WAPR

Worldwide AI Proliferation Registry
catalogue built 2026-11-02 03:29 UTC
build …latest run build f7813e9identity detector v4condition rules WCI-1
Data catalogue

Every instrument catalogued so far

Every instrument catalogued so far (37), whether or not a page shows it yet. 1 data folder(s) have no entry here yet; catalogue.json names them. Counts are measured from the files themselves each time this page is built.

Machine-readable: series/index.json (the tape's index) · catalogue.json (this catalogue) · csv-index.json (the index of CSV downloads of the instruments listed here)

The tape 16 series

Each series below shows its latest row on the tape and how many days the tape holds for it; its CSV holds every row. A row's time is when that row was appended to the tape, not when its instrument last checked its source — the catalogue below gives each instrument's check.

serieslatest readingfeed and datafile
On-site power for data centres, MW stated in a primary documentannounced_mw 13,139.25MWn 9 register rows · denominator 18 register rows feed row appended 2026-11-02 03:25 UTC data vintage register as last edited 2026-09-15T07:40:16Z (the board's own record) this period has no stated end date announced_mw.csv36 days on the tape14 restatements
Sites with statistically distinctive crane activity this yearcranes_distinctive_sites 350sites with distinctive crane activityn 3,189 filings · denominator 680 sites with any filing feed row appended 2026-11-02 03:25 UTC data vintage 2026 this period has not ended cranes_distinctive_sites.csv36 days on the tape1 restatement
Filed DMR values that broke a permit limit in the last 12 monthsdmr_exceedances_12mo 4filed values over their permit limit, 12 monthsn 16 data-centre-coded NPDES permits · denominator 81 data-centre-coded NPDES permits feed row appended 2026-11-02 03:25 UTC data vintage 2025-10-31..2026-10-31 (collected 2026-11-01) dmr_exceedances_12mo.csv35 days on the tapeno restatements
Water actually discharged as filed in monthly DMRs, sum of 12-month meansdmr_flow_mgd_12mo 1.247MGD, sum of 12-month mean reported flowsn 4 data-centre-coded NPDES permits · denominator 81 data-centre-coded NPDES permits feed row appended 2026-11-02 03:25 UTC data vintage 2025-10-31..2026-10-31 (collected 2026-11-01) dmr_flow_mgd_12mo.csv35 days on the tapeno restatements
Share of mapped facilities with a recorded operatorfacilities_attributed_pct 60% with operatorn 3,019 mapped facilities · denominator 5,041 mapped facilities feed row appended 2026-11-02 03:25 UTC data vintage 2026-11-01 facilities_attributed_pct.csv36 days on the tape7 restatements
Mapped facilities in the declared framefacilities_mapped 5,041mapped facilitiesn 5,041 (unit not stated) feed row appended 2026-11-02 03:25 UTC data vintage 2026-11-01 facilities_mapped.csv36 days on the tape14 restatements
LLM-tagged repo creation events on Hugging Face per collection window — a re-created id counts again (vintage gives the window and its hours)hf_llm_new 274LLM-tagged repo creation events in the collection windown 274 repo creation events of any pipeline tag in the collection window · denominator 1,888 repo creation events of any pipeline tag in the collection windowagainst the row before it: windows of different length (25 h, then 12 h) — the counts are not comparable feed row appended 2026-11-02 03:25 UTC data vintage 2026-11-01T15:15:03Z..2026-11-02T03:15:04Z (12.0 h) hf_llm_new.csv36 days on the tape7 restatements
Share of those repackaged as quantizedhf_quantized_pct 45.26% of LLM-tagged repo creation events in the collection windown 124 LLM-tagged repo creation events in the collection window · denominator 274 LLM-tagged repo creation events in the collection window feed row appended 2026-11-02 03:25 UTC data vintage 2026-11-01T15:15:03Z..2026-11-02T03:15:04Z (12.0 h) hf_quantized_pct.csv36 days on the tape7 restatements
Share of those flagged as safety-strippedhf_safety_stripped_pct 8.76% of LLM-tagged repo creation events in the collection windown 24 LLM-tagged repo creation events in the collection window · denominator 274 LLM-tagged repo creation events in the collection window feed row appended 2026-11-02 03:25 UTC data vintage 2026-11-01T15:15:03Z..2026-11-02T03:15:04Z (12.0 h) hf_safety_stripped_pct.csv36 days on the tape7 restatements
Ireland: data centres' share of metered electricity, by quarterireland_dc_share_pct 23.646% of metered electricityn 1,991 GWh · denominator 8,420 GWh feed row appended 2026-11-02 03:25 UTC data vintage 2025Q4 · ended 2025-12-31 ireland_dc_share_pct.csv36 days on the tape1 restatement
Solar share of the grids under operators' sites, 01:00–05:00 localnight_solar_pct 0.022% solar, night window, mean over operatorsn 6 operators on the grid-mix panel · denominator 6 operators on the grid-mix panel feed row appended 2026-11-02 03:25 UTC data vintage 2025-10-01..2026-09-30 night_solar_pct.csv36 days on the tape8 restatements
US discharge permits held by facilities reporting the data-centre sector codenpdes_permits_518210 45NPDES permits (NAICS 518210)n 45 (unit not stated) feed row appended 2026-11-02 03:25 UTC data vintage as held by ECHO at 2026-11-02T03:23:50Z this period has no stated end date npdes_permits_518210.csv35 days on the tape13 restatements
Average flow recorded on those permits, million gallons per daynpdes_reported_flow_mgd 0MGDn 0 permits · denominator 45 permits feed row appended 2026-11-02 03:25 UTC data vintage as held by ECHO at 2026-11-02T03:23:50Z this period has no stated end date npdes_reported_flow_mgd.csv35 days on the tape7 restatements
Announced on-site power a primary document says has NO utility connection — announced, never runningonsite_islanded_mw 7,525.85MW announced with no utility connectionn 3 on-site power rows the board counted · denominator 9 on-site power rows the board counted feed row appended 2026-11-02 03:25 UTC data vintage register as last edited 2026-09-15T07:40:16Z (the board's own record) this period has no stated end date onsite_islanded_mw.csv35 days on the tape8 restatements
On-site power with a primary OPERATING record, MW — BLIND until one existsoperating_mw BLIND — no value on the tapedenominator 18 register rowsMW with a primary OPERATING record (air-permit operating statement, CAMPD unit, EIA-923 generation); announcements and permit ceilings never count feed row appended 2026-11-02 03:25 UTC data vintage register as last edited 2026-09-15T07:40:16Z (the board's own record) this period has no stated end date operating_mw.csv36 days on the tape14 restatements
Virginia data-centre generator NOx per year, in the unit DEQ's emissions table states (each row's unit column), by vintageva_nox_t BLIND — no value on the tapeBLIND — no input this run: board panel va.nox_sum_t feed row appended 2026-11-02 03:25 UTC va_nox_t.csv36 days on the tape8 restatements

