exascale.build

Describe valid filters, groupings, metrics, detail fields, and citation fields for ai_infrastructure.construction.

Describe valid filters, groupings, metrics, detail fields, and citation fields for ai_infrastructure.equipment_trade.

Describe valid filters, groupings, metrics, detail fields, and citation fields for ai_infrastructure.production.

Describe valid filters, groupings, metrics, detail fields, and citation fields for ai_infrastructure.trade.

Describe any served capability by name — the generic twin of the named describe tools. Pass `capability` as either a capability id from list_capabilities_v1 (e.g. "power.capacity") or a query primitive name (e.g. "query_power_capacity_v1"). Returns the same schema payload as the named describe tool: valid filters, groupings, metrics, detail fields, and citation fields. Use the generic pair (this + query_capability_v1) when list_capabilities_v1 names a capability that has no named tool in your client's tool list — clients cache tool lists, and capabilities shipped after that cache are still fully reachable here.

Describe capacity factor: net generation / (operating nameplate × hours), joined across EIA-860M and EIA-923.

Describe the EIA Henry Hub and state electric-power gas-price atoms, units, aggregate grain, missingness, vintages, and citations.

Describe annual EIA-860 ownership filters, convention, metrics, and exact-cell citations.

Describe MISO accreditation document kinds, resource classes, ratios, suppressions, and citations.

Describe PJM ELCC document identities, class dimensions, metrics, and citations.

Describe MISO PRA planning-year, season, zone, parameter, and citation fields.

Describe PJM RPM delivery-year/LDA filters, UCAP metrics, and citations.

Describe valid filters, groupings, metrics, detail fields, and citation fields for power.capacity.

Describe valid filters, groupings, metrics, detail fields, and citation fields for power.demand_rollup.

Describe valid filters, groupings, metrics, detail fields, and citation fields for power.demand.

Describe EIA-923 receipt/cost filters, raw units, missing semantics, and exact-cell citations.

Describe valid filters, groupings, atoms, metrics, detail fields, and citation fields for power.generation.

Describe valid filters, groupings, metrics, detail fields, and citation fields for power.interconnection_queue_caiso.

Describe valid filters, groupings, metrics, detail fields, and citation fields for power.interconnection_queue_ercot.

Describe valid filters, groupings, metrics, detail fields, and citation fields for power.interconnection_queue_isone.

Describe valid filters, groupings, metrics, detail fields, and citation fields for power.interconnection_queue_nyiso.

Describe valid filters, groupings, metrics, detail fields, and citation fields for power.interconnection_queue_pjm_cycle.

Describe valid filters, groupings, metrics, detail fields, and citation fields for power.interconnection_queue_pjm.

Describe valid filters, groupings, metrics, detail fields, and citation fields for power.interconnection_queue_spp.

Describe valid filters, groupings, metrics, detail fields, and citation fields for power.interconnection_queue.

Describe native-XBRL Form 1 plant-cost facts, units, coverage, and exact-fact citations.

Describe valid filters, groupings, metrics, detail fields, and citation fields for power.price_ercot.

Describe valid filters, groupings, metrics, detail fields, and citation fields for power.retail_sales.

Describe valid filters, groupings, metrics, detail fields, and citation fields for robotics.adoption.

Describe valid filters, groupings, metrics, detail fields, and citation fields for robotics.trade.

Describe valid filters, groupings, metrics, detail fields, and citation fields for space.satellite_filings.

Describe valid filters, groupings, metrics, detail fields, and citation fields for space.satellite_notices.

Fetch and hash-verify the raw source row behind a returned citation. Pass a citation object inside `params`. Two ready-to-pass shapes come straight from the query tools: each aggregate row's `citations[ref].verify` object, or a detail record's `citation` (from include_records). Either proves the number with no re-query. The tool verifies the raw workbook SHA-256 before returning source-header row values. Use this when an agent must prove an answer from the underlying source row.

List available exascale.build data capabilities for agent discovery before querying. Also call this BEFORE stating that a capability is not available — client tool lists are cached and this surface grows; anything listed here is reachable via query_capability_v1 even if your tool list predates it.

