# Building-energy demo — data sources & provenance

Every building-energy figure the MCP returns traces to the real, permissively-licensed public
dataset below. Raw files are **not** committed — only the curated, derived slice under
`public/data/hvac/`, rebuilt with `scripts/build-hvac-bldg59.py`.

> Representative public data — not connected to any live building. BMS/BAS ingestion for a
> customer building is a per-deployment step; see `describe_connector_requirements`.

## LBNL Building 59 (Wang Hall), 2018–2020 — `bldg59.json`

- **Source:** *A three-year building operational performance dataset for informing energy
  efficiency*, Hong, Luo, Blum, Wang — Lawrence Berkeley National Laboratory. Two instrumented
  office floors (2 × 2,325 m²) of a 2015 office building in Berkeley, CA. Companion paper:
  Sci Data 9 (2022), doi:10.1038/s41597-022-01257-x.
- **URL:** https://zenodo.org/records/5951008 (mirror of Dryad doi:10.7941/D1N33Q) ·
  **License:** CC0-1.0 (verified on both Dryad and Zenodo) · **Retrieved:** 2026-07-07
  (`Building_59.zip`, SHA-256 prefix `1e224dd7`).
- **Slice taken:** daily and monthly aggregates for 2018-01-01 → 2020-12-31 — per day: kWh by
  electrical end-use (`hvac_N`, `hvac_S`, `lig_S`, `mels_N`, `mels_S`, from the 15-minute
  `ele.csv`, kW × 0.25 h; days with <22 h of rows dropped), outdoor air temperature
  mean/min/max + mean solar (from `site_weather.csv`), and peak occupant count of the two
  instrumented floor-halves (from the 1-minute `occ.csv`).
- **Note:** occupancy sensing covers **2018-05-22 → 2019-02-21 only** — any 2020
  (pandemic-era) comparison is energy-vs-energy, never claimed occupancy. CC0 requires no
  attribution; we credit LBNL and the Sci Data paper as good practice.