UFOData sources and the choices behind them
Nothing the tool draws is invented. Each value comes from a named source, read at a stated moment, and where no record exists the interface says so instead of filling the gap.
The principles#
- A source is read only when it is needed. Catalogues are built once into the repository; services are called only for a recording that needs them (a place, a date in their span, a button pressed), and never for a date before their record begins.
- What may disappear is hosted here. Data that a third party could take down or that cannot be read from a browser (orbital elements, 3D models) is copied, reduced, and served from ufoathome.org with its credit, so any page embedding a scene gets it.
- Measured beats modelled, and modelled beats assumed. A brightness is a published observation where one exists; a formula fills in only what nobody measured, and an object with neither is computed but not drawn.
- Every service is swappable. Weather, elevation, imagery, place, time zone, models and drafting go through one registry: the source is a visible choice, with its credit next to it, and a second one can be added without touching the rest.
- The limits are stated. Each source below says where it stops: before 1940 for the weather, before 1957 for satellites, before February 2021 for individual passes.
The sky · The air · Satellites · Aircraft and their trails · Weather · The ground · Place and time · Decor, pictures and sound · Drafting and assessment
The sky
| Source | What it provides | When it is read | Hosting and credit | Choices and limits |
|---|---|---|---|---|
| astronomy-engine | Positions and magnitudes of the Sun, Moon and planets; the Moon's phase; sky coordinates; and, for an eclipse, how much of the Sun the Moon hides from this place at this instant. | In the page, for every recording with a date. | Bundled library, MIT. | A full ephemeris rather than approximate formulas: the reason to reconstruct a sky is to be right about it. Its precession also carries the year-2000 catalogues (stars, the Galaxy, meteor radiants, novae) to the date observed. |
| HYG Database | Star positions, magnitudes and colours; names of stars down to magnitude 3. | Built once (build:stars). The page loads stars to magnitude 7.5, and to 9 only for an instrument that reaches that deep. | Served with the scene bundle. CC BY-SA. | Cut at 9 because the catalogue itself thins out beyond: star counts stop growing as they should. |
| JPL Horizons | The orbit of each naked-eye comet since 1910, as it was at that apparition's perihelion. | Built once (build:comets); the page propagates the orbit itself, no request. | Generated into the code. NASA/JPL. | Brightness is NOT taken from JPL: its magnitude parameters put NEOWISE 4 magnitudes too faint. Peak magnitudes observed at the time are entered by hand, with their dates. A tail is drawn only where its length was recorded. |
| AAVSO, via Strope, Schaefer & Henden 2010 | The measured light curve of every nova since 1891 that reached magnitude 5, binned from the AAVSO's observations. | Built once (build:novae) with the supernovae below; the page interpolates the curve, no request. | Generated into the code. VizieR catalogue J/AJ/140/34. | A recorded curve rather than a decline formula: the standard laws miss most real novae, and would draw DQ Herculis bright for a year through the weeks dust had dimmed it ten magnitudes. Nothing is drawn before the first observation or after the last. RS Oph 2021 and T CrB 1866 reuse the curve of their own previous eruption, shifted onto their recorded peak. |
| Historical supernovae | The supernovae of 1006, 1054, 1181, 1572, 1604 and 1987A: where they stood, and how bright on each night. | Built once (build:novae). | Generated into the code. Positions from SIMBAD; 1572 and 1604 reduced by Ruiz-Lapuente (2004, 2017) from Tycho's, Kepler's and the Korean astronomers' own estimates; 1006, 1054 and 1181 from the chronicles (Stephenson & Green); 1987A from the Open Supernova Catalog. | Each curve says which kind it is, because they differ by two orders of magnitude in quality: 1054 and 1181 are a peak and the day they were last seen, and the decline of 1006 is borrowed from 1572, the same kind of explosion. Dates in the Julian calendar are read as such. V1369 Cen (2013), whose AAVSO record is not reachable by a script and has no published table, is read out of the vector markers of a published figure (Izzo 2017, arXiv:1704.07214), within the part that figure shows. |