Query verified U.S. private construction spending ($ millions) for data centers and semiconductor/computer-electronics manufacturing plants, from the U.S. Census Bureau's Value of Construction Put in Place (C30). Use this for "how much is being spent BUILDING data centers (or chip fabs) in the US" questions — the construction buildout in dollars, not capacity or investment. Filter by `category` ("data_center" — Census's named subcategory under Office; or "computer_electronic_electrical" — the semiconductor/computer-electronics manufacturing line under Manufacturing), `basis` ("seasonally_adjusted" = a seasonally-adjusted ANNUAL RATE, or "not_seasonally_adjusted" = the NOT-adjusted MONTHLY LEVEL), `data_month` (one month, ISO first-of-month e.g. "2026-04-01") or the `data_month_from`/`data_month_to` range, `year`, and `revision_status` ("preliminary", "revised", or "final"). Group by any of `category`, `basis`, `data_month`, `year`, or `revision_status`. Pass each parameter as a top-level key of `params` (flat — not nested under a `filter`, `filters`, or `where` key). Example: `{"category": "data_center", "basis": "seasonally_adjusted", "data_month": "2026-04-01"}` for one month; add `"group_by": ["data_month"]` over a `data_month_from`/`data_month_to` range for a series. Returns JSON aggregates with citations and optional row-level records when `include_records` is true — every value cites the exact Census workbook, sheet, row, and column. The two categories are DISTINCT series and are never conflated: `data_center` is data-center buildings; `computer_electronic_electrical` is the chip/electronics-manufacturing (fab) line — the CHIPS-Act build-out. `basis` is the other fork: the seasonally-adjusted series is an ANNUAL RATE (what the current monthly pace annualizes to), while the not-seasonally-adjusted series is the actual MONTHLY LEVEL. `revision_status` carries Census's own preliminary/revised/final marking verbatim. Data is monthly; the data-center series begins 2014-01. The response `as_of` is the release vintage (Census revises monthly); pin `as_of` to an earlier vintage to reproduce what was served then. NOT additive: `construction_spending_musd` is a published per-(category, basis, month) reading, so a total that mixes the two bases (an annual rate + a monthly level), or that sums the seasonally-adjusted ANNUAL-RATE series across months, is not a real figure — such a result carries a `construction_aggregation` scope note and ranking remainders omit the metric. Filter to one `basis` and `group_by data_month` for a series over time. Does not determine total data-center INVESTMENT (servers, chips, cooling, equipment — this is construction put-in-place only; Census does not publish an investment total), data-center MW capacity, count, square footage, or location (use the power.* capabilities for capacity and the interconnection queue), which company or project is building (Census C30 has no operator breakdown), public or government construction (this is PRIVATE construction only), or construction outside these two categories.

Query verified U.S. monthly IMPORTS of semiconductor-manufacturing EQUIPMENT (HS-8486) — customs value (USD) by country of origin — from the U.S. Census Bureau's International Trade data. Use this for "is the fab buildout actually tooling up, and who supplies the machines" questions — the equipment leg of the fab lifecycle: construction spending (ai_infrastructure.construction) measures the shell, this measures the tools flowing in, and chip imports (ai_infrastructure.trade) measure the output side. HS-8486 covers machines and apparatus used solely or principally to MANUFACTURE semiconductor boules/wafers, devices, and integrated circuits — AND flat-panel displays (Census does not split them at this level); it is NOT the chips themselves (those are HS-8542). Filter by `country` (the verbatim Census name, e.g. "JAPAN", "NETHERLANDS", "KOREA, SOUTH"), `cty_code` (the Census country code), `country_level` ("total" = the all-countries TOTAL, "country" = an individual country, "grouping" = a Census bloc/continent like ASIA / APEC / EU), `year`, `data_month` (one month, ISO first-of-month e.g. "2026-04-01") or the `data_month_from`/`data_month_to` range. Group by any of `country`, `cty_code`, `country_level`, `data_month`, or `year`. Pass each parameter as a top-level key of `params` (flat — not nested under a `filter`, `filters`, or `where` key). Example: `{"country_level": "country", "group_by": ["country"], "order_by": "general_value_usd", "top_n": 5}` for the top tool-supplying countries; `{"country_level": "total", "group_by": ["data_month"]}` for the national trend. Returns JSON aggregates with citations and optional row-level records when `include_records` is true — every value cites the exact Census response row, re-verifiable via get_source_evidence_v1. Measures: `general_value_usd` (general imports value) and `consumption_value_usd` (imports for consumption) — value only; no tool counts, and no tool-type or vendor breakdown (one HS4 heading: no lithography-vs-deposition-vs-etch split, no per-manufacturer series such as ASML). NEVER SUM across country rows: Census's groupings (ASIA, APEC, EU, OECD, ASEAN, the continents) OVERLAP each other and the individual countries, and the all-countries TOTAL contains everything — so adding rows double-counts; a cross-row sum returns a country_aggregation note and nulls the metric in ranking remainders. Filter `country_level=total` for the U.S. national figure. Country is the country of ORIGIN (Census attribution), not which U.S. fab, state, or operator receives the equipment — there is no U.S. place breakdown. Imports only (not exports), customs value (not landed/CIF/duty), and recent months are preliminary and revised in later Census releases.

