Listening for events…
← All Docs

Dex source triage — agency · URL · best dex

Started 2026-06-19. The working ledger for the admin-URL-draining campaign: every real datasource (144 with observations, the AutoSense stub tail excluded) mapped to the best dex family for it. Column C is the recommendation; we lock each as Mike works the backlog. Bright line throughout: measured reality only (feedback_measured_reality_only) — forecasts, reanalysis, scenario/economic models are held out, never dexed.

Five dex families: Eventdex (slot = one event) · Yeardex (slot = one year) · Locationdex (slot = one fixed place/station) · CelestialObjectDex (slot = one celestial object, e.g. an asteroid; docs/celestialdex-framework.md) · YearLocationdex (slot = one place × time cell; docs/yearlocationdex-framework.md). Not every source is dex-shaped: a continuous measured point-series with no event/year/place/object structure is a paper-feed (stays a live source, papers draw from it directly, no dex).

A. Already dexed — done, no action

Agency Source URL Dex
USGS usgs_earthquake earthquake.usgs.gov/fdsnws/event/1/query Eventdex eq
NOAA SPC spc_tornado_history spc.noaa.gov/wcm/ Eventdex tor (+ NCEI severe-wx 15 kinds)
Smithsonian smithsonian_gvp webservices.volcano.si.edu Eventdex vol
NOAA/Kyoto dst_index · kyoto_dst_realtime cdaweb.gsfc.nasa.gov/hapi Eventdex gst
LIGO/Virgo gwosc_events gwosc.org/eventapi Eventdex gw
IceCube icecube_neutrinos dataverse.harvard.edu Eventdex neutrino
Pierre Auger auger_uhecr opendata.auger.org/catalog Eventdex uhecr
NOAA NCEI ibtracs_hurricanes ncei.noaa.gov/.../ibtracs Eventdex tc
FEMA fema_disasters fema.gov/api/open/v2 Eventdex fema
NASA/CNEOS/GMN cneos_fireballs · gmn_meteors · nasa_neo · meteorite_falls ssd-api.jpl.nasa.gov etc. Eventdex entry
NASA DONKI donki_flare api.nasa.gov/DONKI/FLR Eventdex flare
NASA DONKI donki_cme api.nasa.gov/DONKI/CME Eventdex cme
NOAA/Blitzortung goes18/19_glm_flashes · blitzortung_lightning noaa-goes*.s3 / blitzortung.org Eventdex lightning
NOAA NWS nws_alerts api.weather.gov/alerts Eventdex flood + severe-wx
Caltech/ZTF (Fink) fink_transients api.fink-portal.org Eventdex transient (ede746e, 2,350 objects; SN/kilonova/unknown — Solar System excluded)
The Space Devs launch_library ll.thespacedevs.com Eventdex launch (<pending>, 86 measured launches; scheduled excluded, spaceport-geocoded)
NUFORC nuforc nuforc.org Eventdex nuforc (<pending>, 80,324 sighting slots 1906-2014; report log w/ ground point; spine-parquet first, dossiers Brick D)
USDA ERS usda_landuse ers.usda.gov/.../major-land-uses Yeardex landuse
USDA ERS usda_dairy ers.usda.gov/.../dairy-data Yeardex dairy
World Bank worldbank_forest data.worldbank.org Yeardex forest_cover
World Bank worldbank_energy data.worldbank.org Yeardex energy_use
NASA/JPL CNEOS cneos_cad (NEW spine; nasa_neo/jpl_sbdb lacked approach dates) ssd-api.jpl.nasa.gov/cad.api CelestialObjectDex neo (<pending>, 18,749 asteroids / 30,362 approaches; 0.05 au, 2000-2100)
CelesTrak celestrak_satcat celestrak.org/pub/satcat.csv CelestialObjectDex satcat (<pending>, 69,352 space objects; satellite=object w/ orbital lifecycle, Mike's ruling; spine-parquet)
NMDB nmdb_stations · nmdb_cosmic_rays nmdb.eu Locationdex neutron_monitor
NOAA CO-OPS noaa_tide_stations · noaa_tides tidesandcurrents.noaa.gov Locationdex tide_gauge
INTERMAGNET intermagnet imag-data.bgs.ac.uk Locationdex magnetic_observatory (0324523, 154 obs)
USGS usgs_water waterservices.usgs.gov/nwis/iv Locationdex streamgauge (eca9c7f, 9,478 gauges)
EPA RadNet epa_radnet · epa_radnet_stations · epa_radnet_gamma data.epa.gov/efservice (ERM) + radnet/radnet-csv-file-downloads (gamma) Locationdex radiation_monitor (b841077, 333 stations; EURDEP→RadNet pivot). +2nd series 2026-06-27: near-real-time gamma (dose rate + count bands, 17.9M readings, 150 slots)
USDM / NCEI cdc-historical (USDM-MOD archive) + usdm_drought (live) ncei.noaa.gov/.../nidis/CDC/historical/ YearLocationdex drought (<pending>, 4,198,880 county×week cells, 3,220 counties 2000-2024; binary D1+ flag + area_pct 2000-2021; SPEI held out; 1st YearLocationdex; live edge deferred)
EPA NARS — Nat'l Lakes Assessment (new build; not a prior PG source) epa.gov/national-aquatic-resource-surveys/data-... YearLocationdex nla_water_quality (cdb218e+2022 backfill, 3,572 lake×survey-year cells, 2007+2012+2022; raw chem/clarity/chl-a IN, design-based condition estimates + MMI OUT; 2nd YearLocationdex; NTL mg/L→µg/L harmonized; 2017 + NRSA/NWCA siblings deferred)
EPA NARS — Nat'l Coastal Condition Assessment (new build; not a prior PG source) catalog.data.gov/.../national-coastal-condition-assessment-2015-datafiles YearLocationdex ncca_water_quality (861dc7e, 1,093 coastal-site×year cells, 2015 estuarine+GreatLakes; raw chem/clarity IN, POPESTIMATES population weights OUT; 3rd YearLocationdex; TN/TP mg/L→µg/L shared w/ NLA; NCCA 2010 + NWCA sibling deferred)
EPA NARS — Nat'l Rivers & Streams Assessment (new build; not a prior PG source) epa.gov/national-aquatic-resource-surveys/data-... YearLocationdex nrsa_water_quality (e8d17de, 5,904 river-site×cycle cells, 3 cycles 2013-14/2018-19/2023-24; raw chem IN, *_allcond + MMI OUT; 4th YearLocationdex, 1st NARS kind w/ real year axis; full NLA vocab minus Secchi; TN mg/L→µg/L but TP already µg/L; utf8-lossy; NRSA 2008-09 deferred)
EPA NARS — Nat'l Wetland Condition Assessment (new build; not a prior PG source) epa.gov/national-aquatic-resource-surveys/data-... YearLocationdex nwca_wetland_chemistry (778dda5, 2,849 wetland-site×cycle cells, 3 cycles 2011/2016/2021; 5th YearLocationdex, last NARS sibling — breaks the water-only mold: wetlands' headline signal is SOIL (~92% of sites) not water (~55%), so Mike ruled ONE kind carrying both blocks — shared water vocab ptl/ntl/cond/ph/chla + new soil block soil_carbon/nitrogen/sulfur_pct/ph_h2o/cec; soil-only & water-only sites both kept ("data is data"); TN mg/L→µg/L, TP already µg/L, soil C/N/S percent; 2011 soil = surface horizon, 2016 = stddepth core, 2021 no soil → null; condition estimates + veg MMI OUT)
EPA — 2011 PR coral-reef survey (new build; not a prior PG source) catalog.data.gov/.../2011-pr-survey-data Locationdex coral_reef_station (465570c, 64 reef stations, southern PR coast 2011; 6th Locationdex kind; measured reef condition IN — stony-coral %live/bleached/diseased, fish density+richness, rugosity; design-based regional condition estimate not carried; expanding multi-survey registry w/ survey-namespaced slots 46357c0; secondary taxa + sweep deferred)
NOAA NCEI ncei_coop_hpd (retired DSI-3240 /hourly_precip-3240/<state>/ → v2) ncei.noaa.gov/data/coop-hourly-precipitation/v2 Locationdex precipitation_station (<pending>, 2,093 COOP rain gauges, 54 states/territories, records 1940-2026; 11th Locationdex kind; ROSTER-based like radiosonde — identity + coords + POR + %-good from the v2 station-inventory, hourly series referenced not embedded; measured gauge depths IN; the many per-state /hourly_precip-3240/NN/ URLs all collapse into this ONE kind. Series deep-pull + live edge deferred)