| IMO meteor shower list | Radiant, activity window, peak hourly rate and speed of the annual showers; a sporadic background. | Table in the code, computed for any date. | In the code. International Meteor Organization. | The hourly rate is a multi-year average, not a forecast for one night: outbursts and storms are not in it. |
| Global Meteor Network, trajectories | The fireballs its cameras triangulated since December 2018: where each began and ended, at what height, for how long, and how bright at its peak. | Reduced once (build:fireballs). A scene fetches the month of its recording only, and nothing for an earlier date. | Served from ufoathome.org/fireballs/, credited in the player's credits. CC BY 4.0. | Of three and a half million meteors, only absolute magnitude -3 and brighter are kept: the fainter are the sporadic background the scene computes. Brighter than -9 is dropped: past -8 the counts stop falling as meteors do and level into a tail reaching -45, the photometry of a saturated camera. One event solved twice is kept once. The brightness drawn is the peak all along the path, eased in and out, since the record gives the peak and not the light curve. FRIPON's detections are archived too but not used: its site asks to be contacted for any use. |
| Milky Way, zodiacal light, sky brightness | The glow of the Galaxy and of the zodiacal dust, and the brightness of the sky by twilight and moonlight. | Computed in the page, no data fetched. | Models from the literature: Leinert et al. 1998 and Hong 1985 (zodiacal light), Patat et al. 2006 and Krisciunas & Schaefer 1991 (sky brightness against which the glows are judged; the light of the sky itself is under the air). | No photograph of the Milky Way: a model Galaxy integrated along the line of sight, checked against three measurements rather than fitted to one. |
| World Atlas of the artificial night sky brightness (Falchi et al. 2016) | How bright the night sky of a place is with its towns, the zenith of a moonless night in magnitudes per square arcsecond; what turns a clear sky into a polluted one. | Cut once into small tiles (build:light-pollution). The editor fetches one tile, a few kilobytes, when a place has no stated value; the player reads the value the recording carries. | Tiles served beside the page, then from ufoathome.org, with the credit kept in their index. The atlas is CC BY-NC 4.0, a non-commercial licence: its use on a public site is still to be settled. | A measurement of 2015 stands for any date: the atlas cannot say how a town lit up or dimmed since. It is only what the editor proposes: a value the witness or the investigator states (a Sky Quality Meter reading) wins, and a recording with none is played under a natural sky. |
| NASA LOLA shape map of the Moon (LDEM_16) | The relief of the Moon's edge, so that the limb is not a perfect circle: the mountains and craters that let the Sun's last light through as Baily's beads. | Cut once (build:lunar-limb) into a file loaded with the scene, the first time an eclipse comes near totality. | Served with the scene bundle, about 2 MB. NASA, Planetary Data System. | 16 samples a degree, and only the band the limb can ever fall in (libration and the observer's parallax included), kept in units of 100 m: enough for the beads, a fraction of the 33 MB map. Which eclipse there is, and how much of the Sun is covered, is never listed: it is computed from the real positions of the two bodies. |
The air
The light of the sky is computed, not drawn: sunlight and moonlight traced through a spherical atmosphere of air, haze and ozone at fifteen wavelengths, then shown as an adapted eye or a camera would. Every input below is a published measurement or model; the two display anchors are the only choices.
