How the numbers are made

Projects and hybrids

A project is one map entry. When one EIA plant ID reports both operating generation and batteries, it is one hybrid project. A separate storage-only plant joins a generation plant only when all of these hold:

• its 2025 EIA-860 storage record says it directly supports another unit;
• the named plant and generator exactly match an operating generator at a mapped generation plant;
• both plants are in the same state and within 3 km of each other;
• every link from that storage plant points to the same single target.

Distance alone never merges plants. The current catalog has 14,233 EIA plant records grouped into 14,225 projects, 529 of them hybrids. The project ID is the parent EIA plant ID; component IDs resolve to it, stay searchable, and keep their own records in the Details tab.

Project capacity is EIA-860M nameplate MW: generation MW plus battery MW. Fuel totals sum disjoint generator ratings, and fleet totals count each project once. Battery energy (MWh) and maximum discharge rate come from exact matches in the 2025 annual storage schedule.

The operating date shown is the earliest reported operating year and month among a project’s operating units.

Map locations and plant footprints

Markers use the EIA coordinates of the plant’s operating units. For 714 solar and wind plants whose EIA point lies more than 1 km outside their own mapped footprint, the marker moves to a display centroid: the area-weighted centroid of the plant’s USGS solar array polygons, or the centroid of the convex hull of its USGS turbine points. The original EIA coordinates stay in the plant details.

Footprints join on the source’s own EIA plant ID (USGS eia_id, OpenStreetMap ref:US:EIA), never on name or distance. 11,446 current plants link to 87,119 records: 6,619 solar arrays, 70,532 turbines and 9,968 OpenStreetMap plant outlines. Records too far from the EIA point, or with a state or technology conflict, go to review instead of being matched.

Capacity factor

The denominator is the nameplate in service at the time, not today’s: same-year EIA-860 ratings for 2010 to 2025, and same-month EIA-860M ratings in 2026. So capacity factors start in 2010. A month gets no factor when the dated inventory cannot support the whole month, for example a commissioning, retirement or unresolved uprate in that month, or a missing rating. Current capacity is never substituted.

For hybrids, battery net output is subtracted and battery MW is left out of the denominator. Pumped storage gets no factor. Values below 0% or above 100% stay visible. This differs from EIA’s published capacity factor, which uses net summer capacity.

Battery flows use battery rows only: charging, gross discharge and storage net. Battery utilization (%) = discharge MWh ÷ (dated battery MW × hours) × 100. A discharge-to-charge ratio can be calculated for a period, but it is not round-trip efficiency.

ERCOT hourly generation

For ERCOT projects in the reviewed crosswalk, the Hourly view shows ERCOT settlement-metered net energy (SMNE) for the project’s reviewed resource IDs. An hour needs four complete quarter-hours from every contributing resource; otherwise the combined hour is unavailable. Combined-cycle train readings take precedence over component readings so nothing is counted twice. Days follow the ERCOT operating day (Central time).

The crosswalk covers 611 grouped projects above 20 MW in ERCOT. 589 have reviewed resource IDs and nodes for every eligible generator, 7 are partial and 15 unresolved. Identity needs evidence tying the EIA plant to named ERCOT resources; name similarity and nearest nodes do not count. Daily capacity factor in this view uses current nameplate and needs every hour.

Solar

NSRDB irradiance, temperature and wind (ERA5 where NSRDB is unavailable) drive a model of plane-of-array irradiance, cell temperature, DC output and inverter clipping, using EIA-860 equipment and tracking data. Assumptions: angular reflection losses, 0.5% a year module aging from commissioning, and a 10% DC loss allowance that includes 2% soiling.

Onshore wind

HRRR winds where available and ERA5 elsewhere, interpolated to hub height and adjusted for air density. Equipment comes from EIA-860 and the USGS turbine database, with published power curves where a turbine model matches and stated proxies where it does not, minus a 10% farm-loss allowance.

Recent days

From September 1, 2026 a recent series uses NOAA GOES satellite irradiance and HRRR 80 m wind scaled to hub height. It refreshes about every six hours, with roughly a three-hour source delay.

Thermal

June to September use EIA net summer capability, December to February net winter capability, and other months the midpoint. Dated EIA-860 and EIA-860M inventories set which generators count. Thermal potential is monthly only and covers 3,473 projects from January 2010 to August 2026.