Query verified U.S. semiconductor & electronic-component PRODUCTION and CAPACITY UTILIZATION — the Federal Reserve's monthly G.17 industrial-production index (2017=100) and capacity-utilization rate (percent) for NAICS 3344 — from the Board's own release, history to 1972. Use this for "are the domestic fabs actually producing / how hot are they running" questions — the OUTPUT leg of the fab lifecycle: construction spending (ai_infrastructure.construction) measures the shell, equipment imports (ai_infrastructure.equipment_trade) the tools flowing in, chip imports (ai_infrastructure.trade) what crosses the border; this measures domestic production and how much of the installed capacity is in use. NAICS 3344 is "semiconductor and OTHER electronic component" manufacturing — the finest split the Fed publishes here (broader than semiconductors alone, and NOT the same slice as QCEW's 334413). Filter by `series_kind` ("ip" = the production index, on both bases; "capacity_utilization" = percent of capacity in use, seasonally adjusted only; "capacity" = the capacity index behind the rate), `series_name` (the verbatim Fed series, e.g. "IP.G3344.S", "CAPUTL.G3344.S"), `basis` ("seasonally_adjusted" / "not_seasonally_adjusted" — IP only), `year`, `data_month` (ISO first-of-month, e.g. "2026-05-01") or the `data_month_from`/`data_month_to` range. Group by any of `series_name`, `series_kind`, `basis`, `data_month`, or `year`. Pass each parameter as a top-level key of `params` (flat — not nested). Example: `{"series_kind": "capacity_utilization", "group_by": ["data_month"], "data_month_from": "2024-01-01"}` for the utilization trend; `{"series_kind": "ip", "basis": "seasonally_adjusted", "group_by": ["year"]}` for the production index by year (an average per year). Returns JSON aggregates with citations and optional row-level records when `include_records` is true — every value cites the exact Fed SDMX observation, re-verifiable via get_source_evidence_v1. Measures are avg/min/max per reading — `avg_ip_index`, `avg_capacity_utilization_pct`, `avg_capacity_index` (+ min/max variants): an index or a rate is INTENSIVE, so multi-month figures are AVERAGES, never sums (the Fed publishes its own quarterly/annual aggregations, which this block does not serve — monthly grain only). An index is not dollars and not unit counts (2017=100). Capacity and utilization exist seasonally adjusted only — their not-seasonally-adjusted cells are structurally absent, never zero. Averaging the IP index across both bases returns a production_aggregation note — filter or group by basis instead. National industry aggregate: no state, county, fab, or company breakdown. Every monthly release revises history (as_of carries the vintage).

