Real Price Desk

Method

How Real Price Desk gets its data, computes its numbers, uses AI, and handles errors.

Updated

On this page 10 sections
  1. What we do
  2. Who makes it
  3. Sources and terms
  4. Formulas
  5. Definitions we flag in every post
  6. Missing and revised data
  7. Falsifiers and the scorecard
  8. Reproducibility
  9. What we do not do
  10. Updates to this page

What we do#

Real Price Desk takes public US statistics, recomputes the headline figures itself, and adds one computed layer the source does not publish: for example inflation-adjusted prices, the share of months above a baseline, or the gap between two series. Every post says in one sentence what is new compared with the source, states what would prove the main claim wrong, and flags where definitions differ.

Who makes it#

Posts are bylined Editorial by GrandPrixx, the named reviewer. Content is written with Grok Bot, an AI. Grok Bot downloads the data, runs the analysis scripts, builds the charts and drafts the text. Before any post is published, GrandPrixx checks the headline numbers against the raw files, reviews the wording, and gives an explicit go. Nothing is published without that go. No other author is implied and no author profiles exist. A short disclosure sits at the end of every post, and the same facts are on the About and disclosure page.

Sources and terms#

  • U.S. Energy Information Administration (EIA): Electric Power Monthly and Short-Term Energy Outlook. EIA says its publications are in the public domain and asks users to acknowledge the source with the publication date. We cite the table or series ID and the release date on every page.
  • U.S. Bureau of Labor Statistics (BLS): Consumer Price Index and average hourly earnings, retrieved through the BLS Public Data API. BLS says what it publishes is public domain and asks for citation. Its API terms require us to state the retrieval date and this sentence: BLS.gov cannot vouch for the data or analyses derived from these data after the data have been retrieved from BLS.gov.
  • U.S. Department of Agriculture, Economic Research Service (ERS): Food Price Outlook. Planned for later posts. Our source log records only partial verification of USDA reuse terms: the USDA policies page could not be opened when we checked, so the terms are not verified. We will verify them before using ERS data in a post.

We use no agency logos or emblems. Real Price Desk is not affiliated with or endorsed by EIA, BLS or USDA. Our calculations are ours, and we say so.

Formulas#

  • Real price = nominal price x (CPI-U all items in the base month) / (CPI-U all items in the price month). The base month is stated in every post. Post 1 uses July 2026 dollars and series CUUR0000SA0.
  • Percent change = (new / old - 1) x 100. Real-price calculations and baseline comparisons use unrounded values. Percentage changes are stored to six decimal places before gaps are calculated, and other outputs use field-specific rounding. The post rounds these stored results for display. The July comparisons and year-over-year changes compare the same calendar month across years to reduce seasonal effects. The baseline comparison is different: it uses a mean of all 48 months of 2016 to 2019 and is therefore not seasonally neutral. Across those years July and August averaged above the all-month mean, so a single summer month compared with it carries a seasonal component, and posts say so. The monthly baseline and the charts use data that are not seasonally adjusted.
  • Months above baseline = the count of months in a test window whose real price is above the baseline average (or baseline peak), divided by the months with data.
  • Year-over-year gap = EIA year-over-year percent change minus BLS CPI electricity year-over-year percent change, in percentage points.
  • Hours of earnings = (price in cents per kWh x quantity in kWh / 100) / average hourly earnings. The quantity is a stated round number, never an average household.
  • Estimate versus first published observation (plan versus operated) = a recorded STEO estimate or forecast compared with the first published Electric Power Monthly estimate, with both publication vintages named. The first EPM estimate remains subject to revision. We record the forecast and the test threshold before the data arrive and score both afterwards.

Definitions we flag in every post#

Residential versus commercial versus industrial versus all-sector prices; nominal versus real, and which deflator; EIA average price (utility revenue divided by kWh sold) versus the BLS CPI index; food at home versus food away from home versus ERS categories (for later posts); preliminary versus final data.

Missing and revised data#

If a value is missing we write not found and leave it out of calculations. We never fill it in. Example: BLS did not publish October 2025 CPI because of the 2025 lapse in appropriations. EIA marks recent months preliminary and may revise them; we label preliminary values and log changes to published numbers on the Corrections log. To report an error, email corrections@realpricedesk.com.

Falsifiers and the scorecard#

Each post separates checks of historical calculations from follow-up tests that use later observations, and says when each check can be made. The outcome of each follow-up test is recorded on the Corrections log under Falsifier checks, whether we were right or not. Thresholds for scoring a forecast, such as a tolerance in cents, are editorial choices and are labeled as such.

Reproducibility#

Downloaded source files are stored with their source address, retrieval time and checksum. The five-series EIA JSON file is an extract from the preserved bulk ZIP and is identified as derived in the manifest. A script recomputes the headline numbers from those files and writes the results to a data file; a second script draws the chart; a third checks selected numeric strings and several counts in the post against the data file. That check does not verify every number, its context, dates, units or interpretation, so a human review is still required. Raw files and scripts are kept in a repository that is currently private.

What we do not do#

  • No investment, health, legal or personal-finance advice.
  • No pages generated in bulk for states, cities or ZIP codes.
  • No claims about causes unless the source gives them; otherwise we label a hypothesis.
  • No author names we cannot back up. No fabricated people. The only named person is the reviewer, GrandPrixx, and we describe the role of GrandPrixx as it is.
  • No advertising or tracking scripts at launch.

Updates to this page#

Last reviewed 2026-10-19. Changes to this page are logged on the Corrections log.