
- Updated benchmark configuration files, documentation, and reproduction scripts.
This release supports the manuscript:
ReplayBench-PG: A Fault-Aware Deterministic Replay Benchmarking Framework for Policy-Gated Multimodal AI Pipelines
Files
Kharisara/mer-jit-llm-fgcs-v1.0.5.zip
Additional details
Related works
- Is supplement to
- Software: https://github.com/Kharisara/mer-jit-llm-fgcs/tree/v1.0.5 (URL)
Software
- Repository URL
- https://github.com/Kharisara/mer-jit-llm-fgcs