16 readings, each as a daily file. One row per UTC day a board was built — including days when nothing moved and days when the instrument was blind — so a gap is a gap and a flat line is a flat line. Value, unit, count, denominator, vintage and lamp travel together; a number never leaves this site without them. A past row the rules later derive differently is never overwritten silently: it is appended to the restatements file with the row that replaced it and why. A past row that a collector's own replay of the same collection later contradicted stays on the tape as published — it is what the board served that day — and is listed, per series, under superseded_by_source_ledger with the collector ledger line that supersedes it, the value published and the value the replay derives.

Restatements — restatements.jsonl, 116 entries across all series.

What each CSV column holds
date
UTC day of wapr_run: the last board built that day
value
the reading; empty when BLIND
unit
unit of value
n
how many items the value was computed from — or, where the value is a share of a quantity (ireland_dc_share_pct), the numerator itself in denominator_unit
denominator
for a share, what it is a share of; for a sum or count, the population (in denominator_unit) that the summed or counted items are drawn from — cranes_distinctive_sites counts sites, drawn from the sites with any filing, while its n counts filings; empty when neither applies
denominator_unit
not described in the index
vintage
the period or source state the reading DESCRIBES (a 2024 emissions year, a 2025Q4 quarter, a collection window, the register state the board recorded), which is not the day it was collected
lamp
live / degraded / blind
basis
how the number was derived, in one sentence
source
the public record it was read from
wapr_run
the board build (generated, UTC) the row was read from
How this is known

Before a scheduled run deploys, build/series.py --check rebuilds the tape from the repository's own history and compares every row before the current board's day with the appended copy; a mismatch stops that deploy. The latest row is checked on the next run.

restatements.jsonl: each entry names when the restatement was made (restated_on, UTC), board_run — the board-data.json generated stamp in the working tree when the restatement was written (the board it was checked against; not necessarily a deployed board) — the series, the day, the row it superseded, the row that replaced it and the reason.

Instruments 37

An instrument listed as collected-not-yet-displayed has no served payload yet; what it holds on disk is still measured and shown. We list it rather than wait, because what an instrument holds and what a page happens to render are different facts and conflating them is how registries start overstating themselves.

Not yet in this catalogue — data folders: water_causes.

on a page 23

Mapped facilities

Every building volunteer mapmakers have tagged as a data centre, in the parts of the world we have extracted so far.

holdings
5,041 · facilities · counted from the file generated 2026-11-01
grain
weekly (tools/refresh_weekly.sh applies Geofabrik diffs and re-filters every Sunday; refused when the extracts are unchanged, put back when its joins cannot follow)
first run logged
2026-09-14 · the earliest day in its run manifest
last run
2026-11-01T09:17:05Z · ok
feed checked 2026-11-01 09:17 UTC
last success
2026-11-01T09:17:05Z
re-derived
2026-10-28T07:05:12Z — re-derived offline from files already held (not a collection)
source
OpenStreetMap via Geofabrik extracts -> osmium tags-filter; 9 extract(s): us, canada, mexico, europe, asia, africa, south-america, australia-oceania, central-america
licence
ODbL — attribution + share-alike; keep OSM geometry separate from proprietary estimates

open the raw data →

How this is known
last success
the latest of its own manifest rows with state 'ok' (an extract that published the register); last_run and last_state are its latest attempt, whatever its state. Rows before 2026-09-17 recorded no extract_sha256, and a rerun over the same extract files also wrote 'ok', so the latest such row dates the last run that published the register, not provably the last read of changed extracts; since then 'ok' needs changed digests of the derived files; a new osmium pass over an unchanged extract gives the same bytes (28 Oct 2026: a re-run over the same extracts was refused as byte-identical), so a quiet week is refused and does not reset it

