The work behind the protocol.
Everything here is meant to be checked. The papers are on Zenodo, the models are on Hugging Face, and the protocol is on GitHub under Apache-2.0. Same principle as a signed fact: read it, run it, decide for yourself.
Reproduce it, live.
The claims in the papers run as public endpoints right now. Check them from your terminal; no key, no account.
What the responder can do today
curl -s https://emem.dev/v1/capabilities # models loaded, GPU state, live extensionsThe transparency log's signed tree head
curl -s https://emem.dev/v1/log/sth # every fact lands in a witnessed, Merkle-proofed logThe benchmark the papers describe
curl -s https://emem.dev/v1/benchmark # run the numbers yourself, not our summary of themWhy we publish
The whole protocol rests on one idea: a fact you cite is a fact you can check. That only holds if the method is in the open too. So the design is written up, the models are downloadable, and the code is readable. If something here is wrong, it can be found and corrected, which is the point.
The papers are authored by Jaya Kumari and Avijeet Kumar Singh, published on Zenodo under a Creative Commons licence: DOI 10.5281/zenodo.20706317 and DOI 10.5281/zenodo.20706893. The models are LoRA fine-tunes of Gemma for grounding a language model in Earth observation. The protocol itself is emem.