B. Dex CANDIDATES — actionable, not yet dexed (work these)

Agency Source URL Best dex Note
EMSC / ISC / GFZ emsc · isc · gfz_geofon seismicportal.eu / isc.ac.uk / geofon.gfz Eventdex eq (cross-source merge) Earthquake catalogs → merge into eq like NCEI→tor.
NOAA NHC nhc_storms nhc.noaa.gov/CurrentStorms.json Eventdex tc (live edge) Live TC fixes → feed the existing tc kind.
USGS usgs_volcanoes volcanoes.usgs.gov/hans-public Eventdex vol (merge) or Locationdex Volcano activity/alerts → feed vol, or volcano-monitoring places.
EU JRC global-surface-water-explorer (GSWE) jeodpp.../GSWE Yeardex (surface-water extent) Satellite-MEASURED water occurrence. Annual. Measured.
World Bank world_bank (generic) api.worldbank.org/v2 Yeardex (more indicators) Same engine as forest/energy; pick measured indicators.
NOAA NCEI igra_soundings ncei.noaa.gov/pub/data/igra ✅ Locationdex radiosonde_station — built 2026-06-27 07d4ced, 2,921 stations, roster-based Fixed launch sites, measured profiles. Place = the site.
GeoSphere Austria geosphere dataset.api.hub.geosphere.at Locationdex weather_station (regional add) Austrian met network, measured. weather_station kind now BUILT from US NCEI LCD (<pending>, 2,784 US stations); geosphere is a future regional sibling.
Safecast safecast api.safecast.org/measurements.json Locationdex radiation (deferred) ~half roving drive-surveys → binned-mobile special case later.
EPA CASTNET castnet-outgoing-data gaftp.epa.gov/castnet/CASTNET_Outgoing/data/ Locationdex air_quality_station 159 rural sites (geocoded); ozone 1987-2026 + gases + met + filter-pack chem conc MEASURED. OUT: model_output_vd_* (Vd), total_deposition_mlm (MLM), deposition FLUX (conc×modeled Vd). PG's 2.49M obs = AutoSense catalog stubs, not data. (Triaged 2026-06-26.)
EPA ORD benthic-cyanobacteria stormwater (Tatters 2021) data.gov/dataset/benthic-cyanobacterial-accumulations-...-coastal-urban-stormwate (pasteur 10.23719/1532064) water_quality_station (BLOCKED on geolocation) MEASURED (IN): SPATT cyanotoxins ([ATX]/[HATX]/[CYL]/[RR]/[LR], 41 rows), CBL nutrients (BH01-BH10), cyano genus presence, LC/MS/MS microcystin congeners. Tatters et al. 2025 Harmful Algae 144:102833. Not built 2026-07-01: fails Locationdex place gate — ~10 informal pond labels (Pond 1, JW, Bridge, Sabine, UNTP, with typos), no coordinates in any file, one pond network, single Fall-2021 campaign (no roster mass, no series). Not a model reject (it's measured). Revisit if pond coords surface (paper SI / map) or fold into a future broad STORET/WQX cyano-toxin pull.

