Skip to content
CarbonStream Nexus — Compute Accountability Infrastructure

Research

Research and publications

CSN's position is developed in writing first. This register lists its documents in the prescribed sequence, with publication evidence limits stated separately from technical authority.

Publications

Library

Register order is authoritative and is not sorted by DOI publication dates. CAR Technical Specification is normative over CAR Overview for technical conflicts only. RC1 is for technical review, not ratified; the prototype claims no conformance. No wider authority hierarchy is asserted.

Compute Accountability (CAPS-Core v1.0)

The foundational statement of the compute accountability problem and the record-centred approach CSN proposes.

Published
DOI
10.5281/zenodo.20466837
Access
Published on Zenodo

CarbonStream Nexus Interoperability Standards v1.0

A specification-level document defining interoperability conventions for compute accountability records and governance exchanges.

Published
DOI
10.5281/zenodo.21851002
Access
Published on Zenodo

CSN Compute Accountability Standard (CCAS) v1.0

CSN standard listed in the document register; its source text is not incorporated into this application.

Published
DOI
10.5281/zenodo.22886163
Access
Published on Zenodo

CAR Overview / CAR-OV-001

Compute Accountability Record (CAR) Overview and Adoption Guide — public website document, not a Zenodo publication.

Read CAR Overview and Adoption Guide

Access
Public website document; no Zenodo DOI

CSN-CAR Core v1.1, Release Candidate 1 (Technical Review Release)

In development

Compute Accountability Record Specification — RC1, technical review only; not ratified.

Technical review window and Submit review comments

Published
DOI
10.5281/zenodo.23366684
Access
Published on Zenodo — technical review only, not ratified

Open questions

Unresolved by the current specification

Published deliberately. These are the points where the protocol could be wrong, and where outside critique is most useful.

  • Q1

    How should facility overhead be declared when the operator has no visibility into it?

  • Q2

    What is the minimum a record must carry before an attribution figure is meaningful to a third party?

  • Q3

    How should temporal grid variability be represented without implying measurement precision that does not exist?

  • Q4

    What allocation rule is defensible when a single accelerator serves many tenants and many workloads?

  • Q5

    What would make a compute accountability record acceptable as evidence in an institutional process?