GRACE-LAKSA Public Verification Bundle
GRACE-LAKSA - FMLDS 2026 Verification Release This release provides the public verification artifacts accompanying the paper: GRACE-LAKSA: A Retrieval-Augmented Graph Framework for Long-Tail Civic Complaint Understanding and Routing Daffa Farras Putra Tarigan, Kevin Fransisco, Arya Swastika, and Evaristus Didik Madyatmadja IEEE FMLDS 2026 Included Public verification script (verify.py) Aggregate experimental results reported in the paper Main-score reconstruction checks Clean-versus-leaky evaluation delta checks Rolling-CV significance-field checks Public-release file allowlist and privacy-sensitive column checks The verification suite requires only the Python standard library: python verify.py Verification Scope Passing the verification suite confirms the internal consistency of the released aggregate results and supported quantitative summary claims. This release does not reproduce model training or provide complaint-level experimental data. Data Availability The original Tangerang City complaint-level dataset is not included due to municipal data-sharing and privacy restrictions. The public release intentionally excludes complaint text, identifiers, timestamps, labels, locations, dataset split membership, per-complaint predictions, embeddings, retrieval indices, model weights, and checkpoints. Reproducibility Boundary This release should be interpreted as a public verification bundle, not a complete training reproducibility package. The synthetic retrieval-latency experiment and later paired holdout randomization test are not included in the automated public verifier because durable source artifacts for these analyses are not currently available under the released outputs/ provenance boundary. Version v1.2-fmlds2026 represents the public verification snapshot associated with the FMLDS 2026 version of GRACE-LAKSA. Future repository changes may extend the verification tooling, while this release remains the reference snapshot for the corresponding paper. License Verification software is released under the Apache License 2.0. The license does not grant rights to the original municipal complaint dataset.