# US grid demand, generation, TabICLv2 predictions since 2015

**Hourly electricity demand, generation by fuel and TabICLv2 demand predictions for U.S. grid operators since July 2015.** The main table covers the balancing authorities that report to the U.S. Energy Information Administration (EIA). TabICLv2, a zero-shot model for tabular data, predicts each hour EIA has not published by correcting the operator's day-ahead forecast. Three more tables track the…

US grid demand, generation, TabICLv2 predictions since 2015 is an open, machine-readable dataset published on Mostly Right as immutable Parquet versions, with a documented column schema, a JSON query API and a direct download. Its first rows are free to read and it is open to AI agents.

## Facts

- Slug: `us-electricity-grid-hourly-demand-generation-mix-and-corrected`
- Canonical page: https://mostlyright.md/datasets/us-electricity-grid-hourly-demand-generation-mix-and-corrected
- Publisher: mostlyrightmd (https://mostlyright.md/publishers/mostlyrightmd)
- Published: 2026-11-21T11:21:26.452Z
- Tables: 4
- Formats: Apache Parquet (download), JSON (query and sample)
- Cost: the first 500 rows of each table are free to everyone; querying and downloading need a free account and count against a monthly free allowance unless the workspace is subscribed
- Metadata API: `GET https://mostlyright.md/api/v2/public/datasets/us-electricity-grid-hourly-demand-generation-mix-and-corrected`
- Topics: electricity, power grid, demand, generation mix, renewables, forecasting, predictions

## Table directory

Tables follow the publisher's curated order. Display labels do not change table identifiers.

Start here: Hourly grid data — table id: 11c65b8c-4b35-4466-8540-e8a9d8cb7793

### Grid data

- Hourly grid data — table id: 11c65b8c-4b35-4466-8540-e8a9d8cb7793; published name: us_grid_hourly; https://mostlyright.md/datasets/us-electricity-grid-hourly-demand-generation-mix-and-corrected/tables?table=us-grid-hourly — One row per operator and hour since July 2015: demand, the operator's forecast, generation by fuel and trade, with predicted demand for the hours EIA has not published yet. Rebuilt daily.
### Prediction track record

- Live accuracy — table id: 80bbc630-6981-4bca-839e-219010c7bdb1; published name: us_grid_prediction_accuracy; https://mostlyright.md/datasets/us-electricity-grid-hourly-demand-generation-mix-and-corrected/tables?table=us-grid-prediction-accuracy — Each live prediction beside the demand EIA later published: both errors, band hit, and whether the correction beat the operator. Hours are scored one to two days after they are predicted.
- Prediction log — table id: 83ed1768-91cf-4101-8b16-917af68fa5d5; published name: us_grid_prediction_log; https://mostlyright.md/datasets/us-electricity-grid-hourly-demand-generation-mix-and-corrected/tables?table=us-grid-prediction-log — Every live prediction exactly as it was made, captured once per UTC day since 20 November 2026 and never rewritten.
- 12-week backtest — table id: 00668af4-bd14-4ca7-b672-727c057a3833; published name: us_grid_prediction_backtest; https://mostlyright.md/datasets/us-electricity-grid-hourly-demand-generation-mix-and-corrected/tables?table=us-grid-prediction-backtest — Pre-launch evidence: 24 August to 15 November 2026, each week predicted by models trained only on earlier hours, beside the published demand.

## How to use it

1. Read a table's schema — no account needed.
2. Fetch its published sample rows — no account needed.
   The first 500 rows of every table, in stored order, are free to everyone, with or without an account. Without an account you get that sample and ranked row search on public tables (search_table). An empty search query pages through a table in stored order. There is no filter, sort or download without an account.
3. Log in at https://mostlyright.md/login, connect this dataset to a workspace on its page, and create an `mr_use_` key.
4. Query bounded JSON, or download the complete Parquet snapshot and read it locally.
   A free workspace may read 100,000 rows and run 1,000 queries a UTC month past the sample. The allowance is shared by everyone in the workspace, and a free workspace may have any number of members. Reads count against the workspace whose data or key they use. A subscribed workspace is never counted, for anyone working in it.