| Source | What it provides | When it is read | Hosting and credit | Choices and limits |
|---|---|---|---|---|
| Hillaire 2020; Bruneton & Neyret 2008 | The method: single scattering along each line of sight, and every higher order folded into a table of the light arriving at each height for each height of the Sun. | Tables built on the graphics card once per page and per haze level (about a second of GPU, shared by every scene of the page); the sky view redrawn when the Sun, the Moon or the observer move. | Implemented in the code, twice: a reference in TypeScript and its GPU twin. | Chosen for being fast enough for a web page. Checked against a Monte Carlo trace of the same atmosphere: within 0.25 magnitude and 0.012 in colour from a hazy noon to a Sun at -16°, except the Earth's shadow in nautical twilight, 0.7 magnitude too dark. The table resolution was measured, not picked: coarser tables put a -10° zenith more than half a magnitude off. |
| Bodhaine et al. 1999 | How strongly air scatters light at each wavelength. | Computed in the page. | Formula from the paper. | Derived from the refractive index of air, measured to many figures: nothing to adjust. |
| Serdyuchenko-Gorshelev ozone cross-sections | Ozone's absorption across the visible (the Chappuis band), which makes a twilight zenith blue. | Averaged once into the code. | IUP Bremen, Gorshelev et al. 2014 and Serdyuchenko et al. 2014 (Atmos. Meas. Tech. 7). | Taken at 223 K, the stratosphere's temperature; the band changes by 3 % at most between 193 and 293 K. Without it, a zenith six degrees after sunset comes out nearly white (Hulburt, 1953), and a test checks that it does. |
| ASTM E 490-00a | Sunlight above the atmosphere, wavelength by wavelength. | Averaged once into the code. | The standard's table 3. | Averaged over 20 nm, which removes the Fraunhofer lines: right for a sky, wrong for a spectrograph. It adds up to the Sun's measured illuminance, 128 to 133 thousand lux, which a test checks. The Moon's sky uses the same spectrum, scaled by the Moon's magnitude. |
| Kasten 1969; Hänel 1976 | How milky the clear sky is: the haze's optical depth, from the relative humidity of the weather record. | With the weather: the humidity comes from ERA5's temperature and dew point (Magnus formula, Alduchov & Eskridge 1996). | Growth law from the literature. | Haze particles swell with water as the air nears saturation; a typical continental burden swells as (1 - humidity) to the power -0.3, giving the usual 0.1 at 50 %. How many particles there were (a town, a fire, the wind's history) is in no record, and is not guessed. Capped at 95 %: beyond, it is mist. |
| Patat et al. 2006 (Paranal) | The measured brightness of the zenith through twilight, the check a real sky gives the model. | In the tests, and in the page for whether the Milky Way and the zodiacal light could be seen. | The paper's V-band fit, table 1. | Never fitted to: the model meets it within half a magnitude from -6° to -16°. Reading the published fit also corrected a table in the tool that had civil twilight ten times too dark. |
| Krawczyk et al. 2005; Thompson et al. 2002 | How an eye adapted to that sky shows it: a compressive response that follows the sky's own brightness, and the shift to a dim blue-grey as the rods take over. | In the page, for every pixel of the sky. | Blend of the rods from Krawczyk, Myszkowski & Seidel; night blue from Thompson, Shirley & Ferwerda, taken halfway toward white. | The only choices in the chain, stated as such: a clear day zenith and a moonless night zenith are pinned where the tool showed them before, and everything between follows. A camera (a recording medium) gets no rods, and its gain over an eye brightens what it records. |
Satellites
Two levels. For every date since Sputnik, what the Earth's shadow allowed and how many objects were in orbit. From February 2021, which satellites really crossed the sky, where, and how bright.
| Source | What it provides | When it is read | Hosting and credit | Choices and limits |
|---|---|---|---|---|
| CelesTrak SATCAT | Launch and re-entry date of every tracked object: the number in orbit each month, and when each class existed (Echo, Iridium flares, ISS, Starlink). | Built once (build:satellites); also gives launch dates to the orbital element archive. | Generated into the code. CC BY 4.0. | Only the two date columns are read here (the re-entry list below also reads the inclination). Its orbit columns describe each object's LAST state: Echo 1 appears at 400 km when it flew near 1,500. Debris is not counted. |