Query verified U.S. monthly IMPORTS of integrated circuits (HS-8542) — customs value (USD) by country of origin — from the U.S. Census Bureau's International Trade data. Use this for "how much $ of chips did the US import (from Taiwan / South Korea / in total) and how is it trending" questions. HS-8542 is ALL integrated circuits (processors, memory, amplifiers, parts) — NOT AI-accelerator / GPU-specific. Filter by `country` (the verbatim Census name, e.g. "TAIWAN", "KOREA, SOUTH"), `cty_code` (the Census country code, e.g. "5830"), `country_level` ("total" = the all-countries TOTAL, "country" = an individual country, "grouping" = a Census bloc/continent like ASIA / APEC / EU), `year`, `data_month` (one month, ISO first-of-month e.g. "2026-04-01") or the `data_month_from`/`data_month_to` range. Group by any of `country`, `cty_code`, `country_level`, `data_month`, or `year`. Pass each parameter as a top-level key of `params` (flat — not nested under a `filter`, `filters`, or `where` key). Example: `{"country_level": "country", "group_by": ["country"], "order_by": "general_value_usd", "top_n": 5}` for the top source countries; `{"country_level": "total", "group_by": ["data_month"]}` for the national trend. Returns JSON aggregates with citations and optional row-level records when `include_records` is true — every value cites the exact Census response row, re-verifiable via get_source_evidence_v1. Measures: `general_value_usd` (general imports value) and `consumption_value_usd` (imports for consumption) — value only; HS-8542 reports no meaningful quantity at this level, so there is no chip count. NEVER SUM across country rows: Census's groupings (ASIA, APEC, EU, OECD, ASEAN, the continents) OVERLAP each other and the individual countries, and the all-countries TOTAL contains everything — so adding rows double-counts. Filter `country_level=total` for the U.S. national figure, `country_level=country` for individual countries, or group_by country for the per-country series; a cross-row sum returns a country_aggregation note and nulls the metric in ranking remainders. Country is the country of ORIGIN (Census attribution), not where a chip is installed — there is no U.S. state/county breakdown. Imports only (not exports), customs value (not landed/CIF/duty), and recent months are preliminary and revised in later releases.

Query any served capability by name — the generic twin of the named query tools, reaching every capability including ones newer than your client's cached tool list. Pass `capability` as either a capability id from list_capabilities_v1 (e.g. "power.price_ercot") or a query primitive name (e.g. "query_power_price_ercot_v1"), and `params` as the same flat JSON object the named query tool accepts — call describe_capability_v1 first for valid filters, e.g. {"capability": "power.capacity", "params": {"state": "TX", "group_by": ["energy_source_code"]}}. Returns the identical cited envelope as the named tool: same rows, same citations, same as_of.

Query verified U.S. capacity factor — how hard a fleet actually runs — by joining EIA-860M capacity and EIA-923 generation. Requires `data_month`: one ISO month start, e.g. "2026-01-01". If the user names no month, ask which one (or state the month you chose); if a month is not covered, the error lists the months that are — do not retry blindly. capacity_factor = net generation (MWh) / (operating nameplate capacity (MW) × hours in the month), computed over plant×fuel present in BOTH sources, so scope is auto-aligned. Optional `group_by` of `state` and/or `fuel_group`, and `state`/`fuel_group` filters. Returns the capacity factor per group with its generation and capacity, a `coverage` declaration (what share of in-scope capacity/generation matched), and a citation to BOTH the capacity and the generation source row. Basis is nameplate; storage is excluded; the capacity snapshot is matched to the month. Does not determine per-generator capacity factor, a net-summer/winter basis, or months absent from either source.

Query verified EIA natural-gas price records from two public series. `atom="henry_hub_daily_spot"` returns one trading-day Henry Hub spot price in EIA's published $/MMBtu unit (series RNGWHHD). `atom="state_electric_power_monthly"` returns one state-month average price paid by electric-power consumers in EIA's published $/Mcf unit; `{"atom":"state_electric_power_monthly","state":"TX"}` is the Texas power-sector proxy slice in one call. Filter either atom by `date`, `date_from`/`date_to`, series, state, or price status. The state series is an aggregate of what the state's power sector paid that month, not a plant-level fact. EIA null values remain explicit `price=null` records with `price_status="source_missing"`—never zero or imputed. The two source units are never converted or blended; mixed-unit avg/min/max are null. This tool does not assign proxies to plants or derive heat rates, $/MWh, spark spreads, or any ratio. API revisions are preserved as immutable capture vintages selected by `as_of`. Every price cites its exact archived response record, series id, period, value cell, and SHA-256 for source-evidence verification.

Query verified annual EIA-860 generator ownership. Returns the Owner schedule's raw owner names and ownership shares for each `eia_plant_id` + `generator_id`. `percent_owned` is the workbook's raw fraction of one (0.6 = 60%), not a whole-number percent. Filter by plant, generator, exact raw owner name, state, owner state, annual vintage, or source-reported balancing authority code; `{"state":"TX","balancing_authority_code":"ERCO"}` returns an ERCOT slice in one call. Owner-name matching is exact and intentionally performs no normalization or entity resolution. Critical EIA convention: the Owner schedule contains only jointly owned generators and generators wholly owned by an entity other than the operator. A generator absent from it is wholly owned by the operator in that same annual EIA-860 Generator schedule. An exact plant+generator query exposes this as `ownership_resolution`; it does not fabricate an Owner row or a 1.0 source share. Annual vintages remain independently queryable. Every returned share cites its exact ZIP member, sheet, row, and `Percent Owned` cell for SHA-256 verification.

