Research collective
Halo is a research collective first, a protocol second. Below: the whitepaper, field notes from production, and a handful of recorded talks. New writing roughly monthly; subscribe via RSS or follow on X.
The 32-page protocol spec. Identity, attestation, weighting, slashing, governance — formally defined with proofs. The canonical reference for anyone integrating, attacking, or extending Halo.
Read →When agents act for themselves, we lose every form of trust we built around humans. This is the first essay in a series on what comes next — and why a credit-bureau model doesn't.
Read →A formal analysis of how the half-life parameter τ affects Sybil resistance under varying stake-distribution profiles. With simulation code on GitHub.
Read →Why keeping reputation in a SaaS dashboard makes it useless for the very systems that need it most. Public, queryable, slashable — or nothing at all.
Read →Forty-five minutes with Anatoly Yakovenko, Mert Mumtaz, and the team. Transcript + slides. A useful primer if you want the verbal version of the whitepaper.
Read →Most reputation systems can punish — none can punish atomically. Atomic slashing collapses six months of off-chain dispute into a single transaction. This is the unlock.
Read →Reading list
The v3.1 whitepaper covers the protocol's full surface: identity registration, attestation formats, the weighting algebra, slashing flow, and governance. It is the reference we point integrators & auditors to.
Whitepaper · v3.1 · April 2026
Co-authored by E. Marais, K. Aoki, and M. Velasquez. 32 pages, formal definitions, three proof appendices, simulation code on GitHub. Apache-2.0.
@misc{halo2026,
author = {Marais, Elena and Aoki, Kenji and Velasquez, Mira},
title = {A reputation primitive for autonomous agents},
year = {2026},
url = {https://halo.research/whitepaper.pdf},
}Halo is open research. The protocol, the spec, the simulation code — all under Apache-2.0. If you find a hole in the math, we want to know.