Munim Technologies

Research/Energy

An audit of the US Utility Rate Database

Current residential tariffs in NREL's Utility Rate Database reproduce what households paid, with a customer-weighted median ratio of 0.98. Tariffs entered before 2023 bill a median 0.81, and 1,726 of 63,298 records cannot be billed as documented.

Oct 16, 2026 · 6 min read

The US Utility Rate Database (URDB), published by NREL on OpenEI, holds more than 63,000 electricity tariffs. REopt, SAM, dGen and many bill calculators use it to price electricity for solar, storage and efficiency studies. We audited the whole bulk file and checked its residential tariffs against what households actually paid. Where the entries are current they work well: billing each utility's standard residential rate at its average household use gives a customer-weighted median ratio of 0.98 to EIA-861's realised price. Rates entered before 2023 run low, at a median 0.81.

The structure is mostly sound too. 1,726 of 63,298 records (2.73%) cannot be billed as documented, but only 161 of 32,339 current ones (0.50%). The practical risk for users is not broken records so much as old ones: off OpenEI's list of annually updated utilities, most customers sit at a utility whose newest residential rate is from before 2023 or missing.

0.98

customer-weighted median ratio of URDB bill to EIA-861 price, 883 core utilities

0.81

median ratio when the chosen rate took effect before 2023

2.73%

of 63,298 records cannot be billed as documented (0.50% of current ones)

6.9%

of records are unapproved, and the bulk JSON file includes them

The question

A URDB tariff is billed by looking up, for each hour of each month, a period index in a 12 by 24 weekday or weekend schedule. The index selects a period in the rate structure, and that period's tiers set the price. If a schedule points at a period that does not exist, or a tier can never be reached, a calculator has to fail or guess. We asked three things: how many records can be billed as written, how current the residential rates are, and whether the prices they produce match what people paid.

What we did

A structural audit read the bulk file on 15 October 2026 and checked every record, past and current, in every sector including lighting. It separates errors, which make a rate unbillable as documented, from review flags, which are legal but implausible. A price check, written up with eight predictions before it ran, then chose a standard residential rate for each utility without looking at any price, billed it at the utility's average monthly use, and divided by EIA-861's 2025 bundled residential revenue per kWh. Further checks weighted freshness by households, counted the review queue, compared each record's prices with the same utility's other records, and looked for records whose name contradicts their sector.

What we found

Most all-time errors are end dates before start dates (1,197 records, none current). Among current records, the largest classes are an unreachable tier after an open-ended one and a demand schedule pointing at a period that does not exist. All 18 current Cleveland Electric Illuminating rates carry the same unreachable-tier error.

Largest error classes in the bulk file

CheckAll recordsCurrent records
End date before start date1,1970
Energy tier unreachable after an open tier30588
Demand schedule uses a missing period17344
Energy tier limits not increasing2212
Energy schedule uses a missing period145
Flat-demand months use a missing period148
Any error1,726 of 63,298161 of 32,339
Current means no end date, or one after 2026-10-15. Source: urdb-audit results/2026-10-15/audit.md.

The price check covered 883 core utilities, where at least 90% of residential customers are bundled, and 87% of their 99.7 million households. 88% of priced households are at a utility whose URDB bill is within 25% of EIA's figure. The spread is mostly about age. Rates effective in 2025 or later reproduce the price (median 1.04, 95% within 25%). Rates from before 2023 cover 767 utilities and 22.4 million households, and 32% of those households are at a utility whose tariff bills under three quarters of what they paid.

Median ratio of URDB bill to EIA-861 realised price, by group

Customer-weighted medians. A ratio of 1 means the URDB tariff, billed at average use, matches what households paid on average in 2025. Source: urdb-audit: results/2026-10-16/price.md

The largest outliers were entry errors, found by hand. Elmhurst Mutual (Washington) and Tri-County (Oklahoma) carry energy prices of 0.503 and 0.728 dollars per kWh in 2014 records; the co-ops' own pages give 0.069 and 0.1095. Mississippi Power's current record has a fixed charge of 8.34 dollars per day where its R-60 schedule says 1.23.

Freshness follows the same line. Of 133.8 million in-scope residential customers, 67.9% are at a utility whose newest URDB residential rate took effect in 2025 or later. On OpenEI's annual-update list the figure is 85.8%; off it, 7.6%. The review queue does not explain the gap. 4,360 records have approved set to false, but approving all of them would lift the in-scope share only to 72.7%, and off the list from 7.6% to 8.5%. Those rates were mostly never entered.

Residential customers at a utility with a 2025+ URDB rate

ScenarioAll in scopeOn OpenEI listNot on list
Approved records only67.9%85.8%7.6%
Plus unapproved records72.7%91.7%8.5%
Source: urdb-audit results/2026-10-16/pending.md.

Two smaller findings matter for anyone loading the data. Both bulk links say they download all approved rates. The CSV is exactly the 58,920 approved records; the JSON file also holds the 4,360 unapproved ones and 18 with no flag. And 250 of 14,707 Residential records (1.7%) are named as something else, such as general service or irrigation, mostly in records created before 2015.

What it means

For current tariffs at large utilities, URDB is a good source. For the long tail of co-ops and municipal utilities, an old effective date is common, and an old tariff tends to understate today's bill by about a fifth. Studies that price electricity from URDB should check the effective date of each rate they use, filter the JSON file on its approved field, and treat a ratio far from EIA's average as a reason to look. We publish the record-level flag lists so the database itself can be corrected.

More research