Interconnection facilities

Colocation and interconnection sites that network operators register publicly. Enrichment only — it over-represents colo and under-represents hyperscale.

holdings
4,524 · PeeringDB pagination stopped early (status=429 after 5000 rows); 4,524 geolocated rows kept · counted from the file generated 2026-09-14
grain
event
first run logged
2026-09-14 · the earliest day in its run manifest · declared 2026-09-13 in instruments.yaml
last run
2026-09-14T05:15:23Z · partial (truncated) — not a complete collection: PeeringDB pagination stopped early (status=429 after 5000 rows); 4,524 geolocated rows kept
feed checked 2026-09-14 05:15 UTC
last success
no successful run identified
source
PeeringDB /api/fac — research use under AUP; not redistributed in bulk
licence
PeeringDB AUP — research only, permission email pending; no bulk redistribution

open the raw data →

How this is known
last success
the latest facilities.py row whose PeeringDB state is 'ok' (a complete pagination); a row cut short by the API (status 429) is partial and does not count

Virginia regulated sites

Sites the Virginia environmental regulator has flagged as data centres, with what their backup generators actually emitted.

holdings
199 · sites · counted from the file generated 2026-11-01
grain
day (sites) / year (292 — upstream overwrites itself; WAPR snapshot is the only history)
first run logged
2026-09-14 · the earliest day in its run manifest
last run
2026-11-02T03:15:13Z · fetch_failed
feed checked 2026-11-01 02:15 UTC
last success
2026-11-01T02:15:10Z
source
VA DEQ ArcGIS EDMA layers 294 + 292
licence
public agency data

open the raw data →

How this is known
last success
the latest row with state 'ok' (a DEQ layer read in full)

Power plants ≥250 MW

Large power stations, so a neighbouring plant is never mistaken for the data centre next to it.

holdings
4,784 · plants · counted from the file generated 2026-09-23
grain
event (static WRI v1.3 release, one-time ingest; its recency is measured from the file, see plants.json vintage)
first run logged
2026-09-14 · the earliest day in its run manifest
last run
2026-09-23T07:58:52Z · cached
checked 2026-09-14 10:58 UTC data v1.3; capacity figures to 2019
last success
2026-09-14T10:58:42Z
snapshot
a one-time snapshot, not a feed — vintage v1.3; capacity figures to 2019
source
WRI Global Power Plant Database v1.3 (CC BY 4.0), a static release. Its capacity figures run to 2019 and the latest commissioning year it records is 2020; of the 4,784 plants of at least 250 MW kept here, 22 record commissioning in 2019, 1 in 2020 and 1,456 record no year, so a plant absent here may simply be newer than the database. One value is corrected by WAPR: ESCOMBRERAS GRUPO 1 (WRI1006467) is filed by WRI at latitude 7.5732 and served here as 37.5732. WRI's own value is kept on each corrected row (la_as_filed_by_wri / lo_as_filed_by_wri, with corrected_by_wapr naming the field), and the evidence for each — inside WRI's own file — is in this payload's `corrections` block.
licence
CC BY 4.0 — attribute WRI

open the raw data →

How this is known
last success
the latest row with state 'fetched' (the database downloaded); 'cached' re-reads a file already held
snapshot vintage
plants.json vintage.release and vintage.max_year_of_capacity_data

Basin water stress

How scarce fresh water already is in the basin under each mapped facility. Exposure, not usage.

holdings
5,041 · facilities · counted from the file generated 2026-11-01
grain
event (static 1979-2019 baseline; re-run per facility-frame vintage)
first run logged
2026-09-14 · the earliest day in its run manifest
last run
2026-11-01T09:17:33Z · ok
checked 2026-11-01 09:17 UTC data 1979–2019 baseline
last success
2026-11-01T09:17:33Z
re-derived
2026-09-23T04:55:53Z — re-derived offline from files already held (not a collection)
snapshot
a one-time snapshot, not a feed — vintage 1979–2019 baseline
source
WRI Aqueduct 4.0 baseline water stress (bws), HydroBASINS L6, 1979-2019 baseline — CC BY 4.0, © World Resources Institute. Exposure context, never impact.
licence
CC BY 4.0 — cite WRI Aqueduct 4.0

open the raw data →

How this is known
last success
the latest row with state 'ok' (the basin join completed)
snapshot vintage
water-stress.json source: the period the basin figures describe; the layer's release date is not stated in the payload

Land surface temperature

Satellite land-temperature maps, every 8 days this year and every August back to 2000.

holdings
128 · texture images
grain
8-day and month
first run logged
2026-09-14 · the earliest day in its run manifest
last run
2026-11-02T03:21:48Z · ok
feed checked 2026-11-02 03:21 UTC
last success
2026-11-02T03:21:48Z
re-derived
2026-09-21T19:29:27Z — re-derived offline from files already held (not a collection)
source
NASA GIBS · MODIS Terra L3 Land Surface Temperature (MOD11), 1 km — 8-day and monthly composites. Regional context only; 1 km cannot resolve a campus. No-retrieval (cloud, water, polar night) renders as grey earth.
licence
NASA GIBS open

open the raw data →

How this is known
last success
the latest closing row with state 'ok'

