Emerald AI field-demonstration data attribution

Upstream: https://github.com/ai-emerald/emerald-ai-demo-may-2025
Revision: 34403aef78c2690a8e4f0ba8f7f62d302cdba967
Source: data/dvfs_sweep.csv
Associated work: Turning AI Data Centers into Grid-Interactive Assets:
Results from a Field Demonstration in Phoenix, Arizona
https://arxiv.org/html/2507.00909v1

The preserved upstream CSV and README are distributed under the upstream
repository's Apache License, Version 2.0. A complete copy of that license is
included in LICENSE.emerald.txt. This attribution notice was added by the
GridPass demo authors; it is not an upstream NOTICE file.

GridPass modifications: numeric parsing, grouping by workload, default
selection of ft_llama_8b_dolly, normalization of commanded caps by 400 W,
normalization of measured GPU power by measured draw at the 400 W cap,
inference of GPU counts from aggregate/per-GPU draw, and addition of
provenance and limitations. The source CSV is preserved unchanged.

Interpolation and all site-scale, demand, timing, actuator, and recovery
behavior belong to the GridPass simulation. They are not additional
measurements or performance claims by Emerald AI. The upstream license
does not grant trademark rights or imply Emerald AI endorsement of GridPass.

This notice identifies the license scope of the incorporated upstream
material; it does not change the license of the surrounding GridPass project.