(EDGAR removed from candidates 2026-06-26 → ruled OUT, see §C. magnetic_observatory / streamgauge / radiation_monitor shipped → moved to §A.)

C. Held out — model / forecast / reanalysis / scenario (NOT dex material)

The measured-reality bright line. These stay live platform sources but are barred from dexes and papers (feedback_measured_reality_only).

  • Open-Meteo family — open_meteo (forecast), open_meteo_aqi, open_meteo_flood, open_meteo_marine — forecast model output.

  • ERA5 / ECMWF / Copernicus — era5_cloud (reanalysis), ecmwf (forecast).

  • NASA POWER — nasa_power — modeled/satellite-derived meteorology.

  • NOAA SPC outlook — spc_outlook — convective forecast.

  • EU JRC flood/climate models — cems-glofas, cems-glofas-ra-streamflow-analysis, copernicus-emergency-management-service-european-flood-awareness-system (EFAS), floods (JRC), petseta-iv (PESETA climate-impact projections), ensemble, ensembles-bc (climate model ensembles), succast, gec-outlook, sherpa (AQ scenario), mappe, eurad (AQ model), figaro-e3, figaro-reg (economic IO models), transuranus (fuel-rod model).

  • EU EDGAR — edgar — emissions inventory, activity-derived estimate (RULED 2026-06-26).

  • EPA GHG-by-industry (FLOWSA) — data.gov/dataset/2012-2020-greenhouse-gas-national-and-state-level-emission-totals-by-industry (EPA pasteur 10.23719/1529805, GHGs by 114/540 NAICS sector + households/government, US

    • by-state, 2012-2020). RULED OUT 2026-07-01 on model, both counts: the base is the U.S. GHG Inventory (mostly emission-factor estimation, activity × factor, the same line held out at facility level in power_plant/eGRID) AND the sector attribution is FLOWSA v2.0.0 input-output allocation (the About sheet says "generated using... FLOWSA v2.0.0" flow-by-sector methods). No measured subset to carve out (unlike eGRID's CEMS). The measured version of "who emitted how much" is the EPA GHGRP direct facility reports / CEMS-metered CO2 subset — already the spine behind the power_plant Locationdex; grow that, not this attribution product.

  • EPA lead-exposure hotspots — data.gov/dataset/a-u-s-lead-exposure-hotspots-analysis (EPA pasteur 10.23719/1529907, supplement to Zartarian et al. 2024, ES&T 7(7):3311-3321). RULED OUT 2026-07-01 on model: the distributed product is RandomForest v1/v2 model predictions.csv — a machine-learning estimate of lead-exposure risk per census tract (GEOID + Census/ACS predictors → RF..._pred). The bundled MI/OH training tables do carry an Observed elevated-blood-lead column (a real measurement), but it's a two-state slice packaged as training data, not the source of record. Measured path = CDC / state childhood blood-lead surveillance (a separate authoritative feed to add if a measured lead-exposure layer is wanted); the demographic predictors belong to Census/ACS directly.

  • EPA S+N atmospheric deposition 1990-2010 — data.gov/dataset/1990-2010-atmospheric-deposition-of-sulfur-and-nitrogen (EPA pasteur 10.23719/1434279, data files for Zhang, Mathur, Bash, Hogrefe, Xing & Roselle, ACP: long-term US S+N deposition trends). RULED OUT 2026-07-01 on model: the product is a WRF-CMAQ reconstruction. Bulk files are gridded model output (17,632 regular cells, incl. open-ocean points with no monitor); dry deposition (DDEP) is inferential (conc × modeled deposition velocity) and total deposition (TDEP) = wet + modeled-dry inherits it; the emissions-trend file is inventory. The only measured content is 170 stations' _Obs wet-dep rows (NH4/NO3/SO4/precip), a re-gridded evaluation slice repackaged for the paper's model-vs-obs comparison, not the source of record (same shape as the lead-hotspots reject). Measured path = NADP/NTN wet deposition directly (~250+ long-run precip-chem sites) + CASTNET concentrations → a station × year YearLocationdex kind (sibling of nla_water_quality); dry/total dep stay out.

  • EPA USEEIO import emission factors 2017-2022 (from EXIOBASE) — data.gov/dataset/useeio-models-with-import-emission-factors-for-greenhouse-gases-for-2017-2022-from-exiobas (EPA pasteur 10.23719/1531676; USEEIOv2.3-GHG models built with useeior v1.6.0). RULED OUT 2026-07-01 on model, no measured slice at all (the cleanest reject yet). Three stacked models: USEEIO is an environmentally-extended input-output economic model; EXIOBASE is a modeled global multi-regional IO database; the Import Emission Factors are calculated kg-CO2e per dollar of imported commodity (an intensity coefficient), mapped EXIOBASE->BEA->USEEIO by concordance files and "incorporated into these USEEIO models ... transformed into producer price ... found in the M_n sheet." An emission factor is not a measurement of physical emissions, and import-embodied emissions from foreign supply chains are inherently un-measurable from the US side, so there is no measured substitute for this concept. The measured neighbor already dexed is direct US point-source emissions via CEMS in power_plant (Locationdex), a different thing.