Overnight grid demand

Hour-by-hour electricity demand for the PJM grid and 20 of its sub-regions (Dominion's zone among them), reduced to the dead-of-night window where a data centre's load shows up clearest.

holdings
14,033 · distinct respondent-days (14,033 rows in 1 file(s))
grain
hour -> daily overnight window (fixed 01:00-05:00 LOCAL Eastern, DST-verified)
first run logged
2026-09-14 · the earliest day in its run manifest
last run
2026-11-02T03:15:53Z · ok
feed checked 2026-11-02 03:15 UTC
last success
2026-11-02T03:15:53Z
source
EIA API v2 rto/region-data (PJM parent total, type D) + region-sub-ba-data (the sub-regions reporting under parent PJM, DOM among them), hourly demand
licence
EIA open (key in .env)
How this is known
last success
the latest closing 'derive' row with state 'ok' (a run that fetched and reduced every window)

Texas interconnection queue

How much generating capacity is queued to connect in Texas against how much has signed a binding agreement.

holdings
2 · queue vintages
grain
month (new vintage only on new DocID)
first run logged
2026-09-14 · the earliest day in its run manifest
last run
2026-11-02T03:18:57Z · no_new_report
feed checked 2026-11-02 03:18 UTC
last success
2026-11-02T03:18:57Z
source
ERCOT MIS reportTypeId 15933 = GIS Report (GENERATOR interconnection queue — NOT large-load; research misattributed it, see TRACKER 2026-09-14)
licence
ERCOT public MIS
How this is known
last success
the latest row that settled the run's vintage: step 'vintage' with 'no_new_report' (the newest report is already held) or step 'derive' with 'ok' (a new vintage downloaded and derived); the 'doclist' request alone does not count

Plant emissions, daily

What neighbouring power plants actually emitted and generated, day by day, from the regulator's own monitors on the chimneys themselves — for the large plants near a mapped facility that WAPR matched to a monitoring record. The payload states the coverage it reached.

holdings
221,610 · rows in 223 file(s)
grain
day (published-quarter lag; endDate clamped to ceiling)
first run logged
2026-09-14 · the earliest day in its run manifest
last run
2026-11-02T03:20:59Z · run_summary
feed checked 2026-11-02 03:20 UTC
last success
no successful run identified
source
EPA CAMPD apportioned daily emissions (x-api-key; concurrency 1)
licence
EPA open

open the raw data →

How this is known
last success
not stated: no rule here says which of campd.py's manifest states is a completed collection (the manifest is a pre-registered holdout this catalogue does not read for content)

Crane filings

Every tall-crane notice filed with the aviation authority near a mapped facility — construction shows up here months before a building does.

holdings
18,379 · distinct case filings (18,379 rows in 2 file(s))
grain
day (case events)
first run logged
2026-09-14 · the earliest day in its run manifest
last run
2026-11-02T03:15:16Z · gate_failed
feed checked 2026-11-02 03:15 UTC
last success
2026-11-02T03:15:16Z
re-derived
2026-09-21T19:28:29Z — re-derived offline from files already held (not a collection)
source
FAA Obstruction Evaluation / Airport Airspace Analysis — Part 77 case archives, all nine regions
licence
public FAA data

open the raw data →

How this is known
last success
the latest row with state 'derived' written by a run that fetched (offline re-derivations are excluded)

Model publication

Every AI model published to the main public library, including ones with their safety training stripped out.

holdings
216,911 · distinct model creation events (repo id + createdAt) (216,911 rows in 37 file(s)); 196,224 distinct repo ids — a repo deleted and re-created keeps its id and is a second event
grain
day
first run logged
2026-09-14 · the earliest day in its run manifest
last run
2026-11-02T03:15:09Z · ok
feed checked 2026-11-02 03:15 UTC
last success
2026-11-02T03:15:09Z
source
Hugging Face /api/models createdAt cursor
licence
HF API public
How this is known
last success
the latest closing row with state 'ok'

Crawler identity

The keys AI companies publish so a website can tell their crawlers from impostors, archived daily because nobody else keeps them.

holdings
10 · endpoint documents
grain
event (12h probe cadence)
first run logged
2026-09-14 · the earliest day in its run manifest · declared 2026-09-13 in instruments.yaml
last run
2026-11-02T03:15:05Z · same
feed checked 2026-11-02 03:15 UTC
last success
2026-11-02T03:15:05Z
source
10 agent/crawler identity endpoints (OpenAI WBA JWKS, gptbot/searchbot/chatgpt-user, Anthropic, Google x3, Perplexity x2)
licence
public documents
How this is known
last success
the latest row in which an endpoint document was fetched and compared ('same', 'new', 'changed' or 'baseline'); there is no closing row, so a run in which only some endpoints answered still counts

Ireland's data-centre electricity

Ireland is the only country that officially counts data-centre electricity as its own category. Everywhere else the sector vanishes into general business use.

holdings
132 · distinct observations (132 rows in 1 file(s))
first run logged
2026-09-14 · the earliest day in its run manifest
last run
2026-11-02T03:21:31Z · derived
feed checked 2026-11-02 03:21 UTC
last success
2026-11-02T03:21:31Z
source
Central Statistics Office, Ireland — Data Centres Metered Electricity Consumption
licence
no licence declared by the publisher's API; copyright holder Central Statistics Office, Ireland — unverified, see the payload's source.licence note

open the raw data →

How this is known
last success
the latest closing row with state 'derived'

