Limits

gridprims gives screening estimates, not utility studies. No tool has yet been checked against real bills or a utility’s own figures, and your problem reports are how we find our errors. Tell us when a number looks wrong

Tools by how well checked they are: Unvalidated 0, Self-checked 6, Cross-checked 7, Field-verified 0. No tool here is field-verified. That needs real bills, utility figures, or practitioner checks, which only arrive once people use the site.

Generated Oct 31, 2026 from commit 276cf4e. The same record as JSON

Every tool, and how well checked it is

Stability, validation level, and practitioner review answer different questions: whether a tool may change, what it has been checked against, and whether a practitioner in its field has signed off on it. One does not imply another.

Tool	Stability	Validation level	Practitioner review	Evidence	What would raise it	Known limits
Energy unit converter calc.energy-units.convert	Stable	2 · Cross-checked	Review pending	Five worked examples, built from the cited NIST and EIA constants, pass within their tolerances. Btu, watt-hour, kilowatt-hour, therm, and quad conversions agree with NIST SP 811 Appendix B.8 to every figure NIST prints. Fuel conversions reproduce EIA's published averages, which have no independent source. last passed Oct 31, 2026	Three recorded real-world cases where a utility, a regulator, or a practitioner made the same conversion (for example a gas bill's therm-to-kWh line) would raise it to field-verified.	Fuel results use national averages; the gas or oil you buy can differ by a few percent. The coal figure is EIA's estimate for coal burned in power plants, not for home or industrial coal. Volumes and weights of fuels are not converted to each other except through their energy.
Emissions from electricity use calc.energy-units.emissions	Preview	2 · Cross-checked	Review pending	A worked example and the spec's 800 lb/MWh case (362.87 metric tons for 1,000 MWh) pass. CO₂e computed from each subregion's CO₂, CH₄, and N₂O rates with the technical guide's GWPs matches eGRID's own published CO₂e rate within 0.1% in all 27 subregions. last passed Oct 31, 2026	Three recorded cases where a utility or an emissions inventory used the same eGRID subregion rate would raise it to field-verified.	Average annual rates miss the hour-by-hour changes in the grid's mix; use hourly data when you have it. This is location-based accounting; contracts for clean power (market-based accounting) are not reflected. Rates are eGRID2023's; a newer edition will change them.
Power factor correction calc.power-factor.correct	Preview	2 · Cross-checked	Review pending	The spec's 236.8 kVAR, 641.0 and 526.3 kVA, and $1,384.62 penalty cases pass, and a target below the current power factor returns no correction. Two published worked examples (DOE and WSU/NEEA) are reproduced within 1 kVAR. last passed Oct 31, 2026	Three recorded installations where the sized bank and the bill change were checked against a utility's bills would raise it to field-verified.	The power factor is entered or derived from one kW and kVAR or kVA pair; a file with reactive readings gives a better basis. Only a tariff rule that scales billing demand to a power-factor threshold is priced here; kVA billing and per-kVAR charges are priced in the workbench's bill. Harmonics, resonance, and switching transients are not assessed.
Transformer sizing and loading calc.transformer.size	Preview	1 · Self-checked	Review pending	Hand-computed cases for single-phase, three-phase, and amps from kVA match to 0.1 kVA, and 498.8 kVA converts back to 600 A within 0.1 A. No freely published worked sizing example has been found to register. last passed Oct 31, 2026	A freely published worked transformer-sizing example from a utility or government source, reproduced within 0.1 kVA, would raise it to cross-checked.	Names no standard rating; the standard list is in a paid IEEE standard. Does not compute permissible overload time, hot-spot or top-oil temperature, or loss of life. Assumes a balanced three-phase load; unbalanced loads need each phase checked.
Battery duration and usable energy calc.storage.basics	Preview	1 · Self-checked	Review pending	The spec's 5.33 h, 0.1875 C-rate, 360 MWh, and 423.5 MWh cases pass. last passed Oct 31, 2026	A published battery worked example reproduced with the same convention would raise it to cross-checked.	Real duration depends on power level, temperature, and age; these are nameplate figures. Auxiliary loads and standby losses are not modeled.
Levelized cost of energy calc.cost.lcoe	Preview	2 · Cross-checked	Review pending	The spec's $41.10/MWh case passes, with each term shown. NLR ATB 2025's published LCOE for nine technologies (utility, commercial, and residential PV, land-based and offshore wind, geothermal, hydropower, biopower, and CSP) is reproduced from each case's own inputs within 1%. last passed Oct 31, 2026	Three recorded project cases where the LCOE was checked against a developer's or utility's own figure would raise it to field-verified.	A simple LCOE: no degradation, construction period, or tax detail beyond what the fixed charge rate carries. No defaults are offered; read the ATB for costs and capacity factors by technology.