  • EPA Ohio landfill water balance / "moisture contents" — data.gov/dataset/assessment-of-ohio-landfill-moisture-contents (EPA ORD, pasteur 10.23719/1526474, J. Environ. Management 2023; Water balance Science Hub.xlsx, 17 figure-source sheets). RULED OUT 2026-07-03 (Mike) on model. The title says "moisture contents" but the data dictionary says otherwise: precip AND ET are both NLDAS-2 (land-surface reanalysis, held-out class = ERA5 / NASA POWER), moisture content is "Calculated" (back-computed from the water balance, not a lab gravimetric measurement — the headline quantity is itself a model output), the runoff coefficient CRO is "Assumed" (whole figures built on "Assumes CRO = 1/10/30/50%"), waste-in-place uses biodegradation decay (k = 0.04/0.057/0.09 yr⁻¹), Fig S5/S6 are disposal-to-leachate lag scenarios (+0..+12 mo), and Fig 8/S9 partition water into infiltration vs "water in waste assuming MC = 25%". Fundamentally a landfill water-balance MODELING study, same footing as the MRB water-temperature reject (water-balance state variables) and the reanalysis hold-outs. The only measured slice is metered leachate volume per landfill per year (Fig 2, m³) — but that is thin landfill-OPERATIONS metadata (same low-value engineering/ops class as the parked floating-solar PV + GWOSC uptime), no coordinates in the file, and entangled with the model framing; it does not rescue the set or earn a kind. Landfills are otherwise already covered by the landfilled_waste_composition YearLocationdex (measured state×year tonnages).

D. Paper-feeds — measured but continuous point-series, no event/year/place shape

Stay live; papers draw directly. No dex.

  • NOAA SWPC / space-weather: dscovr_solar_wind, goes_xray, noaa_space_weather, nasa_donki (general bundle).
  • Solar/sun: silso_sunspots (SIDC Belgium).
  • Carbon/climate indices: noaa_co2, noaa_climate_indices (could go Yeardex if an annual subject is wanted; default paper-feed).
  • Radio/propagation: wspr, hamqsl_propagation, lwa_spectra.
  • Drought: usdm_drought — now ALSO feeds YearLocationdex drought (§A) as the eventual live edge; stays a live paper-feed unchanged (parallel organization, not redundant). The county×week historical backfill came from cdc-historical (the USDM-MOD archive, an AutoSense catalog stub in PG — the real CSVs are fetched by the build script).
  • Satellite/misc measured: asf_sentinel, gmis, magnetic_poles (NCEI pole positions).
  • Tidal ephemeris: lunar_tidal (computed astronomical potential — context, not a measurement of weather/Earth state).

E. EU JRC research tail — mostly model/scenario/economic (RULED OUT 2026-06-26 as a block)

~40 jeodpp.jrc.ec.europa.eu/ftp/jrc-opendata/* datasets are EU Commission research products: economic input-output models, land-cover/scenario maps, transport tools, energy models, biodiversity indices. Block ruling 2026-06-26 (feedback_measured_reality_only): model/scenario/economic products = computed estimates, HELD OUT. A few are genuinely measured research datasets that don't fit event/year/place; pull a specific one into B/D only if Mike flags it by name and it clears the bright line on inspection. Examples: alf-bio, algae-industry, anure, carbon-emissions, dopa, drll, droughtobs, eanr, ecml-data, ecoclim, eigl-data, eu-cbm-hat, eucropmap, forest (JRC), fwpi, ghg-inv, gmis, global-dam-watch-database, imap-fp-dataset, jrc-*, kcbd-datasets, liscoast, maes, metis-mix-2030, pvgis, remdb, s2-gold, strategic-trade-atlas, tedam, tnet, tourism, trace, transport-*, trimis, uno-films-xps, upu-gamma, vaia-cbm-database, vecsec-volts, plus the data.gov dataset-* research sets (cyanobacteria, atmospheric deposition, anthropogenic emissions, PM2.5 by county, etc.).

F. Reference / spine sources — no measurements, not a dex on their own

Neither measurement nor model: master directories / cross-walk keys. Useful as plumbing to locate and link other measured feeds, but they carry no measured values, so they are not a dex by themselves. Park; build only when a measured feed needs the spine.

  • EPA FRS (Facility Registry Service) — epa.gov/frs/epa-state-combined-csv-download-files → ordsext.epa.gov/FLA/www3/state_files/state_combined_<st>.zip (+ national_combined.zip). EPA's master directory of regulated facilities (~4M nationally; DC alone 4,578). Per-state ZIP = FACILITY (name/address/coords/FIPS/SITE_TYPE), PROGRAM_GIS (LATITUDE83/LONGITUDE83), ENVIRONMENTAL_INTEREST (which EPA programs touch each REGISTRY_ID + some enforcement dates), NAICS/SIC, CONTACT, ORGANIZATION. No environmental measurements anywhere — pure registry + program cross-walk keyed on REGISTRY_ID. PARKED 2026-06-27 (Mike): record as the master facility spine, build nothing until a measured EPA feed (TRI toxic-release amounts, smokestack emissions, water-discharge monitoring) is queued that needs to join to it. Not measured-reality on its own (feedback_measured_reality_only).

Next Locationdex candidates (Mike's "note them" ask)

Each its own scope freeze + full-roster deep-pull. Done:

  • magnetic_observatory ← intermagnet — shipped 0324523 (154 obs).
  • streamgauge ← usgs_water — shipped eca9c7f (9,478 gauges).
  • radiation_monitor ← EURDEP→EPA RadNet — shipped b841077 (333 stations). Refined 2026-06-27 with a 2nd measured series: RadNet near-real-time gamma (epa_radnet_gamma, 17.9M readings, 150 slots). Still revisit-able for exact monitor coordinates.