Agents can do all of this over MCP at `https://mostlyright.md/api/mcp` — searching, reading schemas and sampling rows work with no credential at all.

## Table: Hourly grid data

- Table id: `11c65b8c-4b35-4466-8540-e8a9d8cb7793`
- Slug: `us-grid-hourly`
- Immutable version: `035c616f-5890-5554-80c0-1ab9169fa528`
- Published name: us_grid_hourly
- Section: Grid data
- Description: One row per operator and hour since July 2015: demand, the operator's forecast, generation by fuel and trade, with predicted demand for the hours EIA has not published yet. Rebuilt daily.
- Schema: `GET https://mostlyright.md/api/v2/public/tables/11c65b8c-4b35-4466-8540-e8a9d8cb7793/schema`
- Sample rows, no key: `GET https://mostlyright.md/api/v2/public/tables/11c65b8c-4b35-4466-8540-e8a9d8cb7793/preview`
- Bounded JSON query: `POST https://mostlyright.md/api/v2/public/tables/11c65b8c-4b35-4466-8540-e8a9d8cb7793/query` with `x-api-key: mr_use_…`
- Full Parquet snapshot: `GET https://mostlyright.md/api/v2/public/tables/11c65b8c-4b35-4466-8540-e8a9d8cb7793/current` with `x-api-key: mr_use_…` (application/vnd.apache.parquet)

### Columns (27)

| Column | Type | Description |
| --- | --- | --- |
| `balancing_authority` | string | EIA code of the balancing authority (grid operator), e.g. PJM, ERCO, CISO. |
| `region` | string | EIA region the balancing authority belongs to, e.g. MIDA, TEX, CAL. |
| `hour_utc` | timestamp_utc | Start of the hour, UTC. |
| `row_kind` | string | measured: EIA published demand for the hour. predicted: not published yet, so the model predicted it. not_reported: no demand and no prediction. |
| `is_prediction` | boolean | True only on rows whose demand figure is a model prediction, not a measurement. |
| `demand_mw` | number | Hourly demand the balancing authority reported to EIA. Empty on predicted rows. |
| `operator_forecast_mw` | number | The balancing authority's own day-ahead demand forecast. Empty where EIA has none, including hours it shows as 0 because the forecast has not been filled in ye… |
| `predicted_demand_mw` | number | Model median of demand for an hour EIA has not published: the operator's forecast corrected by TabICLv2. |
| `predicted_p10_mw` | number | 10th percentile of the model's prediction. |
| `predicted_p90_mw` | number | 90th percentile of the model's prediction. |
| `net_generation_mw` | number | Net generation inside the balancing authority, EIA-adjusted. |
| `interchange_mw` | number | Total interchange with neighbours, EIA-adjusted; positive is net export. |
| `gen_coal_mw` | number | Net generation from coal, EIA-adjusted. Empty before July 2018. |
| `gen_gas_mw` | number | Net generation from natural gas, EIA-adjusted. Empty before July 2018. |
| `gen_nuclear_mw` | number | Net generation from nuclear, EIA-adjusted. Empty before July 2018. |
| `gen_oil_mw` | number | Net generation from petroleum products, EIA-adjusted. Empty before July 2018. |
| `gen_hydro_mw` | number | Net generation from hydropower including pumped storage, EIA-adjusted. Empty before July 2018. |
| `gen_pumped_storage_mw` | number | Net generation from pumped storage alone (from July 2024), EIA-adjusted. Empty before July 2018. |
| `gen_solar_mw` | number | Net generation from solar, with and without integrated batteries, EIA-adjusted. Empty before July 2018. |
| `gen_wind_mw` | number | Net generation from wind, with and without integrated batteries, EIA-adjusted. Empty before July 2018. |
| `gen_battery_mw` | number | Net generation from standalone battery storage (from July 2024), EIA-adjusted. Empty before July 2018. |
| `gen_other_mw` | number | Net generation from other sources, geothermal and other storage, EIA-adjusted. Empty before July 2018. |
| `gen_unknown_mw` | number | Net generation from unknown fuel, EIA-adjusted. Empty before July 2018. |
| `temperature_c` | number | Air temperature at one representative point for the balancing authority (Open-Meteo), 2026 onwards. |
| `prediction_model` | string | Pinned model that made the prediction. |
| `model_trained_through` | timestamp_utc | Newest measured hour the model trained on. |
| `lead_hours` | integer | How far ahead the prediction is: hours after this operator's newest published demand hour. 1 is the first hour EIA has not published for the operator. Empty on… |