Live anomalies

Unusual conditions right now near the mapped sites — storms, drought, space weather. Things happening TO the sites, not caused by them.

holdings
18 · anomalies at the last run · counted from the file generated 2026-11-02 03:21 UTC
grain
event (conditions active at each run)
first run logged
2026-09-14 · the earliest day in its run manifest
last run
2026-11-02T03:21:30Z · ok
feed checked 2026-11-02 03:21 UTC
last success
2026-11-02T03:21:30Z
source
NOAA/NWS active alerts (api.weather.gov); US Drought Monitor weekly GeoJSON (NDMC/UNL, USDA, NOAA); NOAA SWPC
licence
NWS and SWPC — NOAA open; US Drought Monitor — terms not recorded by the collector

open the raw data →

How this is known
last success
the latest closing row (it carries the anomaly count) with state 'ok'

Permitted water discharge

Every Clean Water Act permit or stormwater coverage held by a US facility reporting the data-centre sector code — where, under what permit number, and at what flow where one is recorded.

holdings
45 · permits · counted from the file generated 2026-11-02 03:23 UTC
grain
event (permit register; re-swept per run)
first run logged
2026-09-15 · the earliest day in its run manifest · declared 2026-09-14 in instruments.yaml
last run
2026-11-02T03:23:50Z · derived
feed checked 2026-11-02 03:23 UTC
last success
2026-11-02T03:23:50Z
re-derived
2026-09-21T20:44:59Z — re-derived offline from files already held (not a collection)
source
US EPA Enforcement and Compliance History Online — Clean Water Act permitted facilities reporting sector code 518210 (data processing and hosting)
licence
EPA public domain

open the raw data →

How this is known
last success
the latest closing row with state 'derived' written by a run that queried ECHO (discharge --from-raw rows are offline and excluded)

Rivers near facilities

Flow and water temperature at public stream gauges near mapped sites — exposure, never a claim that a site warms a river.

holdings
998 · gauges · counted from the file generated 2026-11-02 03:21 UTC
grain
latest value each gauge has reported, re-read every run — 77 of 998 dated gauges last reported more than a year before collection (oldest 1984-12-31)
first run logged
2026-09-14 · the earliest day in its run manifest
last run
2026-11-02T03:21:49Z · derived
feed checked 2026-11-02 03:21 UTC
last success
2026-11-02T03:21:49Z
source
US Geological Survey National Water Information System — instantaneous values from instrumented stream gauges
licence
USGS public domain

open the raw data →

How this is known
last success
the latest closing row with state 'derived'

The heat effect, to scale

The most-requested claim about data centres, drawn at true size against the instruments that would have to detect it — a published negative result.

holdings
10 · distance rings · counted from the file generated 2026-11-02 03:25 UTC
grain
static (a physical bound, not a measurement)
first collection
2026-09-14 · as declared in instruments.yaml
last run
none recorded
feed checked: time not stated
last success
no successful run identified
source
WAPR environmental research synthesis (datacenter-environmental-layer): ground-level release bound, 0.079 K at 4.5 km and 0.016 K at 10 km for a 130 MW campus; per-campus systematic offset that does not shrink with sample size. Water: WAPR water-districts research (the ICPRB 2025 study's rate for data centres in Loudoun Water's service area, per effective MW; literature WUE scenario). Computed here, not asserted.
licence
WAPR derivation over public research

open the raw data →

How this is known
last success
not a collection: built from other instruments' payloads, with no run of its own

Grid fuel mix by operator

What kinds of power station are physically running on the grid where each company's data centres sit — not what the company says it buys.

holdings
26 · operators · counted from the file generated 2026-11-02 03:20 UTC
grain
hour, aggregated over the most recent complete 12 calendar months; rebuilt per run
first run logged
2026-09-14 · the earliest day in its run manifest
last run
2026-11-02T03:20:59Z · ok
feed checked 2026-11-02 03:20 UTC
last success
2026-11-02T03:20:59Z
source
EIA-930 Hourly Electric Grid Monitor, hourly net generation by balancing authority and energy source
licence
EIA data — US Government work, public domain; the balancing-authority polygons (Esri) are not republished, only the derived facility-to-code assignment

open the raw data →

How this is known
last success
the latest closing 'derive' row with state 'ok'

Water quality near facilities (US)

Samples that state agencies and the US Geological Survey filed from monitoring stations within 10 km of a mapped US facility, a reading marked unusual only against that station's own past in the same season — the condition of the water nearby, never a measure of what a building did.

holdings
9,002 · stations · counted from the file generated 2026-11-01 09:21 UTC
grain
sample (every sample since 2019 for a fixed characteristic set, per station)
first run logged
2026-09-15 · the earliest day in its run manifest
last run
2026-11-01T11:16:03Z · derived
feed checked 2026-11-01 11:16 UTC
last success
2026-11-01T11:16:03Z
re-derived
2026-09-26T10:36:15Z — re-derived offline from files already held (not a collection)
source
Water Quality Portal (waterqualitydata.us) — the joint EPA / USGS / NWQMC store of ambient water-quality samples filed by state agencies and USGS (surface-water samples, field measurements, and field replicates and duplicates of them; blanks, spikes and other quality-control checks, rejected results, other media and non-level statistics are excluded and counted)
licence
not recorded

open the raw data →

How this is known
last success
the latest closing row with state 'derived' written by a run that queried the Water Quality Portal (--cache-day rows are offline and excluded)