Remaining, priority order:

  1. radiosonde_station ← igra_soundings (fixed launch sites) — or keep as paper-feed.
  2. weather_station — BUILT 2026-07-03 from US NCEI LCD (<pending>, 2,784 US stations, roster-based; identity + coords + elevation + POR, LCD series referenced not embedded). Built from the structured LCD CSV feed after Mike pasted the 2023 LCD annual PDF directory. geosphere (Austria) is a possible future regional sibling; per-station climatology deep-pull deferred.
  3. radiation ← safecast (deferred, binned-mobile special case).

PARKED on extraction cost (measured, but not a clean CSV):

  • engineering_climatology ← NCEI Engineering Weather Data (EWD) — ncei.noaa.gov/pub/access/cebrequests/engineeringwx/. ~1,262 airport/stations worldwide (998 US "K"), two PDFs each (IP + SI metric) = 2,525 files; handbook at doc/EWD_Handbook.pdf. Each 18-page PDF is one station's multi-decade climatology design summary over a fixed period of record (e.g. KORD = 1988–2017), derived from that station's measured hourly observations. Slot = station → Locationdex. Measured-vs-model split applied field-by-field: IN = empirical occurrence percentiles (0.4/1/2/97.5/99/99.6%), median extreme highs/lows, mean-coincident values, hour-counts above AC thresholds, and the monthly dry-bulb-hours frequency tables (pages 5+) + monthly mean temp/dewpoint/precip — all descriptive statistics of observed hours, same class as climate normals / USDM assessed grid. OUT = the "Other Site Data" return-period extrapolations (rain rate 100-yr, basic wind 50-yr gust, ground snow load 50-yr) and the derived Ventilation Cooling Load Index — statistical estimates beyond the observed record / engineering computations, not measurements (feedback_measured_reality_only). PARKED 2026-06-27 (Mike): build later. Reason = the payload is locked in 2,525 PDFs (page-1 design tables + pages 5+ frequency tables are text-extractable; pages 2–4 are raster charts carrying no extra numbers), a heavier and more error-prone parse than a CSV feed. When built: pilot page-1 design bundle for US stations first, then extend to the monthly frequency tables / global stations.

PARKED on value (measured + admissible, but not worth a kind now):

