whopays: the frozen files, unchanged
=====================================

This folder holds every file listed in prereg-v1.sha256, byte for byte, in the same folder layout
the list uses (analysis/, pipeline/, docs/, data/). From inside this folder:

    sha256sum -c prereg-v1.sha256

should print OK for all 47 files. The fingerprint of prereg-v1.sha256 itself was timestamped with
OpenTimestamps (prereg-v1.sha256.ots) before the test was run. The run log (RUNLOG.md, in /proof/)
records the fingerprints of the four input files it used, and of the two extra files below.

Re-running the test: the code in analysis/ and pipeline/ is here, with a fixed random seed, so
running analysis/run.py on these exact files gives the same numbers. (Its safety guards expect the
original git history and will refuse a copy without it; to re-run, read the guard messages, or ask us.)

Also here, not in the 47 but fingerprinted in the run log:
  data/derived/fetch_manifest.csv
  data/derived/panel_provenance.json

WHERE THE DATA COME FROM
  * data/raw/eia861m, eia923, eia_gas: U.S. Energy Information Administration (EIA) releases.
    EIA data are U.S. government works in the public domain. Please cite EIA.
  * data/derived/exposure_pjm_2023.csv and exposure_pjm_2024.csv: numbers transcribed by us from
    PJM Interconnection's Load Forecast Reports (data-centre forecast, tables B-9 and B-10). PJM is a
    private company, not EIA; credit PJM. PJM's own reports (PDFs and workbooks) are NOT republished here.
  * the other data/derived files: tables we built from the EIA files above with the code in pipeline/.

A CORRECTION TO fetch_manifest.csv
  Its "license" column says "US government work, public domain (EIA)" on every row, including the rows
  whose source begins "phase1:pjm/". Those rows describe PJM publications, not EIA, and the licence label
  is wrong for them. The file cannot be edited because its fingerprint is recorded in the run log.

WHAT THE GROUPS MEAN
  Where a file marks a utility as "High" (fast-growth) or "Low" (comparison), that describes how fast
  business electricity use grew from 2019 to 2022 compared with other utilities in the same region. It is
  not a finding about that utility. The test compares groups; it says nothing about any one utility or household.

This is a first test of one question, with the limits stated on the site (whopays.colinduncan.xyz).
Results are not a finding about causes. CSV files are plain numbers; none of their cells begins with a
spreadsheet formula character (checked), and they should be opened as data, not trusted as programs.