Overhead conductor ampacity calc.line-rating.ampacity	Preview	2 · Cross-checked	Review pending	Higher wind raises the rating, 1,000 A at 230 kV is 398.4 MVA, and US and SI inputs agree within 0.01%. 50 cases spanning conductor size, surface, temperature limits, weather, elevation, latitude, bearing, date, and hour match two open IEEE 738 implementations, linerate 5.0.0 and pypacity 0.1.3, within 1% (the largest difference is under 0.3%). last passed Oct 31, 2026	Three recorded line ratings from a utility's published facility ratings, reproduced under the same assumptions, would raise it to field-verified.	Steady state only: no transient or emergency rating. Resistance must be entered from the manufacturer's data; no public-domain resistance table exists. Clear-sky solar gain; haze, industrial atmosphere, and ground reflection are not modeled.
Revenue requirement explainer ratepayer.revenue-requirement.explain	Preview	1 · Self-checked	Review pending	The spec's 7.5% cost of capital, $150,000,000 return, $33,333,333 income tax, and $13,333,333 and $0.89 one-point sensitivity pass. needs a fresh run	Reproducing a public rate case order's revenue requirement summary within 0.5% from its own inputs would raise it to cross-checked.	A simplified cost-of-service formula: no deferred taxes, tax credits, or riders. Figures from FERC Form 1 are total-company and may cover more than one state or service.
Service headroom electrify.service-headroom.check	Preview	2 · Cross-checked	Review pending	The spec's 480 V scenario (126.3 kVA, 157.9 kVA, 189.9 A, 210.1 A, 174.7 kVA) and a twelve-bill example pass. Two published worked examples (EC&M, 2014; Electrical Contractor, 2011) are reproduced to the whole amps they print: 123 A, 173 A, and 27 A spare; 929 A, 1,180 A, and 20 A spare. needs a fresh run	Three recorded cases where an electrician's 30-day recording and our figure from the same site were compared would raise it to field-verified.	Not a code compliance decision; conductor sizing and full load calculations are out of scope. Sites with solar, wind, a battery, or peak shaving cannot use the 30-day option.
Backup power size electrify.backup-power.size	Preview	1 · Self-checked	Review pending	The spec's 9.6 kWh load list, 60 kWh usable and 75 kWh nameplate at 12 hours, and 7.0 kW generator with motor start pass, each hand-computed. needs a fresh run	Matching REopt's resilience results for the same critical load and outage would raise it to cross-checked.	County outage history means some customers in the county were out, not this address; durations are a typical guide, not a worst case. No inverter losses, battery aging, or temperature derating.
Reliability indices reliability.outage-log.indices	Preview	1 · Self-checked	Review pending	The spec's 1.5 interruptions, 200.0 minutes, 133.33 minutes per interrupted customer, and 99.9619% availability pass, and the momentary split. needs a fresh run	Reproducing a published worked example of the sustained indices to its printed precision would raise it to cross-checked.	Indices counting how often each customer was interrupted need per-customer records an event log does not hold; they are reported as unavailable, never estimated. Major event days need daily figures over several years: the workbench computes them from a whole log.
Bill check household.bill-totals.check	Preview	1 · Self-checked	Review pending	Three worked examples from the spec's scenarios, and 21 URDB rates classified and converted by hand, pass to the cent. last passed Oct 31, 2026	Five published residential bill examples (state regulators' typical-bill tables) reproduced to the cent would raise it to cross-checked.	Taxes, local fees, and riders that the tariff data leaves out are not included, so a real bill is usually higher. Time-of-use energy is priced only from the kWh in each period as the bill prints them; a tiered time-of-use rate cannot be priced from totals. Demand charges need meter data and are not priced here. Export credits for net-metered months are not modeled; such a month prices only the charges that do not depend on kWh.
Is my use high? household.usage-benchmark.compare	Preview	2 · Cross-checked	Review pending	Two worked examples from the spec's scenarios pass, and three utilities' averages computed by hand from their EIA-861 rows match. State averages computed from EIA-861 2024 match EIA's published Table 5A average monthly residential consumption in all 51 states. last passed Oct 31, 2026	Three recorded real bills compared with their utility's published average would raise it to field-verified.	An average includes homes of every size, age, and heating type; a large or all-electric home can be well above it and use energy well. EIA-861 publishes no distribution of use, so this shows no percentile, rank, or "homes like yours". The averages are for the latest EIA-861 year, which may differ from the year of your bills.

What we do not compute

We decline these and say why, rather than estimate anyway.

Anything that needs a utility's model of the circuit

Fault current, interrupting capability, voltage and power quality, and transient stability.