| CORDS, The Aerospace Corporation | Every re-entry it followed since 2000: its predicted time with the uncertainty stated for it, and for about a hundred the time it was seen coming down and where from. | Reduced once (build:reentries), with the SATCAT's decay days for every year since 1957; the editor fetches the year of the recording only. | Served from ufoathome.org/reentries/. | Archived by hand, because the site cannot be read by a script. A re-entry is stated with its precision and nothing finer: a sighting on record within an hour and 1,500 km, a prediction whose window meets the observation, or a bare decay day. Only the first says it could have crossed this sky. An orbit whose inclination keeps it out of sight of the observer's latitude is never listed: drag lowers an orbit but leaves its plane, which is why this one SATCAT orbit column is read. |
| Orbital element archive, Laurent Chabin (SCEAU) | The public element sets saved once or twice a day since 22 February 2021: the naked-eye list, every Starlink, and to January 2025 the full catalogue. | Reduced once (build:tle). The page fetches only the two weeks around the observation, and only for a date the archive covers; the propagator itself is loaded only then. | Served from ufoathome.org/tle/, credited in the player's credits. | Hosted here because the original server cannot be read from a browser. 2.8 GB of snapshots are kept as 113 MB: every day for naked-eye objects and for Starlinks in their first 60 days (the trains), one set a week otherwise. Debris and objects nobody measured are dropped. A set is never used more than 7 days from its epoch, so a hole in the archive (three months in late 2025) is reported as a hole rather than filled with a stale orbit. |
| satellite.js (SGP4) | The standard propagator for these element sets, and the Sun's position; the Earth's umbra and penumbra are computed on top. | Loaded on demand, for covered dates only. | Bundled library, MIT. | Held on the release before the one that ships a WebAssembly build that breaks the bundle, for no gain here. |
| Satellite brightness: McCants, Stellarium, Mallama et al. | How bright each object looks: standard magnitudes (McCants, completed by Stellarium's list for recent objects), and published means for the constellations nobody catalogued (Starlink, OneWeb, BlueBird). | Merged into the archive at build time; applied in the page. | Starlink 5.93 then 7.21 (arXiv:2006.08422, 2101.00374), 4.58 below 357 km while raising orbit (2405.12007); OneWeb 7.18 (2012.05100); BlueBird 3.77 then 4.32 (2608.23668). | Without the orbit-raising value no Starlink train would ever be drawn. An unmeasured rocket stage takes the median of at least three identical measured stages; any other unmeasured object is computed but not drawn. Glints and flares are not modelled: they can only make a satellite brighter. |
Aircraft and their trails
From 2022, the aircraft that were in the observer's sky at the hour of a recording, the air they flew in, and the trail their engines left there. What is drawn is a compatible candidate, never an identification, and an empty sky in the record excludes nothing.
| Source | What it provides | When it is read | Hosting and credit | Choices and limits |
|---|---|---|---|---|
| adsb.lol open history | Where every aircraft that broadcast its position was, from 2022: its track, and its type, registration and category. Each is drawn with its real size, lamps (from dusk, by kind of aircraft), the Sun that still lights it after the ground has lost it, the haze between, and the sound it makes if it can be heard. | Reduced once per day (build:aircraft) into one packed file per hour and per 1-degree tile. The page reads only the tiles within 150 km of the observer, for the hour of the recording and the five minutes before it, and never for a date before 2022. The code that draws, labels and plays the aircraft is loaded only when a scene has some. | Built from the daily releases of the adsb.lol history (ODbL 1.0), credited in the player's credits. Not hosted yet: until the archive is served from ufoathome.org/aircraft/, aircraft show only from a local copy. | One feeder instance is enough: prod-0, with staging-0 as a fallback, hold the same flights to 0.02 %. Only the positions of a release's own UTC day are kept, so the result does not depend on the order days were read in; airport vehicles are dropped. What no receiver heard is absent: light aircraft, gliders, balloons and military aircraft that switch their transponder off are under-represented, and about one aircraft in seven states no type, so it is drawn as a generic airliner. |
| amvlab aircraft models (A320, 737, A350, A380, 787) | The airframe of an Airbus A320, A350 or A380, or of a Boeing 737 or 787, for an aircraft of those types that is near enough for its shape to be seen: engines, wing tips, tail. | Fetched by the page only for an aircraft of those types within about 130 times its own length of the observer (5 km for an A320, 10 km for an A380), once for the scene (170 to 250 KB each). Further off an aircraft is a few pixels, and the built-in shape and its lamps are drawn. | Re-hosted on ufoathome.org/models/ with their licence, CC BY 4.0, © 2026 amvlab, and credited in the player's credits. | The plain variants, without a livery, since the airline is not known. One model for the whole A320 family, one for every 737, one for the A350 (-900 and -1000), one for the A380 and one for the 787 (-8, -9 and -10), each scaled to the real length of the type (an A321 is the A320 model drawn longer). Which variant each depicts, and how faithfully, is not documented by their author. Every other type keeps the built-in shape: a model is a catalogue entry and a row of a table, not code. |