  • solar_pv_performance_station ← UCF/DOE Florida Solar Energy Center floating-solar dataset — catalog.data.gov/dataset/high-resolution-floating-solar-pv-data (files at data.openei.org/files/6179/). Two-plus years of 1-minute MEASURED sensor data from four real floating-solar sites (Oakville & Windsor CA, Altamonte Springs & Orlando FL) + one co-located land-based reference array (Cocoa FL): module temperature (15 sensors/site), on-site met, system-level PV parameters, per-inverter output (12 inverters at Oakville). Clears the bright line — it's instrument telemetry from hardware that physically exists, not a floating-solar potential/suitability model. Slot = one instrumented site → Locationdex shape. PARKED 2026-07-03 (Mike) on value, not admissibility: it's an engineering-performance dataset (how five specific arrays perform), not an environmental phenomenon TerraPulse tracks; it's tiny (5 sites) and closed (a one-time DOE research window, no live edge, no growth); it cross-matches with nothing on the platform (power_plant is national annual generation, not device-minute data). The one genuinely interesting measured question, does floating cool the panels enough to beat land, is barely powered: only ONE site (Cocoa land vs. Orlando floating) is a co-located pair, so n=1. Revisit only if a broader floating-solar performance feed appears to join it to, or a specific paper wants the Cocoa-vs-Orlando cooling pair with eyes open on n=1.
  • EPA elevated-temperature landfill (ETLF) gas characteristics (measured, but not dex-shaped) — catalog.data.gov/dataset/elevated-temperature-landfill-gas-characteristics (EPA ORD, pasteur 10.23719/1527938, J. Environ. Management 2022; ETLF Gas SciHub Final.xlsx). MEASURED field gas-well data (CH₄/CO₂/O₂/H₂/CO %, temperature °C, ~20k records) at 5 elevated-temperature landfills, reported by operators to state agencies and EPA-compiled. No modeling (contrast the landfill water-balance reject); the only non-data bit is Fig 6's alert/caution threshold rubric (management guidance). PARKED 2026-07-03 (Mike): measured but unplaceable. The 5 landfills are anonymized "ETLF A–E" with no names and no coordinates — operator-submitted data about a stigmatized, litigation-prone condition (subsurface exothermic/smoldering reactions, the Bridgeton/ West Lake archetype), de-identified under confidentiality so EPA could publish the science. That fails the Locationdex place gate (can't resolve "ETLF C" to a map point, same wall as the benthic- cyano block) and fits no other dex spine (not annual-national, not events, not identifiable objects); subject is a niche waste-management pathology, ops-flavored monitoring. Revisit only if a landfill- methane paper wants it as a paper-feed (rides un-located), never as a dex anchor.
  • EPA coral ocean-acidification tank experiment (measured, but a lab manipulation) — catalog.data.gov/dataset/carbonate-chemistry-water-quality-coral-measurements (EPA ORD, pasteur 10.23719/1411864, Enzor et al. 2018 J. Exp. Marine Biol. & Ecol.). MEASURED carbonate chemistry (pH, total alkalinity, pCO₂), water quality (temp, salinity, Ca, Mg, phosphate, ammonia, nitrate), and coral response (buoyant weight, calcification rate, tissue growth) on Caribbean reef corals over a 3-month exposure. PARKED 2026-07-03 (Mike): measured but out-of-scope in spirit. It's a recirculating ocean-acidification EXPOSURE SYSTEM — a lab tank with pCO₂ imposed as a treatment, built to estimate what would happen to corals under projected acidification. Every number is real (not a computer model, so not a clean bright-line reject like the landfill water balance), but TerraPulse ingests measured environmental reality (passive observation of what is physically happening), and a manipulated aquarium is neither passive nor environmental — the "what would happen" side of the line reached with tanks instead of a computer. It's the BFB wetland experiment pattern taken further (BFB was at least a real field wetland; this is fully artificial), no reef sites / no coordinates, and it is NOT a reef survey so it does not fold into coral_reef_station (field reef stations). No dex home. Revisit never as a dex; only relevant if a paper explicitly wants OA-experiment coral-calcification response as literature context.
  • Elliott et al. 2019 NOx/NH₃ δ¹⁵N source-signature review compilation (reference table, not dex-shaped) — catalog.data.gov/dataset/2019-elliott-et-al-nox-and-nh3-isotopic-signature-review-paper-dataset (EPA, pasteur 10.23719/1518391, Sci. Total Environ. 2019; NOx and NH3 isotopic signature.xlsx). A literature end-member library: published δ¹⁵N values for NOx and NH₃ emission source TYPES (lightning, vehicle exhaust, coal plant w/ & w/o scrubber; livestock waste, volatilized fertilizer, human waste…), each as a (Paper, ‰) pair citing a different upstream study (Hoering-1957, Heaton-1987, Felix-2012, Chang-2016…). Plus an IPCC-2013 budget sheet in TgN/yr labeled "TAR (model 1)/AR4 (model 2)" — explicit model output, held out. PARKED 2026-07-03 (Mike): measured- underlying but not dex-shaped. The δ¹⁵N values are real mass-spec, but this is a SECONDARY compilation (each datum traces to a different original paper — fragments the "every datum → one clean source" provenance standard, feedback_provenance_every_datum), organized by source TYPE with no place/time/event/entity spine to key against, and its purpose is to feed isotope mixing / source- apportionment MODELS on measured atmospheric N (a modeling technique) — the same reference- coefficients-for-a-model character as the FLOWSA/USEEIO rejects, except the numbers happen to be measured. Closer to a constants table than a phenomenon. Revisit only if a nitrogen-deposition source-apportionment paper wants the end-member library as literature context, never as a dex.