These need the utility's own circuit model and device ratings, which are not public and are often protected infrastructure information.

docs/research/12-der-interconnection-screens.md, screens 2.2.1.4, 2.2.1.5, and 2.2.1.9
18 CFR 388.113(c)(2)

Nodal or locational prices

They need the grid operator's own network model, which we will not publish, and most operators forbid redistributing their price data.

docs/research/15-browser-grid-modeling.md §6
docs/research/13-governance-and-deferred-items.md, item 13h

Resource-adequacy figures

Loss-of-load expectation, expected unserved energy, or an adequacy verdict. The reliability indices we compute from your own outage log are a different thing.

Unit-level forced-outage data is confidential, the only freely locatable class averages we found are a 2010 vintage, and one operator's otherwise-free feed forbids redistribution.

This absence could not be fully confirmed: Whether NERC publishes a current free class-average download could not be confirmed; its pages did not render outside a browser.

docs/research/15-browser-grid-modeling.md §5 (NERC GADS confidential; ISO-NE class averages, 2010 vintage)
docs/research/13-governance-and-deferred-items.md, item 13f

A pass or fail interconnection verdict

Six of the twelve standard screens, including every screen that needs a model of the circuit, cannot be answered from public data.

docs/research/12-der-interconnection-screens.md, summary

Utility studies, engineering sign-off, or legal, tax, or investment advice

gridprims gives screening estimates; a utility study, a licensed engineer, or a professional adviser is the final word.

PRODUCT.md, Capabilities and Constraints

Real-time or operational grid data

Live feeds need servers and most operators restrict redistribution; we publish dated snapshots and link to live sources.

openspec/changes/plan-launch-and-value-proof/design.md D4

Forecasts of prices, loads, or peak days

Public data can describe what happened but cannot support a forecast we would stand behind.

PRODUCT.md, Capabilities and Constraints

Anything that needs protected infrastructure information

We publish only public data and never Critical Energy/Electric Infrastructure Information.

18 CFR 388.113
openspec/changes/establish-platform-foundation/specs/platform/data-snapshots/spec.md, CEII screen

Feeder-level or address-level results that we publish

What you compute from your own data on your own device is yours; we do not publish such results.

Feeder and address detail can identify customers and infrastructure, and the data behind it is not public in most places.

docs/research/12-der-interconnection-screens.md, minimum-load data

Known problems

Is the automation running?

Every data refresh and check runs on a schedule with nobody in the path. This is when each last succeeded. Your device judges whether it is overdue.

Scheduled job	Runs every	Last success
refresh-data	24 hours	2026-10-31 15:21 UTC
link-check	168 hours	2026-10-31 02:14 UTC
canary	168 hours	2026-10-31 02:15 UTC
heartbeat	24 hours	Not yet

The full record

Late data

2020 ZCTA to County Relationship File (U.S. Census Bureau) was due Dec 9, 2022 (judged from its usual schedule); 1422 days late.
Emissions & Generation Resource Integrated Database (eGRID) was due Jan 31, 2026 (posted date); 273 days late.

No data refresh is failing.

Not yet populated: The corrections log is not yet populated; add-trust-and-continuity fills it.

Not yet populated: The tariff conformance suite will list the charge types its reference engines cannot check.

Claims we rely on but have not confirmed

Each of these comes from our research, is still unconfirmed, and is used by the page named.

Claim	Used by	What would confirm it
PG&E's Green Button CSV starts with about five metadata rows, then TYPE, DATE, START TIME, END TIME, USAGE (kWh), COST, NOTES. docs/research/07-methods-and-formats-verification.md §3	/files/ (file intake recognizes this layout)	One real PG&E export, with personal fields removed, sent through a problem report.
Smart Meter Texas 15-minute exports carry ESIID, usage date, start and end times, kWh, an estimated flag, and a consumption or surplus column. docs/research/09-audiences-and-data-access.md §1a	/files/ (file intake recognizes this layout)	One real Smart Meter Texas export, with the ESIID removed.
EIA posts the next EIA-861 release each August (an inferred date, not a posted one). docs/research/01-data-sources-and-licensing.md, EIA	/data/eia-861/ (the expected next release)	EIA posting a release date for the 2026 data.
The 2026 NEC moves 220.87, Determining Existing Loads, to 120.87, with the same 1-year and 30-day options and the same bar on the 30-day option for sites with solar, wind, or peak shaving. docs/derivations/service-headroom.md (Mike Holt's 2026 table of contents; Minnesota's 2026 adoption review)	/tools/electrify/service-headroom/check/ (the 2026 edition option)	Reading 120.87 in the 2026 NEC through NFPA's free access.

What the levels mean

0 · Unvalidated: The method is cited, but nothing independent has checked it yet.
1 · Self-checked: Our own worked examples and consistency tests pass, but no independent check has been run yet.
2 · Cross-checked: Results agree with an independent implementation or a published worked example within a stated tolerance.
3 · Field-verified: Results agree with real bills, utility figures, or practitioner calculations in at least 3 recorded cases.