Query cited MISO Schedule 53 class-average and indicative-DLOL seasonal accreditation ratios, including published unit counts, storage variants, and explicit suppression atoms. Requires an API key ($49/mo or $490/yr, 7-day full refund, at https://exascale.build/pro) while the source's redistribution terms are under review. This data point is not included in the 25 free queries.

Query cited PJM marginal ELCC class ratings by document kind, auction, delivery year, and resource class. Preliminary, final, and incremental-auction documents remain separate; no value is preferred or reconciled. Requires an API key ($49/mo or $490/yr, 7-day full refund, at https://exascale.build/pro) while the source's redistribution terms are under review. This data point is not included in the 25 free queries.

Query cited MISO Planning Resource Auction seasonal zonal ACP, verbatim ERZ values, published CONE parameters, and IMM conduct thresholds. Requires an API key ($49/mo or $490/yr, 7-day full refund, at https://exascale.build/pro) while the source's redistribution terms are under review. This data point is not included in the 25 free queries.

Query cited PJM Base Residual Auction UCAP clearing prices, RTO cleared MW, reliability requirement, and published administrative price collar by delivery year and LDA. Requires an API key ($49/mo or $490/yr, 7-day full refund, at https://exascale.build/pro) while the source's redistribution terms are under review. This data point is not included in the 25 free queries.

Query verified U.S. generator-level operating, planned, retired, or canceled power capacity from EIA-860M. Use this for capacity questions by state/jurisdiction, county FIPS, source-reported balancing authority code, fuel, prime mover, technology, lifecycle, or year. Pass filters inside the `params` object. The operating/planned/retired/canceled selector is `lifecycle` (e.g. `lifecycle: "operating"`, the default) — there is no `status` or `status_group` parameter. Returns JSON aggregates with citations and optional generator-level records when `include_records` is true. Does not determine electricity supplied, generation MWh, real-time dispatch, capacity factor, battery storage throughput/duration, demand/load, prices, data-center load, or transmission deliverability. For capacity REQUESTED in an ISO interconnection queue (projects pending interconnection, not yet built), use the relevant ISO's queue tool: query_power_interconnection_queue_v1 (MISO), query_power_interconnection_queue_pjm_v1 (PJM — or query_power_interconnection_queue_pjm_cycle_v1 for PJM's cluster/cycle grid), or query_power_interconnection_queue_caiso_v1 (CAISO).

Query verified U.S. hourly electricity demand (MW) as EIA's own published national and regional totals from the EIA Grid Monitor (region-data). Use this for "how much load for the whole country, or a region" questions. Filter by `respondent` (US48 = the Lower-48 national total, or one of the 13 EIA regions — CAL, CAR, CENT, FLA, MIDA, MIDW, NE, NW, NY, SE, SW, TEN, TEX), `data_date` (one day) or the `data_date_from`/`data_date_to` range, and `hour_number`. To pin one specific UTC hour, combine `data_date` + `hour_number`. Group by any of `respondent`, `respondent_level` (national vs region), `data_date`, `hour_number`, or `datetime_utc`. `datetime_utc` and `respondent_level` are grouping/output axes only — not filters. Pass each parameter as a top-level key of `params` (flat — not nested under a `filter`, `filters`, or `where` key). Example: `{"respondent": "US48", "data_date": "2026-06-10", "hour_number": 14}` for the US48 total at one hour; add `"group_by": ["datetime_utc"]` over a `data_date_from`/`data_date_to` range for a series. Returns JSON aggregates with citations and optional row-level records when `include_records` is true. `demand_mw` is EIA's OWN published demand total, served verbatim — the Adjusted series (the same canonical definition as power.demand's `demand_mw`), NOT a sum exascale computed. This closes power.demand's refusal of national/region totals (BA demand is non-additive across balancing authorities). `demand_forecast_mw` is the same respondent-hour's day-ahead forecast, so forecast-vs-actual misses need no second query. History runs hourly from 2019-01-01 onward — this published series begins about 3.5 years later than power.demand's balancing-authority history — and is served by default; the response `as_of` is the knowledge cut. A query with NO calendar window and no calendar-axis `group_by` defaults to the latest day with reported demand and says so in a `default_latest_day` note — group by `data_date` or `datetime_utc`, or pass a date range, for a series over time. Pin `as_of` to an earlier vintage to reproduce what was served then. INVERTED additivity: `demand_mw` is ALREADY a published total, so it is NOT additive across respondents — US48 already equals the sum of the 13 regions. A result spanning more than one respondent without grouping by it carries a `respondent_aggregation` scope note and ranking remainders omit the demand metrics: filter `respondent=US48` for the national total, or group by `respondent` for the per-respondent series. Occasional source-quality anomalies (an hour EIA did not publish; a rare impossible value EIA published) are served verbatim and cited, never altered. Does not determine balancing-authority-level demand (use power.demand for the BA series), demand before 2019-01-01, the raw un-Adjusted series (this route publishes the Adjusted series only), plant, generator, county, or state attribution, installed capacity (use power.capacity), monthly plant generation (use power.generation), retail sales, revenue, or customers (use power.retail_sales), wholesale prices, or long-horizon forecasts (the forecast is day-ahead only).