| Open-Meteo, historical forecast API | Temperature, humidity and wind at 500, 400, 300, 250, 200 and 150 hPa, hour by hour, from 2022: the air the aircraft fly in. | Fetched by the page, once per recording that has aircraft, for the observer's place and the hours of the window: never for a date before 2022, and never for a scene with no aircraft. | Open-Meteo.com, CC BY 4.0, no key, credited in the player's credits. | Not the reanalysis the weather uses: its archive serves no pressure levels. These are the forecasting models' own analyses, which are known to underestimate how often the air at cruise level is supersaturated over ice, so a trail that lasts in the sky may be one the record says should not. Read at the observer's place only: an aircraft a hundred kilometres off flies in air the model may state differently. The humidity is relative to water (checked against the ceiling the data make at saturation over ice, near 60 % at -47 C). |
| Schmidt-Appleman criterion (Schumann 1996) | Whether an engine's exhaust leaves a trail in that air, and whether it lasts. The exhaust cools along a line fixed by the engine's efficiency and the pressure; a trail forms if that line crosses saturation over water, and it lasts only where the air is saturated over ice. | Computed in the page for each position of each aircraft, at the pressure its barometric altitude is (a barometric altitude is a pressure), and carried by the wind at that level. | Formula and constants in the code, with the saturation pressures of Murphy and Koop (2005). | The engine is a typical one for the kind of aircraft, not the one of that airframe: overall efficiency 0.38 for a wide-body, 0.34 for a narrow-body, 0.30 for a business jet, 0.22 for a fighter. Propellers, rotors, gliders and balloons are given no trail. How a trail ages is a parameterisation of observed orders of magnitude, not a simulation: its width grows with the wind shear between the two levels either side of the aircraft (how much the wind changes with height) and a fixed turbulent diffusion, its depth is assumed, and it is drawn as wide across the line of sight whichever way the shear points; how it fades and how long a short one lasts are assumed too. A trail drawn is estimated, never a fact; and one left by an aircraft that had gone before the five minutes the record is asked for is missing. |
Weather
| Source | What it provides | When it is read | Hosting and credit | Choices and limits |
|---|---|---|---|---|
| Open-Meteo: ERA5 (Copernicus/ECMWF), then its forecast model for the last days | Hourly cloud cover by level, precipitation, snow, weather code, wind, temperature and dew point, at the place and hour of the observation. | Fetched by the editor once a date and a place are known, from 1940 only. For the last five days, which ERA5 has not caught up with, the forecast model's own analysis is read instead (the file names which one answered). | Open-Meteo archive and forecast APIs, keyless. © Copernicus/ECMWF. | A reanalysis on a ~28 km grid, not a station: a local shower can be missing. Cloud bases are derived (temperature and dew point for the low deck), not measured. Every field can be unlocked and replaced by the account. No vertical profile is available in archive mode, which is what radar propagation anomalies would need. |
The ground
| Source | What it provides | When it is read | Hosting and credit | Choices and limits |
|---|---|---|---|---|
| AWS Terrain Tiles (SRTM, ETOPO1) | Relief around the observer. | Fetched when a scene has a real place. | Amazon S3, keyless. © AWS Open Data. | No key and no rate limit, which an embeddable component needs. The relief model is too coarse for trees and buildings: a picture of the place lined up in the scene is the way to state those. |
| Esri World Imagery | Aerial imagery draped on the relief, and the observer map. | Fetched with the relief, and when the map is opened. | Esri tiles, keyless. © Esri, Maxar, Earthstar Geographics. | Today's photograph of the ground, whatever the year of the account; the scene darkens it to the hour. EOX Sentinel-2 cloudless can be picked instead. |
Place and time
| Source | What it provides | When it is read | Hosting and credit | Choices and limits |
|---|---|---|---|---|
| Nominatim (OpenStreetMap) | Searching a place by name, and naming a point. | Only when the author presses Enter or the search button, never at each keystroke. | © OpenStreetMap contributors. | It knows hamlets, farms, roads and airfields, where accounts happen; its usage policy is why nothing is sent while typing. |