## Table: Live accuracy

- Table id: `80bbc630-6981-4bca-839e-219010c7bdb1`
- Slug: `us-grid-prediction-accuracy`
- Immutable version: `a8a54dd2-ca0d-5758-917d-6e0d094bb878`
- Published name: us_grid_prediction_accuracy
- Section: Prediction track record
- Description: Each live prediction beside the demand EIA later published: both errors, band hit, and whether the correction beat the operator. Hours are scored one to two days after they are predicted.
- Schema: `GET https://mostlyright.md/api/v2/public/tables/80bbc630-6981-4bca-839e-219010c7bdb1/schema`
- Sample rows, no key: `GET https://mostlyright.md/api/v2/public/tables/80bbc630-6981-4bca-839e-219010c7bdb1/preview`
- Bounded JSON query: `POST https://mostlyright.md/api/v2/public/tables/80bbc630-6981-4bca-839e-219010c7bdb1/query` with `x-api-key: mr_use_…`
- Full Parquet snapshot: `GET https://mostlyright.md/api/v2/public/tables/80bbc630-6981-4bca-839e-219010c7bdb1/current` with `x-api-key: mr_use_…` (application/vnd.apache.parquet)

### Columns (17)

| Column | Type | Description |
| --- | --- | --- |
| `captured_on` | date | UTC day the prediction was captured. |
| `balancing_authority` | string | EIA code of the balancing authority. |
| `region` | string | EIA region. |
| `hour_utc` | timestamp_utc | Start of the predicted hour, UTC. |
| `kind` | string | live: captured before EIA published the hour. |
| `predicted_demand_mw` | number | Corrected forecast as captured. |
| `predicted_p10_mw` | number | 10th percentile as captured. |
| `predicted_p90_mw` | number | 90th percentile as captured. |
| `operator_forecast_mw` | number | The operator's own day-ahead forecast at capture. |
| `prediction_model` | string | Pinned model. |
| `model_trained_through` | timestamp_utc | Newest measured hour the model had seen. |
| `lead_hours` | integer | How far ahead the prediction was: hours after the operator's newest published demand hour on the day it was captured (its first predicted hour is 1). Counted p… |
| `actual_demand_mw` | number | Demand EIA published; empty until it is published. |
| `prediction_error_pct` | number | Corrected forecast minus actual, percent of actual. |
| `operator_error_pct` | number | Operator forecast minus actual, percent of actual. |
| `actual_in_band` | boolean | Whether the actual fell inside the captured p10 to p90 band. |
| `closer_than_operator` | boolean | Whether the corrected forecast was closer to the actual than the operator's. |

## Table: Prediction log

- Table id: `83ed1768-91cf-4101-8b16-917af68fa5d5`
- Slug: `us-grid-prediction-log`
- Immutable version: `7daacaca-a212-56ad-9215-247f5b35451c`
- Published name: us_grid_prediction_log
- Section: Prediction track record
- Description: Every live prediction exactly as it was made, captured once per UTC day since 20 November 2026 and never rewritten.
- Schema: `GET https://mostlyright.md/api/v2/public/tables/83ed1768-91cf-4101-8b16-917af68fa5d5/schema`
- Sample rows, no key: `GET https://mostlyright.md/api/v2/public/tables/83ed1768-91cf-4101-8b16-917af68fa5d5/preview`
- Bounded JSON query: `POST https://mostlyright.md/api/v2/public/tables/83ed1768-91cf-4101-8b16-917af68fa5d5/query` with `x-api-key: mr_use_…`
- Full Parquet snapshot: `GET https://mostlyright.md/api/v2/public/tables/83ed1768-91cf-4101-8b16-917af68fa5d5/current` with `x-api-key: mr_use_…` (application/vnd.apache.parquet)

