The datasets behind the poems, kept honest: what each one actually holds, the flagship built from it, and the fields still on the table. The dark panel at the bottom lists cross-dataset combinations worth building next, picked only where the data is genuinely compatible.
NASA / JPL SBDB · 41,845 NEOs
Near-Earth objects with orbital elements and JPL designations. 6,000 plotted, every one named.
field.json: points[a, e, i, diameter, class, pha, name] · total · pha_count
Flagship: Orbits (living web of 40K drifting bodies)
USGS · 3.7K quakes
Recent quakes with magnitude, depth, and location. The kept file bins by depth band.
field.json: points[mag, depth, lat, lon] · bands · magRange · depthRange
Flagship: The Depth of the Break (earth cross-section by depth)
NOAA NCEI · 200 volcanoes
Significant volcanoes with elevation, structural type, explosivity (VEI), and last eruption year.
volcanoes_clean.json: [name, country, type, elevation, vei, last_eruption]
Flagship: The Skyline of Fire (200 cones, height by elevation)
OBIS · 200K records (binned)
Deep-sea occurrences reduced to honest depth bands; only records with a recorded depth are placed.
depth_column.json: bands[band, minDepth, maxDepth, total]
Flagship: The Descent + The Abyss (depth column)
NASA · 46K meteorites
Recorded meteorite landings with mass, year, and fall/found class.
mass(g), year, lat, lon, recclass, fall
Flagship: Rain of Stone (log-mass fall)
Megalithic Portal · lat/lon
Standing stones, dolmens, and stone circles by location and type.
name, type, lat, lon, country
Flagship: The Standing Stones
NOAA (Kaggle) · 44.6M strikes
Real strike counts by location and time. Earlier sample data was rejected and replaced.
lat, lon, strike_count, date
Flagship: A Continent Lit by Lightning
lighthouses + bioluminescence · lat/lon
Two beacon datasets overlaid: human lighthouses against living light. The proof that combinations work.
lighthouses[lat, lon] · biolum[lat, lon, organism]
Flagship: Two Lights (combination, built)
GBIF · 8,439 real records
Carnivorous-plant specimens grouped by genus and trap mechanism (pitfall, snap, flypaper, suction).
garden.json: specimens[genus, trap, count] · meta
Flagship: The Herbarium of Hunger + The Carnivore Garden
peppers · 175 cultivars
Pepper cultivars by species and Scoville heat.
name, species, scoville_min, scoville_max
Flagship: A pepper that blooms open
Global Shark Attack File
Recorded attacks by date, location, activity, and outcome.
date, lat, lon, activity, fatal, species
Flagship: The Fin
Glottolog · 23K languoids
Language families sized by member count, with tiny families and isolates split out.
forest.json: families[name, size] · tiny_families · isolates · totals
Flagship: The Forest of Tongues
world languages · 7K
Languages by speaker count and endangerment status.
name, speakers, status, lat, lon
Flagship: The Vigil (endangerment stream)
WALS · 76K feature values
Typological features across languages: word order, case, phonology, more.
language, feature, value, lat, lon
Flagship: The Loom of Grammar (feature radial)
PHOIBLE · 105K phonemes
Phoneme inventories with phonological features and per-language segment counts.
phoneme_field.json: phonemes[segment, features] · seg_counts · curated
Flagship: The Phoneme Field (feature space)
The Onion · 2,103 headlines
Satirical headlines reduced to a frequency wordfield.
wordfield.json: words[word, count] · total_headlines
Flagship: Ink (drifting wordfield)
World Bank · Gini, 36 countries
Income inequality (Gini) as a time series per country.
gini_timeseries.json: countries[name, series[year, gini]] · indicator
Flagship: The Gap Moves (36 strands of Gini)
The promoted social poems (The Few, The Bigger Crime, The Bent Line, The Missing Girls, The Empty Desks, The Places Without, The Crossing, The Weight of a Number) are aggregate-stat pieces: each is its own front door on the hub, built from a single national/state statistic rather than a point cloud.
Cross-dataset poems, picked only where the two sources are genuinely compatible (shared geography, shared time axis, or a real thematic tension). Most datasets are better left alone; these few earn the overlay.
Human beacons against living light, both as lat/lon point fields. Already shipped as proof the form works.
Quake depths and volcanic cones share the Pacific rim. Plot quakes as a depth cross-section with volcanoes anchored at the surface above their tectonic neighbors.
Two ways the sky touches the ground, both dense lat/lon fields over decades. One slow and rare, one constant. A split-sky field.
Glottolog family sizes already carry the tree; world-languages carries which branches are dying. Color the forest by how much of each family is endangered.
PHOIBLE inventories and WALS grammar features both key on language. A poem where a language's sound-size pulls against its grammatical complexity.
Both coastal lat/lon. The fear and the wonder of the same sea: attacks massing into the fin, living light scattered beneath.
Ghost Fleet already places wrecks by recorded depth; deep-sea life bins the same axis. One descent past both the lost ships and the things that live beside them.
World Bank Gini over time against NYC rent burden. The macro inequality strand and the lived rent line as one figure widening together.