# ISNAD

ISNAD (Isnād) is an open-source provenance framework for multi-agent AI systems, adapting classical hadith-science methodology — graded transmitter registries, weakest-link chain evaluation, independent corroboration — to AI pipelines.

It grades every transmitter in an AI claim's chain (source, retriever, model, tool, agent), propagates the weakest link to one of four verdicts (serve / caveat / review / quarantine), and emits signed, tamper-detecting audit records (SHA-256 self-hash, HMAC/Ed25519, append-only Merkle log).

**ISNAD grades who handled a claim, not whether it is true, and issues no certifications** — it produces evidence supporting EU AI Act Art 12 logging and Art 19 retention (≥6 months).

## Key facts

- Founder: Ali Zahid Raja (ORCID 0009-0003-7875-4590)
- License: Apache-2.0
- Version: 3.0.5
- Software DOI: 10.5281/zenodo.21216873
- Paper DOI: 10.5281/zenodo.21211290
- arXiv: 2607.24117 (preprint, not peer-reviewed)
- Validated on 575,060 hadith chains (κ = 0.87, 3-way) and RAGTruth (κ = 0.575)

## Links

- [GitHub repository](https://github.com/alizahidraja/isnad)
- [arXiv paper](https://arxiv.org/abs/2607.24117)
- [PyPI](https://pypi.org/project/isnad/)
- [Hugging Face paper](https://huggingface.co/papers/2607.24117)
- [ISNAD-Bench dataset](https://huggingface.co/datasets/alizahidraja/isnad-bench)
- [Software (Zenodo)](https://doi.org/10.5281/zenodo.21216873)
- [Paper (Zenodo)](https://doi.org/10.5281/zenodo.21211290)

## Contact

- contact@isnadhq.com