  • GWOSC detector status / duty cycle (gw Eventdex companion, NOT a new kind) — gwosc.org/detector_status/. A status portal (no direct dataset/API on the page itself) for the LIGO/Virgo/KAGRA detectors: daily status, per-run observing summaries (O1–O4c, back to Nov 2016), links out to GWOSC segment data, the live online.igwn.org dashboard, and GraceDB. PARKED 2026-07-03 (Mike) on two grounds. (1) The citable layer is an interpretation. The bare operational state (detector locked / in observing mode) and the duty-cycle % derived from it are measured/arithmetic, but the headline product is the Data Quality veto segments (CAT1/2/3, split by CBC vs burst search): the detchar team's analysis-specific JUDGMENT that a stretch of data is good/bad, encoding glitch/artifact/environmental-contamination decisions. That is an interpretation/ classification, held out on the same footing as design-based condition estimates and MMI indices (feedback_measured_reality_only). (2) The clean slice is thin ops bookkeeping, not a phenomenon. The gravitational-wave events (strain measurements) are already the gw Eventdex; detector uptime is machine metadata whose only real use is the exposure denominator (duty cycle × sensitivity) for a gw detection-rate analysis — speculative, none queued, and gw is still held out of some cross-matches pending skymaps. Payload isn't on the pasted URL; do NOT auto-hop to the GWOSC segment API / online.igwn.org. Revisit only if a gw-rate/exposure paper needs the duty-cycle denominator, in which case pull GWOSC's own observing-segment data (gwosc.org, the advertised path) as a gw companion layer, not a standalone kind.
  • BFB lead-contaminated wetland amendment experiment (would be water_quality_station study #9) — catalog.data.gov/dataset/amendment-effectiveness-and-seasonal-variations-in-a-lead-contaminated-acidic-wetland (EPA ORD, pasteur 10.23719/1532242, Applied Geochemistry 2025; two xlsx: BFB-Sci-Hub.xlsx + BFB-Rainfall-Data.xlsx). One Superfund wetland, a controlled lead-remediation experiment: 9 treatment plots (U1–3, W1–3, M1–3) × amendments (Control, Reactive Substrate, Apatite II surface vs mixed), ~26 months (Oct 2021 – Nov 2023). Bright line, mostly measured: IN = soil Pb/P/S concentrations + P:Pb / S:Pb ratios, porewater Pb, pH/ORP/phosphorus/sulfate, hourly water level (3 piezometers + 4 stilling wells) and hourly rainfall (3 gauges) over the full 2 years. OUT = the Pb XAS speciation fractions (Fig 4/5: Anglesite/Cerussite/Galena/Pyromorphite/Plumbojarosite… + R-factor), which are linear-combination fits of X-ray absorption spectra to reference minerals — a model-fit interpretation of the measured spectrum (analyst-chosen basis), same class as the GWOSC DQ flags and the MMI/condition estimates held out elsewhere (feedback_measured_reality_only). PARKED 2026-07-03 (Mike) on value + fit: it's a manipulative EXPERIMENT on 9 plots at ONE wetland (every other water_quality_station study is passive monitoring), ships no coordinates, and the scientific payload people cite (did the amendment work, via Pb speciation) lives in the held-out XAS layer — strip that and the ingestible remainder is fairly generic soil/porewater chemistry from a handful of plots at a single site. Revisit only if a wetland-remediation or Pb-mobility paper wants the measured chemistry + hydrology (XAS still held out), in which case fold as water_quality_station study #9 with plots as null-coord slots (the Georgia CP pattern).
  • EPA Sentinel-2 / chlorophyll-a matchup table (would be a water_quality_station study, but a matchup artifact) — catalog.data.gov/dataset/matched-sentinel-2-spectral-data-and-chlorophyll-a-concentrations-2015-2020 (EPA ORD, pasteur 10.23719/1529420; one workbook S2_chla_matchups.xlsx, 300 rows, 2015–2020). Each row pairs one in-situ chlorophyll-a sample with the Sentinel-2 scene that overpassed near the same time. Bright line: IN = In.Situ.chl ("in situ chlorophyll a concentration µg/L", field-measured, pulled from the Water Quality Portal) + the top-of-atmosphere rhot.NNN bands (measured satellite radiance). OUT/derived = rhos/Rrs (atmospheric-correction-model output) and the MCI/NDCI band-ratio indices. So the parts that matter clear the line. PARKED 2026-07-03 (Mike): measured but not dex-shaped (the same discriminator that BUILT Taylorsville/Harsha cuts the other way here). (1) It is an algorithm-training / matchup artifact, not a monitoring campaign: ~54 of 64 columns are satellite spectra; the file exists to develop a Sentinel-2 chl-a retrieval algorithm (methodology-dataset class). (2) No station spine — 300 records but 234 unique lat/lons spread across the whole CONUS (Lake Erie to Florida to Puget Sound), scattered one-off points selected for a clear-sky overpass, with no site/station ID column at all (only the satellite Scene.ID); that fails the water_quality_station grain (slot = a fixed monitoring station). (3) It's a selection-biased derivative — the in-situ chl-a is extracted from the Water Quality Portal, which is the real station-identified national source we'd ingest if we wanted chl-a stations, not this cloud-free-overpass subset. Contrast the BUILT 8th study (Taylorsville/Harsha): also a satellite chl-a portability study, but its payload was a full limnological survey (nutrients, pigments, ELISA microcystin, sonde, phytoplankton counts) at named sites; here the in-situ payload is chl-a only, at anonymous points. The payload is the discriminator, not the satellite framing. Revisit only if we onboard the Water Quality Portal itself as a station-identified source.