Prime-power register

Power plants built for data-centre load, one row per project, each with its evidence and how it was verified. The board counts a project only where a permit, a regulator's docket, a securities filing or a regulated exchange release carries the figure.

holdings
18 · projects · counted from the file generated 2026-09-14
grain
event (hand-verified register edits)
first collection
2026-09-14 · as declared in instruments.yaml
last run
none recorded
feed checked: time not stated
last success
no successful run identified
source
hand-kept register of generation projects built for data-centre load, each row carrying its evidence and a verification level; documents archived under data/prime/primary
licence
facts from public documents; each row carries its evidence
How this is known
last success
not stated: no manifest records its runs (a hand-kept register, or read live and never archived)

Coral reef heat stress

Ocean heat stress at monitored coral reefs, read from NOAA Coral Reef Watch through WAPR's proxy when the Explorer asks, cached for up to an hour at the edge and up to an hour in the browser — what the ocean is doing, never what a data centre did to it. The readings are not archived; the holdings are the station list.

holdings
20 · reef stations, read live
grain
day
archive
shown live, not archived · first collection declared 2026-09-12 in instruments.yaml
last run
none recorded
feed checked: time not stated
last success
no successful run identified
source
NOAA Coral Reef Watch, via /api/crw/<station>
licence
NOAA open

open the raw data →

How this is known
last success
not stated: no manifest records its runs (a hand-kept register, or read live and never archived)

News strip

Headlines from public feeds that match data-centre and power keywords, shown on the board with publisher, tier, date and link and never rewritten. Nothing is archived: each feed is cached at the edge for up to 30 minutes and the assembled strip for up to 15 minutes in the browser.

holdings
none held
grain
event (headlines as the feeds carry them; nothing archived)
archive
shown live, not archived · first collection declared 2026-09-14 in instruments.yaml
last run
none recorded
feed checked: time not stated
last success
no successful run identified
source
9 feeds (SEC, EIA, Microsoft, Google Cloud, Meta, DCD, Utility Dive, Google News x2)
licence
RSS titles verbatim with publisher+tier+link
How this is known
last success
not stated: no manifest records its runs (a hand-kept register, or read live and never archived)

Earthquakes

Earthquakes of magnitude 4.5 and above from the past seven days, fetched from the USGS feed by the Explorer in the visitor's browser — exposure context only. Nothing is archived.

holdings
none held
grain
event
archive
shown live, not archived · first collection declared 2026-09-13 in instruments.yaml
last run
none recorded
feed checked: time not stated
last success
no successful run identified
source
USGS M4.5+ 7-day GeoJSON (client-side, Explorer layer)
licence
USGS open
How this is known
last success
not stated: no manifest records its runs (a hand-kept register, or read live and never archived)

served as data 14

Weather control

How hot or cold each month actually was, so nobody can mistake a warm summer for a new data centre.

holdings
14,284 · distinct station-days (14,594 rows in 1 file(s))
grain
day (2019 backfill; ~3-day lag; revisions APPEND, last row per (station,date) wins)
first run logged
2026-09-14 · the earliest day in its run manifest
last run
2026-11-02T03:15:42Z · derived
feed checked 2026-11-02 03:15 UTC
last success
2026-11-02T03:15:42Z
source
NOAA NCEI Access Data Service (keyless primary; CDO token fallback) — GHCND daily TMAX/TMIN, 5 airport stations (Dulles, Richmond, DFW, Abilene, Honolulu)
licence
NOAA open
How this is known
last success
the latest closing row with state 'derived'

Model market changes

Models appearing, retiring, changing price or going down — changes only, never the firehose.

holdings
914 · change events in 37 file(s)
grain
event (12h diff, deltas only)
first run logged
2026-09-14 · the earliest day in its run manifest
last run
2026-11-02T03:15:11Z · same
feed checked 2026-11-02 03:15 UTC
last success
2026-11-02T03:15:11Z
source
OpenRouter catalogue, HF trending, OpenAI/Anthropic/GCP status, Epoch notable models CSV, BenchLM RSS
licence
public APIs; Epoch CC-BY
How this is known
last success
the latest row in which a watched page was fetched and compared; there is no closing row, so a run in which only some pages answered still counts

Generator maker filings

What the companies building power generators tell their investors, from binding filings rather than press releases.

holdings
6 · distinct filings (196 rows in 37 file(s))
grain
quarter (event)
first run logged
2026-09-14 · the earliest day in its run manifest · declared 2026-09-13 in instruments.yaml
last run
2026-11-02T03:15:10Z · cached
feed checked 2026-11-02 03:15 UTC
last success
2026-11-02T03:15:10Z
source
data.sec.gov latest 10-Q/10-K for CAT CMI GEV BE
licence
public filings
How this is known
last success
the latest row with state 'fetched' or 'cached': each is written after a live read of the company's EDGAR submissions index, 'cached' when its newest 10-Q/10-K was already held; index_failed, CIK_MISMATCH and doc_failed rows do not count

Discharge monitoring reports

What data-centre permittees reported discharging, month by month: flow, temperature, the intake-to-discharge temperature difference, and every filed value that broke its permit limit — the permittee's own measurement, not a permission.