Query verified U.S. hourly electricity demand (MW) by balancing authority from EIA-930. Use this for "how much load" questions at the hourly balancing-authority grain: filter or group by `balancing_authority_code`, `region`, `data_date` (or the `data_date_from`/`data_date_to` range), `hour_number`, `datetime_utc`, or `is_imputed`. Pass filters inside the `params` object. Returns JSON aggregates with citations and optional row-level records when `include_records` is true. `demand_mw` is EIA's own cleaned (Adjusted) series, with receipts: the as-reported `demand_mw_raw` and the `is_imputed` flag ride every detail record. `demand_forecast_mw` is the same row's day-ahead forecast, so forecast-vs-actual misses need no second query. History runs hourly from 2015-07-01 onward and is served by default: a bare `data_date` anywhere in that window answers from the newest promoted vintage covering it, and the response `as_of` is that knowledge cut. A query with NO calendar window (no `data_date`, `data_date_from`, or `data_date_to`) and no calendar-axis `group_by` defaults to the latest day that has reported demand — not the full history — and says so in a `default_latest_day` note; group by `data_date` or `datetime_utc`, or pass a date range, to read a series over time. Pin `as_of` to an earlier vintage to reproduce exactly what was served then; one response may cite several source files, and every citation carries its own file and vintage. An empty result names the served coverage window in an `empty_scope` note. Demand is NOT additive across balancing authorities: a result summing more than one BA carries a `ba_aggregation` scope note and ranking remainders omit the demand metrics — group by `balancing_authority_code` for the source-grain series. Does not determine plant, generator, county, or state attribution (EIA-930 carries no such IDs, and BA footprints do not follow state lines), US48 or regional totals (computed rollups are refused; EIA's own published series is the named follow-up), installed capacity (MW — use power.capacity), monthly plant generation (use power.generation), retail sales/revenue/customers (use power.retail_sales), wholesale prices, or long-horizon forecasts (the EIA-930 forecast is day-ahead only).

Query verified raw EIA-923 fuel receipts and delivered fuel costs. Returns one Page 5 Fuel Receipts and Costs row per published receipt: plant/month, fuel, supplier, purchase type, source physical quantity, and delivered cost in EIA's stated cents/MMBtu. Filter by plant, month/range, exact source strings, state, fuel, cost status, or source-reported balancing authority code; `{"state":"TX","balancing_authority_code":"ERCO"}` returns an ERCOT slice in one call. Quantity units remain fuel-specific (short tons, barrels, or Mcf). EIA withholds costs for some plants. The raw `.` marker is preserved in `fuel_cost_raw`, the numeric cost is null, and `fuel_cost_status` explicitly reports `withheld` for unregulated receipts. Missing is never zero or imputed. This tool does not derive heat rates, efficiency, marginal cost, generation cost, or $/MWh; combine the cited raw atoms outside exascale.build if analysis requires those judgments. Every quantity or cost can be verified against its exact workbook cell.