← The record · All reference data

Grid & consumer energy-cost reference — US EIA

Committed consumer energy-cost baseline for a site’s corridor: what households in that site’s state pay for electricity and heating fuel, which utility and balancing authority serve the campus, and the scale of the data-center load against those markets. The consumer-price half of the demand thread (issue #91, from the 2026-06-10 facility-design call: “bring in fuel costs at the consumer level due to macro pressures and data-center demand”). Regenerated by watermark eia / watermark grid; the site/scenario read the committed YAML (no live pull).

Every state-, utility- and BA-level value below is resolved per site, not baked in: the reference build keeps the un-slugged reference/eia/*.yaml, and every other site’s copies live under reference/eia/<slug>/. A site in Indiana pulls the Indiana series and its own serving utility; nothing about one site’s grid is asserted of another’s.

Files

The demand → consumer-price-pressure sensitivity

watermark.economics.energy.derive_demand_pressure links the facility’s first-class total facility_draw (watermark.facility.power, issue #87 — IT load × PUE) to this dataset:

• Annual consumption = facility draw × 8760 h × load factor (~0.9, assumption).
• Demand share = that consumption ÷ the site state’s total retail electricity sales (EIA).
• Households-equivalent = consumption ÷ ~10,500 kWh/household/yr (EIA US avg).
• Price-pressure band = demand share × a stylized demand-to-price transmission coefficient (0.5–1.0 %price/%demand).

The demand share and households-equivalent are the robust, EIA-cited headline. The price-pressure band is a stylized screening sensitivity, not a forecast — retail price formation is far more complex than one coefficient.

Known gaps & caveats

• EIA requires a key (WATERMARK_EIA_API_KEY, free). Without one, watermark eia replays the committed offline fixtures (tests/fixtures/economics/eia/); a live keyed pull refreshes to the latest annual vintage. Figures here are EIA annual averages (latest stable annual; EIA annual data lags ~1 year). The committed fixtures + figures are now a genuine recorded keyed pull — 2025 vintage (watermark eia, #120), superseding the earlier hand-transcribed 2023 placeholders.
• The campus buys at wholesale/industrial rates, not the residential price. The residential price is the consumer-impact reference, not the campus’s own bill.
• Retail fuel is state-level natural gas, not gasoline. EIA retail gasoline is PADD/region-level (Midwest PADD 2), not state; we use the site state’s residential natural gas ($/Mcf) as the state-level consumer heating-fuel series and do not fabricate a county/state gasoline figure.
• Series values are connector-pulled and regenerable; EIA columns are selected by name — period plus the series’ own value column (price / sales / value on the /v2/seriesid route), never by index.

grid-profile.yaml specifics (issue #94)

• The serving utility is grounded, never asserted. Each grid-profile.yaml carries the utility with its own source and confidence: document where the site’s record names a tariff outright, reference where the identification rests on the EIA-861 service-territory file plus the state commission’s certified-territory map. A site whose serving LSE is not document-cited says so on the value rather than borrowing a neighbour’s.
• The RTO/BA is read off the utility, not chosen — a utility in a PJM transmission zone is FERC-jurisdictional to PJM, and the profile records which BA applies to that site.
• All three load denominators are connector-sourced (#94/#120). Utility retail GWh + customers + avg price come from the EIA-861 per-utility “Sales to Ultimate Customers” bulk file (watermark.grid.eia861, 2024 vintage; downloaded to econ_cache_dir/eia861/, offline replays tests/fixtures/economics/eia861/); the BA’s annual demand is the EIA-930 daily-demand sum via fetch_ba_annual_load; the state retail denominator is shared with consumer-energy.yaml (#91).
• A restructured utility reports in a split — Bundled (SSO) + Delivery (shopping). Retail sales/customers are the sum, and the average price is the bundled (full-service) revenue/sales: delivery-only rows exclude generation, so a blended price would understate the all-in cost. This applies wherever the state’s retail market is restructured.
• The campus load share derives from the first-class facility_draw (watermark.facility.power, issue #87 — IT load × PUE), at the same ~0.9 load factor as #91.

ba-interchange.yaml specifics (issue #95)

• EIA-930 region-data is hourly; the connector reduces it to window-mean aggregates inside the fetch (demand mean/peak, net-generation mean, net-interchange mean/min/max, net-import-hours fraction) so the cached payload and committed fixture stay tiny. The window is a representative summer month (peak cooling demand, data-center relevant), not the full year — recorded in period_start/period_end.
• Sign convention: total interchange TI is + for net exports, − for net imports (EIA-930). One BA window is committed today — PJM, which is a net exporter on average and is the balancing authority every site registered so far resolves to. A site outside PJM pulls its own BA rather than reading this one.
• Keyed (WATERMARK_EIA_API_KEY); offline replays tests/fixtures/economics/eia930/. The committed fixture + figures are a genuine recorded keyed pull of the 2024-06 window (watermark interchange, #95/#120), superseding the earlier hand-authored aggregates (which had overstated net-import hours as ~24% vs the real ~0.6%).
• The derived CampusInterchangeComparison is a screening view over window means, not an hourly dispatch or locational (LMP/congestion) model — deliverability to the campus bus is the RTO interconnection-queue layer (#96).

Per-site instances

Which utility serves a campus, under which tariff, and what the load works out to against that utility’s book belong to the site being read. They live beside this file under instances/:

• Lima’s instance

Every other site’s answers are in its own reference/eia/<slug>/grid-profile.yaml, with each value carrying the source/confidence it was resolved at.

Cataloged datasets — generated from data/catalog/reference/; run watermark catalog render --apply after editing an entry.

A {…} placeholder in a title resolves to the site reading it — a slug-scoped dataset holds each site’s own copy, so its title names each site’s own county.

eia — Balancing-Authority Interchange Aggregates (EIA-930)

Source: U.S. EIA — Form EIA-930 Hourly Electric Grid Monitor (BA region demand/net-gen/interchange) · License: U.S. Government work (public domain) · Access: keyed · Site scope: basin-shared · Refresh: on-demand

Regenerate: watermark interchange

eia-consumer-energy — {state} Consumer Energy-Cost Baseline (EIA API v2)

Source: U.S. EIA — API v2 seriesid route (state residential electricity/gas price, total retail sales) · License: U.S. Government work (public domain) · Access: keyed · Site scope: slug-scoped · Refresh: annual

Regenerate: watermark eia

eia-demand-pressure — Facility Demand → Consumer Price-Pressure Sensitivity (EIA + facility load)

Source: Derived — facility draw (air permit, #87) sized against EIA state retail sales + residential price (#91) · License: U.S. Government work (public domain) · Access: keyed · Site scope: slug-scoped · Refresh: annual

Regenerate: watermark eia

eia-grid-profile — Per-Site Campus Grid Profile — Serving Utility, BA & Load Share (EIA-861/930)

Source: U.S. EIA — Form EIA-861 (serving-utility retail sales) + Form EIA-930 (balancing-authority hourly demand) · License: U.S. Government work (public domain) · Access: keyed · Site scope: slug-scoped · Refresh: annual

Regenerate: watermark grid