holdings
81 · permits · counted from the file generated 2026-11-01 09:21 UTC
first run logged
2026-09-15 · the earliest day in its run manifest
last run
2026-11-01T09:21:15Z · ok
feed checked 2026-11-01 09:21 UTC
last success
2026-11-01T09:21:15Z
source
EPA ECHO — Discharge Monitoring Reports (ICIS-NPDES) via eff_rest_services.get_effluent_chart; census via dmr_rest_services.get_all_dmr_facilities
licence
EPA Standard Open Data License — public domain, attribution requested

open the raw data →

How this is known
last success
the latest closing row (it carries the run's counts) with state 'ok'

Siting context (US)

For the 1,850 mapped US facilities: the utility territory each sits in, once polygons the layer types FEDERAL (power marketers and irrigation projects drawn over the utilities they supply) are set aside as overlays (1,027 with a single remaining territory, 810 where several remaining territories overlap and none is chosen, 13 with no territory but a federal one or none at all), the nearest transmission line of 230 kV or more (found for 1,850), the principal aquifer beneath it (1,372 have one mapped) and the nearest named stream (found for 1,850) — utility and aquifer by containment, line and stream by distance.

holdings
1,850 · US sites · counted from the file generated 2026-11-02 03:29 UTC
first run logged
2026-09-15 · the earliest day in its run manifest
last run
2026-11-02T03:29:47Z · ok
feed checked 2026-11-02 03:29 UTC
last success
2026-11-02T03:29:47Z
source
aquifer: USGS Principal Aquifers of the United States (1:2,500,000); hv_line: HIFLD Electric Power Transmission Lines (ORNL / DOE-CESER); utility: HIFLD Electric Retail Service Territories, 2023-08 vintage (ORNL / DOE-CESER; per-polygon SOURCE = EIA-860/861 + the utility's filed map); waterway: USGS National Hydrography Dataset, Flowline — Small Scale (layer 4), named features only
licence
aquifer: US Government work, public domain; hv_line: Public use (HIFLD Open); utility: None (Public Use) per the item's licenceInfo; waterway: US Government work, public domain

open the raw data →

How this is known
last success
the latest closing row (it carries the site count) with state 'ok'; a run that answered every site from its cache still counts, and the dossier's section_sources.siting.collected dates the answers themselves

Air-permitted data centres, national

The air-permit sources the states report to the federal compliance system under the data-processing industry codes (NAICS 518210 or SIC 7374) — 495 at the last collection — with each state's permit key, its major/minor class, inspections and violations; 3 of them carry a CAMD program id, the link to EPA's clean-air-markets data.

holdings
495 · permit sources · counted from the file generated 2026-11-01 09:18 UTC
first run logged
2026-09-15 · the earliest day in its run manifest
last run
2026-11-01T09:18:51Z · ok
feed checked 2026-11-01 09:18 UTC
last success
2026-11-01T09:18:51Z
source
EPA ECHO — ICIS-Air facility search (air_rest_services), NAICS 518210 ∪ SIC 7374
licence
EPA disclaimer: freely distributed for non-commercial, scientific and educational use; federal facts

open the raw data →

How this is known
last success
the latest closing row (it carries the run's counts) with state 'ok'

Facility emissions, national (NEI 2023)

The tons of NOx, CO, particulates and VOCs each listed data-centre facility was recorded emitting in the 2023 National Emissions Inventory — the only national per-facility emissions number for the sector, and only as complete as what each state submitted and WAPR's data-centre filter, which nei.json counts state by state.

holdings
277 · facilities · counted from the file generated 2026-11-02 03:25 UTC
first run logged
2026-09-15 · the earliest day in its run manifest
last run
2026-11-02T03:25:16Z · ok
feed checked 2026-11-02 03:25 UTC
last success
2026-11-02T03:25:16Z
source
EPA National Emissions Inventory 2023 — point-source facility summary (EIS report 38234, 21 Jul 2026)
licence
EPA disclaimer — freely distributed for non-commercial, scientific and educational use; federal facts

open the raw data →

How this is known
last success
the latest closing row with state 'ok': each run checks EPA for a newer inventory, and 'fetched' on that row says whether the file was downloaded again

Generators at sites (EIA-860M/923)

Grid-connected generators of 1 MW or more on or beside a mapped data centre: the month each was switched on and its monthly output since — the meter for the campus plants that interconnect, and a named list of what EIA lists instead in the counties of the prime register's on-site power projects.

holdings
160 · plants · counted from the file generated 2026-11-02 03:29 UTC
first run logged
2026-09-15 · the earliest day in its run manifest
last run
2026-11-02T03:29:45Z · ok
feed checked 2026-11-02 03:29 UTC
last success
2026-11-02T03:29:45Z
re-derived
2026-09-21T20:45:13Z — re-derived offline from files already held (not a collection)
source
860m: https://api.eia.gov/v2/electricity/operating-generator-capacity (Form EIA-860M, monthly generator inventory); 923: https://api.eia.gov/v2/electricity/facility-fuel (Form EIA-923, monthly net generation and fuel)
licence
US Government work — public domain

open the raw data →

How this is known
last success
the latest closing row with state 'ok' written by a run that queried EIA (--from-archive rows are offline and excluded)