Other technologies and coverage

Solar thermal (CSP) uses a separate direct-normal-irradiance, turbine and storage model checked against NREL SAM. Conventional hydro is modeled only inside mixed plants and assumes enough water. Historical solar and wind files cover 2010 to 2025; a 2026 archive runs to October 1, 2026 on equipment through August, with September carrying the August inventory forward as a stated assumption. Where equipment timing or ratings are uncertain, the chart shows a sensitivity range (the sum of component bounds, not a confidence interval).

A continuity check of the recent series on four 2025 days (160 projects) found regional solar energy within −3.8% to +11.4% of the reviewed archive and wind within −4.7% to +6.9%. That check does not measure annual accuracy.

How it is calculated

For a target month in year Y, training uses only reported months up to September of Y−1. Each training month gives a ratio r = actual MWh ÷ potential MWh. The plant ratio R is the potential-weighted mean of r over a trailing window, shrunk toward a peer prior P (the median ratio of other plants with the same technology in the same state, or nationally when the state has fewer than 10):

A same-calendar-month ratio S adds seasonality once a plant has at least 12 training months and two observations of that month:

Settings from 2020 on are fixed per technology, chosen on January 2017 to September 2019 data: solar 60-month window, s = 0.5, k = 12; wind 36 months, s = 0.35, k = 6; thermal 36 months, s = 0.65, k = 0. Plants that report annually train only on complete annual totals at least two years old. Plants with little history fall back to peer priors, then to an annual-output prior, then to cold-start priors built from comparable plants’ earlier annual capacity factors. Cold-start estimates are capped at potential and carry a sensitivity range, which is not a calibrated probability interval.

Validation

WAPE = Σ|expected − actual| ÷ Σ|actual|. Signed bias = Σ(expected − actual) ÷ Σ|actual|, positive when the model overestimates. On retrospective months from 2010 to the latest available 2026 month:

How the estimate works

For each plant-hour, the model scales physical potential by a response learned from ERCOT plants:

Prices fall into 13 bins with edges at −100, −50, −30, −23, −10, −5, 0, 5, 20, 50, 100 and 500 USD/MWh. The credit state is the share of plant capacity inside an assumed production tax credit window. 250 fitted ERCOT plants (145 wind, 105 solar) use their own response; every other plant uses the pooled response. Expected curtailed hours is the sum of hourly event probabilities, not an observed duration. Monthly totals sum valid hours in the plant’s local month, and a month is complete only when every hour has an estimate.

The response was fitted on January to April 2026, on dispatch-reduction estimates built from ERCOT’s public 60-day dispatch disclosures (high sustained limit minus telemetered output). Plants whose EIA ID is divisible by five were held out. On July 1 to 28, 2026, fitted wind plants were about 27% low against the reference and fitted solar plants about 27% high. Held-out plants transferred weakly.

The cohort is 1,943 plants in the seven ISO and RTO markets (896 wind, 1,047 solar): CAISO 442, MISO 386, PJM 369, ERCOT 365, SPP 263, ISO-NE 72, NYISO 46. A plant needs a reviewed resource-level pricing node. Mixed-fuel, storage, concentrating solar, offshore wind and zone-priced plants are excluded. Hours run from January 2014 where potential and a complete real-time node price overlap.

With and without the production tax credit

Plants inside a production tax credit (PTC) window earn the credit only when they generate, so they tend to keep running at more negative prices. Credit receipt cannot be observed, so the model shows two bounds:

• Lower bound, with PTC (solid line): each generator earns the credit for 120 months after its original EIA operating month, for wind first operated after 1993 and solar after 2021.
• Upper bound, without PTC (dashed line and band): the same response with no credit in any hour.

Outside a credit window the bounds are equal and no band is drawn. A step in the lower bound when a window ends comes from this assumption, not from observed behavior. Across the covered fleet, 2025 totals 23.0 TWh with the credit and 48.4 TWh without it.

Fleet screens and comparisons

The agent and the MCP server compare modeled curtailment across the whole eligible cohort before ranking. Comparisons use matched calendar months in both periods, need at least 95% valid hours each month, and exclude plants without baseline history (new plants are reported as new entrants).

A change splits exactly into a volume effect, (P_c − P_b) × (r_b + r_c) ÷ 2, and a rate effect, (r_c − r_b) × (P_b + P_c) ÷ 2, where P is potential and r is loss share in the baseline (b) and comparison (c) periods. The split is arithmetic and does not identify a cause.