G. International source links (Mike's saved world list, from 2026-07-19)

New triage axis this batch: commercial-use licensing is now checked up front for every source, because TerraPulse is a commercial product (feedback_commercial_use_license_check). A source that bars/omits a commercial grant is a real disqualifier, not a footnote — it can moot the dex question entirely (see DFO).

  • Afghanistan Meteorological Department (weak direct source → SUPERSEDED by existing IGRA) — www.amd.gov.af. National met service (EN/Pashto/Dari). REJECTED 2026-07-19 (Mike) as a direct source on three grounds: (1) the site is mostly forecasts (3-day/weekly/seasonal = model output, held out); (2) its only measured stream (radiosondes) is already in NOAA IGRA, public domain and cleaner; (3) commercial-use problem — footer "© 2026 … All Rights Reserved", no open license, plus a sanctioned-government posture for a US commercial platform. Action taken instead: backfilled Afghanistan's IGRA period-of-record into the existing global radiosonde_station LocationDex — 5 stations, 3,598,905 rows, 1961-2021 (see [[project_igra_por_backfill]]). No amd.gov.af ingestion; nothing to revisit.
  • Dartmouth Flood Observatory (DFO) (would be a strong flood Eventdex, but LICENSE-BLOCKED) — floodobservatory.colorado.edu. CU Boulder (Brakenridge/Kettner); space-based surface-water measurement. The prize is the Global Active Archive of Large Flood Events — one row per observed flood, ~1985→present, with dates/centroid/cause/dead/displaced/area/severity = textbook Eventdex shape (like tc/eq). Also 335 satellite-discharge stations (LocationDex shape). PARKED 2026-07-19 (Mike) on commercial-use — the terms cannot be resolved from the site. The MediaWiki Disclaimer + About pages are empty stubs (404 "no permission to create"); no license/copyright/commercial statement anywhere; it's a university research product, NOT a federal work (so not auto-public-domain), and the historical framing is "freely available for research and humanitarian use" — which never addressed commercial reuse. Silence ≠ permission → cannot assume commercial-OK. The download URL is therefore moot until cleared. Measured-reality nuance for later: the events archive is clean-observed; the discharge product is a satellite-microwave retrieval w/ a rating-curve step (gray zone). Revisit only with written commercial-use clearance from Kettner (kettner@colorado.edu) — draft-email offered, Mike would send; the Large-Flood Archive is worth that ask.
  • NASA GPM / IMERG (clean license, but POOR DEX FIT) — gpm.nasa.gov. NASA+JAXA global precipitation (2014→present; IMERG Final reprocessed to 2000). License is the clean case — NASA Earth-science data is fully open/public-domain, commercial-OK (free Earthdata login to download only). DPR radar + GMI microwave = clean measured swaths; IMERG = multi-satellite blend (microwave retrieval + IR/motion gap-fill + monthly gauge calibration) — not a model (no forecast/reanalysis/ projection, no atmospheric simulation) but not raw-instrument either; the morphed gap-fill is inferred where no sensor was overhead. PARKED 2026-07-19 (Mike) — NOT on license or the IMERG call, but on dex fit + wrong-tool. GPM is a continuous global raster (0.1°/30-min), no per-object grain → not dex-shaped; only becomes a dex if derived into an extreme-precip Eventdex. On sanity-check that derived idea is weak: IMERG is weakest exactly at the extremes (0.1°/30-min smears intense convective cores, retrieval underestimates peaks, IR gap-fill crudest for short-lived convection), and its 2000→present record is inhomogeneous (TRMM-era vs GPM-era constellation, growing sensor mix) → fake extreme-frequency drift = the coverage-artifact confound we avoid. For extreme rainfall events, direct gauges (GHCN-Daily) beat IMERG on both measurement-purity and accuracy, and we already have a precipitation_station LocationDex frame. Revisit only as a spatial gap-fill paper-feed for ungauged regions (not a dex), or if a gauge-based extreme-precip event dex is ever wanted (that's GHCN, not GPM).
  • European Drought Observatory (EDO/GDO) (EU JRC drought portal — mostly held-out model indices) — edo.jrc.ec.europa.eu → 302-redirects to drought.emergency.copernicus.eu (same operator, rebranded; followed the site's own redirect, not a mirror). JRC / Copernicus Emergency Management Service. License is CLEAN (Copernicus free-and-open, CC BY 4.0, commercial-OK w/ attribution) — not the blocker here. PARKED 2026-07-19 (Mike) on measured-reality + redundancy + the standing EU JRC block. Four headline products, ¾ held out: CDI (classification blend of the other indices — derived), SPEI ("based on ERA5" = reanalysis/model, doubly out — index on a model), soil moisture anomaly (labeled model output). Only fAPAR anomaly is genuinely measured (satellite vegetation greenness). Three more strikes: (1) the EU JRC tail was already RULED OUT as a block (§E, 2026-06-26, mostly model/scenario) and EDO is squarely in it; (2) drought is already dexed — the drought YearLocationdex (USDM), and EDO is the EU analog built on the indices we exclude; (3) the lone measured layer isn't the clean access point — fAPAR anomaly is vegetation-minus-climatology (a drought proxy); if we ever want satellite vegetation we'd pull the primary global Copernicus/MODIS fAPAR product, not this drought portal's derived EU anomaly (the same downstream-product trap as IMERG). Revisit only if a satellite-vegetation dex is ever wanted, sourced from the primary global fAPAR/NDVI product — not EDO.
  • EMSC / CSEM (measured seismology, already ingesting — but COMMERCIAL-LICENSE BLOCKED, feed stopped) — www.emsc-csem.org (data portal seismicportal.eu). European-Mediterranean Seismological Centre, a non-profit NGO of 70+ member institutes. Observed earthquakes + crowdsourced felt reports = clean measured. Was a live source (emsc, 37,537 obs in PG, 2026-03-31→07-19) and a registered-but-empty emsc_eq kind; the plan of record was to fold it into eq as a confirming cross-source (§B line). NOT built — eq is 100% USGS (verified: 800/800 sampled slots cite only usgs_earthquake); the checkpoints claiming emsc_eq/gfz_eq/isc_eq were "built" do NOT match disk (no slots). PARKED + DEACTIVATED 2026-07-19 (Mike) on COMMERCIAL LICENSE. seismicportal.eu/terms.html: default is non-commercial — "Material may be reproduced for personal, academic, educational, non-commercial research or other non-commercial use." Only the felt reports are explicitly CC BY 4.0 (commercial OK), plus a vague "some datasets… CC BY 4.0" that does NOT name the event catalog → the core seismicity we'd merge falls under the non-commercial default = a blocker for a commercial platform. Saving grace: we don't need it — USGS ComCat is public-domain and already the eq spine (13,870 slots); EMSC's only unique add (Euro-Med confirming coverage) is exactly the non-commercial part. Action taken: commented out the (EMSCFetcher,300,"emsc") JOBS entry + its import in scheduler.py (removed from KNOWN_SLUGS), set the datasource active=false + audit note. Needs a terrapulse service restart to flush the running 5-min job. Existing 37,537 staged rows retained (staging only, never surfaced in a dex/site) — not purged (ask before destructive delete). GFZ GEOFON
    • ISC are sibling confirming feeds still running — each has its OWN licence, assess one at a time before touching. Revisit EMSC only if it ever relicenses the event catalog for commercial use.
  • Global Forest Watch (GFW / WRI) (aggregator; dex-shaped layer too-close-to-modeled) — www.globalforestwatch.org (API data-api.globalforestwatch.org). WRI forest-monitoring portal, 65+ datasets, rebranding to "Global Nature Watch." Already a dormant curated source internally (gfw_fires fetcher + datasource, inactive, 0 obs — never ingested; nothing to stop). License mostly fine (fire=NASA public domain; tree-loss/GLAD/RADD = CC BY 4.0 commercial-OK; some partner imagery layers proprietary) — not the blocker. PARKED 2026-07-19 (Mike) on measured-reality. Layer by layer: tree cover loss (Hansen/UMD) is the only clean dex fit — the GFW API serves it as an admin×year×canopy-threshold table (hectares lost) = textbook YearLocationdex shape — BUT it's a per-pixel classification (bagged-decision-tree verdict "canopy removed here in year N" from Landsat), and the hectare total moves with a user-chosen canopy-density threshold (10/25/30/50/75%). A number that shifts when you turn a knob is a derived index, not a reading → "too close to modeled info" (Mike), held out like GPM/IMERG and the EDO indices (feedback_measured_reality_only). Also "tree cover loss" ≠ deforestation (includes logging/fire/harvest/storms). GLAD/RADD alerts + carbon flux = same classification/model class, out. The one genuinely measured layer, VIIRS/MODIS active fire, is clean — but it's NASA FIRMS redistributed; take fire from FIRMS directly (primary, public domain, Phase-2), not the GFW middleman (the downstream-aggregator trap, cf. GPM/EDO). Redundancy note: coarse national forest already dexed (forest_cover Yeardex, World Bank). Revisit only if the measured-reality line ever admits satellite classifications, in which case source tree-loss from primary UMD GFC.
Live Feed