Stable identifiers

One WAPR id per mapped facility, minted once under a written rule and never changed when the map is edited; the crosswalk to the network registry, air permits, discharge permits, crane notices and on-site power, each attached by distance with the metres recorded.

holdings
5,041 · ids · counted from the file generated 2026-11-02
grain
event (minted once per facility; crosswalk rebuilt per run)
first collection
2026-09-15 · as declared in instruments.yaml
last run
none recorded
feed checked: time not stated
last success
no successful run identified
source
WAPR identifiers minted over the OpenStreetMap facility register, crosswalked by distance to PeeringDB, VA DEQ, EPA discharge permits, FAA crane notices and the prime-power register
licence
ODbL 1.0 — derived from the OpenStreetMap register; each linked record keeps its own licence

open the raw data →

How this is known
last success
not a collection: built from other instruments' payloads, with no run of its own

The tape — daily readings

16 readings as daily files: value, unit, count, denominator, vintage and lamp on every row, rebuilt from the archive and checked against it.

holdings
16 · series · counted from the file generated 2026-11-02 03:25 UTC
grain
day (the UTC day a board was built); each row's vintage says what period it describes
first reading on the tape
2026-09-14 · declared 2026-09-15 in instruments.yaml
last run
none recorded
feed checked: time not stated
last success
no successful run identified
source
WAPR's own tape of readings, one row per UTC day a board was built, rebuilt from the repository history by build/series.py --check
licence
WAPR compilation of public records; each row names the record it was read from

open the raw data →

How this is known
last success
not a collection: built from other instruments' payloads, with no run of its own

Dated transitions per site

For every mapped site, the dated public records that mark its life — mapped, listed on the network registry, cranes filed beside it, an air permit, a discharge permit, a year of recorded emissions, the first month a grid-connected generator of 1 MW or more near it produced power — each naming its document and whether the date is the document's own, a bound the document sets (on or before) or the day WAPR first saw it.

holdings
5,081 · ids · counted from the file generated 2026-11-02
grain
event (each transition dated by its document or by first sight, and says which)
first collection
2026-09-15 · as declared in instruments.yaml
last run
none recorded
feed checked: time not stated
last success
no successful run identified
source
WAPR's dated state changes per WAPR id, read from ids.json and the network-registry, permit, crane, emissions and generator instruments
licence
ODbL 1.0 — keyed to the OpenStreetMap register; each event's document keeps its own licence

open the raw data →

How this is known
last success
not a collection: built from other instruments' payloads, with no run of its own

Grid iron beside each site

For each of the 4,796 mapped facilities: how many substations and power plants OpenStreetMap maps within 5 km of it, how many of those substations carry a voltage tag, and the strongest and the nearest tagged substation. The 6,501 substations and plants tagged 100 kV or more are also listed one by one, each beside its nearest facility.

holdings
4,796 · facilities · counted from the file generated 2026-09-15 06:22 UTC
grain
event (one sweep of the cells the register occupies)
first run logged
2026-09-15 · the earliest day in its run manifest
last run
2026-09-15T06:22:27Z · derived
feed checked 2026-09-15 06:22 UTC
last success
2026-09-15T06:22:27Z
re-derived
2026-09-21T19:28:11Z — re-derived offline from files already held (not a collection)
source
OpenStreetMap power infrastructure (power=substation, power=plant) via the Overpass API, licensed ODbL 1.0
licence
ODbL 1.0 — © OpenStreetMap contributors

open the raw data →

How this is known
last success
the latest closing row with state 'derived' written by a run that queried Overpass (--replay-day rows are offline and excluded)

What else produces a water reading

For each characteristic measured in the water, the causes the US Geological Survey's own pages name for a reading like it — a term is kept only if the cited page contains it. WAPR does not decide which cause applies at any station.

holdings
7 · characteristics · counted from the file generated 2026-09-26 10:36 UTC
grain
static reference (each term checked against the page it cites when built)
first collection
2026-09-15 · as declared in instruments.yaml
last run
none recorded
feed checked: time not stated
last success
no successful run identified
source
US Geological Survey Water Science School (public domain), one page per characteristic
licence
USGS public domain

open the raw data →

How this is known
last success
not stated: build/water_causes.py writes no manifest, so nothing here records a completed run. It fetches the USGS page each cause term cites and keeps the term only where that page's article text contains it, and water-causes.json carries the verification itself: every kept term with the sentence it was found in, every term dropped, and the article's sha256

Proposed, under construction and gone

Data centres OpenStreetMap's mappers have tagged as proposed, under construction, disused or gone — kept apart from the register of existing buildings, so nothing here changes that count.

holdings
161 · sites · counted from the file generated 2026-11-01 11:16 UTC
grain
weekly (rebuilt with the register by tools/refresh_weekly.sh)
first run logged
2026-10-28 · the earliest day in its run manifest · declared 2026-09-17 in instruments.yaml
last run
2026-11-01T11:16:10Z · ok
feed checked 2026-11-01 11:16 UTC
last success
2026-11-01T11:16:10Z
source
OpenStreetMap via Geofabrik continent extracts, osmium tags-filter
licence
ODbL 1.0 — © OpenStreetMap contributors

open the raw data →

How this is known
last success
the latest row with state 'ok' (build/lifecycle.py rebuilt the layer from the re-cut extracts)

Counts are measured from the files each time the catalogue is built — never carried over from a document. Where nothing was counted, the row says why — never a zero.