### Columns (11)

| Column | Type | Description |
| --- | --- | --- |
| `captured_on` | date | UTC day the prediction was captured from the live grid table. |
| `balancing_authority` | string | EIA code of the balancing authority. |
| `region` | string | EIA region. |
| `hour_utc` | timestamp_utc | Start of the predicted hour, UTC. |
| `kind` | string | Always live: captured before EIA published the hour. Backtest rows are in us_grid_backtest_12w. |
| `predicted_demand_mw` | number | Corrected forecast as captured (model median). |
| `predicted_p10_mw` | number | 10th percentile as captured. |
| `predicted_p90_mw` | number | 90th percentile as captured. |
| `operator_forecast_mw` | number | The operator's own day-ahead forecast at capture. |
| `prediction_model` | string | Pinned model that made the prediction. |
| `model_trained_through` | timestamp_utc | Newest measured hour the model had seen. |

## Table: 12-week backtest

- Table id: `00668af4-bd14-4ca7-b672-727c057a3833`
- Slug: `us-grid-prediction-backtest`
- Immutable version: `d52313ae-d680-5ba7-b9dc-a5bd26cdb51b`
- Published name: us_grid_prediction_backtest
- Section: Prediction track record
- Description: Pre-launch evidence: 24 August to 15 November 2026, each week predicted by models trained only on earlier hours, beside the published demand.
- Schema: `GET https://mostlyright.md/api/v2/public/tables/00668af4-bd14-4ca7-b672-727c057a3833/schema`
- Sample rows, no key: `GET https://mostlyright.md/api/v2/public/tables/00668af4-bd14-4ca7-b672-727c057a3833/preview`
- Bounded JSON query: `POST https://mostlyright.md/api/v2/public/tables/00668af4-bd14-4ca7-b672-727c057a3833/query` with `x-api-key: mr_use_…`
- Full Parquet snapshot: `GET https://mostlyright.md/api/v2/public/tables/00668af4-bd14-4ca7-b672-727c057a3833/current` with `x-api-key: mr_use_…` (application/vnd.apache.parquet)

### Columns (17)

| Column | Type | Description |
| --- | --- | --- |
| `week_start` | date | First day of the test week; the model trained only on hours before it. |
| `balancing_authority` | string | EIA code of the balancing authority. |
| `region` | string | EIA region. |
| `hour_utc` | timestamp_utc | Start of the predicted hour, UTC. |
| `kind` | string | Always backtest: made after the fact with data from before the week, never shown to anyone in advance. |
| `actual_demand_mw` | number | Demand EIA published for the hour. |
| `operator_forecast_mw` | number | The operator's own day-ahead forecast. |
| `predicted_demand_mw` | number | Corrected forecast (model median). |
| `predicted_p10_mw` | number | 10th percentile. |
| `predicted_p90_mw` | number | 90th percentile. |
| `prediction_model` | string | Pinned model. |
| `model_trained_through` | timestamp_utc | Newest hour the model trained on. |
| `lead_hours` | integer | How far ahead the prediction was: hours after the newest hour the model could see, the hour before the week began (1 to 168). |
| `prediction_error_pct` | number | Corrected forecast minus actual, percent of actual. |
| `operator_error_pct` | number | Operator forecast minus actual, percent of actual. |
| `actual_in_band` | boolean | Whether the actual fell inside the captured p10 to p90 band. |
| `closer_than_operator` | boolean | Whether the corrected forecast was closer to the actual than the operator's. |

## Citing this dataset

Cite the dataset by name and canonical URL (https://mostlyright.md/datasets/us-electricity-grid-hourly-demand-generation-mix-and-corrected), the table, and the immutable `version_id` the rows came from. A query response carries `table.version_id` and `table.content_digest`; a download carries a strong ETag and a `Content-Digest` header.