Operator-reported regional history

Operator-published totals are served separately as market context: CAISO monthly wind and solar curtailment, 2015 to 2025, from its Monthly Renewables Performance Report; and SPP’s annual mean curtailed MW for wind and solar combined (1,483 MW in 2024, 1,382 MW in 2025) from its market monitor’s State of the Market report. In an October 2026 check, the model’s covered CAISO solar fleet was higher than CAISO’s reported totals and moved in the opposite direction from 2024 to 2025. Fleet scope and definitions differ, so this is a diagnostic, not a plant error rate.

Node assignment

A plant’s pricing node is the market location whose published price InfraMap shows for that plant. Assignments are reviewed against ISO registries (some carry the EIA plant ID), the plant’s own EIA-860 node designation when it equals a current official node, ISO unit registries matched on name plus at least two independent attributes, and FERC EQR delivery points. ERCOT plants use a reviewed crosswalk from EIA generators to ERCOT resources. Nearest-node and name-only matches are rejected, and a zone, hub or load node never counts as a resource node.

An October 16, 2026 review gave every plant in the seven markets one of four outcomes: a pricing node; a zone or load price location, labeled as such; no node, with a sourced reason (for example net metering, a distribution program, behind-the-meter load or a pending registration); or unresolved, with a note on what was checked. 97% of market capacity has a price location.

For the largest unresolved plants, research added nodes in labeled tiers, including a reviewed-inference tier where no published source names the node for the plant. Where a plant’s EIA balancing-authority code points to the wrong market, the plant is assigned to the market that prices it. The Prices tab shows each plant’s outcome and its sources.

Comparison hubs

Every priced node gets a comparison benchmark. A published hub membership wins. Otherwise the node gets a disclosed regional comparison. The card and Prices tab name the relationship, for example “Published hub member” or “Same-region comparison hub, not membership”. A hub chosen by the user overrides the default.

Basis

Positive basis means the node priced above the hub; negative means below. A node at $70 and a hub at $50 give +$20/MWh. Average basis uses only hours where both prices exist, and monthly and yearly values weight every hour equally. Zero stays zero, negative prices stay negative, and a missing hub is never replaced by another hub.

Day-ahead and real-time

Day-ahead (DA) and real-time (RT) are separate series and never substitute for each other. Values are total LMP (ERCOT: settlement point price) at hourly grain. Where an ISO publishes sub-hourly RT prices, the hour is the mean of its intervals and needs every interval. Hours are keyed to UTC, and monthly and yearly periods use each market’s local calendar.

A scheduled refresh adds recent days and revisits them 1, 2, 7 and 30 days later, so final prices replace preliminary ones. Archive starts are market-wide; each node’s history starts when the ISO first published it.

Price chains

When an ISO renamed a node, one market replaced another, or a plant was re-registered, the plant keeps one continuous price history: a price chain of dated segments, each one node in one market. Each hour comes from exactly one segment and is paired only with that segment’s hub, so basis is never computed across a node change. Charts mark each change with a dashed line.

A link counts only when a source ties the earlier node to the plant for that window, such as an ISO file showing old and new IDs on the same resource on adjacent days. 590 chains cover 340 plants (92.9 GW), including SPP’s move from the Western Energy Imbalance Service (WEIS) to the SPP RTO on April 1, 2026. Models use predecessor hours only after the plant’s latest repower or physical change.

Map price layers

Prices need a free account. Free accounts see every price and basis with each node under an alias; node names are part of Pro. Node points show the average DA or RT price at each location for a week, month or year, from January 2025. Means use complete hourly prices; a location with missing hours is marked partial. Some nodes have no published location and are drawn at the linked plant. The node card compares a node with its hub when both are complete for the period.

The heatmap interpolates complete-period node averages: each grid point uses up to eight nearest price series from the nearest market within 180 km, weighted by (1 − d²/R²)² ÷ d², where d is distance and R is 180 km. Markets never blend. Colors run through fixed anchors at −50, 0, 25, 50 and 100 USD/MWh.

EQR hourly sales

For 431 plants, the Hourly view can show hourly FERC EQR energy sales for Q2 2026. Sales streams link through reviewed delivery points or through exact FERC company IDs to a single EIA asset. Only energy quantities count, a row counts only when its interval falls within one hour, and multi-hour sales are never apportioned. The chart sums all audited streams for each hour; an hour with an incomplete interval is unavailable.