| Open-Meteo time zones and the IANA database | The time zone of a place, then its legal offset on the date of the account. | Once a place is known; the historical offset comes from the browser. | © Open-Meteo; tz database in the browser. | Today's offset is never used for an old date: France was at UTC+1 all year in 1965. |
Decor, pictures and sound
| Source | What it provides | When it is read | Hosting and credit | Choices and limits |
|---|---|---|---|---|
| IGN LiDAR HD (canopy height model and classification) and BD TOPO (map vectors) | The trees, buildings, bridges and overhead lines round a place, at their measured place, height and width: from the laser, the height of everything that stands above the ground and what each point was taken to be (plant, roof), as pictures of a hundred-metre square; from the map (BD TOPO), the centre line of each bridge, the width of the road it carries, and each electric line with its voltage; the names of what it names (a town hall, a post office, the road a bridge carries). In France the roads of the scene come from the same map, with their names, instead of from OpenStreetMap, with a pavement along each side of its urban streets (the map says which are urban). | By itself, from the survey the author chose with the selector beside the list (the laser and the map together by default, either alone, or none), when the author gives the recording a place the survey covers (France, a box that the service itself narrows by answering nothing where it has not flown), a moment after the last change of the place; never when a recording is merely loaded. | © IGN, Licence Ouverte 2.0, served by the Géoplateforme's WMS, which answers browsers from anywhere. The survey date is not given by the service. The bridges and lines come from the same Géoplateforme's WFS; the pylons stand at the vertices of the line as the map draws them, 22 m tall for 90 kV where the laser did not measure them. | The height model alone reads a roof as readily as an oak (291 m at the Eiffel Tower), so the classification is asked for too and only its plants and roofs are kept. It is a scan of one day, and a recording is of another: what it brings is marked as coming from a survey, replaced when the place changes, and left alone where the author has placed trees or buildings. |
| IGN aerial photograph (ORTHOIMAGERY.ORTHOPHOTOS) | The colour of each building's roof, seen from above: the median of the photograph under its outline, without the greens, lifted a little because the photograph is already lit by the sun. | With the scenery of a survey (see above), in the same request: a square of at most a thousand pixels. | © IGN, Licence Ouverte 2.0, from the Géoplateforme's WMS. The photograph is of the day it was flown, not of the recording. | The shape of a roof (flat, two slopes, four slopes, which way the ridge runs) is read from the laser's heights inside the outline, within 150 m only; a roof that cannot be read stays a box rather than a guess. |
| INSEE, Enquête Emploi du temps 2009-2010 (via Rauch, Économie et statistique 522-523, 2021) | When the people of a town are up, for how many of the windows of its buildings are lit by the hour on the clock: half asleep at 22:40, half up at 7:00, nine in ten asleep from 1:00 to 6:00. | Built into the lit windows of the scene; nothing is fetched. | Published by INSEE; only the three figures above are used. | The survey gives medians and a bound, not a curve: the spread round the medians, the two in a hundred always up, and the half of the windows lit when everybody is up are assumptions, and France's time use is taken for any place. |
| OpenStreetMap buildings (Overpass) | The buildings round a place anywhere in the world, as OpenStreetMap's volunteers drew them: each outline as a rectangle (or rectangles), its height where it states one (height, building:levels), its roof (roof:shape, roof:height, roof:colour), the colour and material of its walls, its name. No trees. | By itself, for a place outside the reach of the IGN's surveys when no survey was chosen (or when the IGN's laser turns out not to have flown there, as over Belgium), or when the author chooses it from the list of surveys. | © OpenStreetMap contributors (ODbL), through the public Overpass instances, one request at a time like the roads. They are busy: a town's buildings can take a minute or more, and the spinner over the picture can be stopped by choosing "None". | Most buildings carry only building=yes: they are drawn with the ordinary measures (two storeys, a flat roof, a pale plaster). Only what is written is drawn as written. The coverage is as good as the volunteers': a Belgian town counts 4,600 buildings in a square kilometre and a half, a village of the American West a handful. |
| OpenStreetMap street lamps (Overpass) | The public lamps round a place: each lamp mapped (highway=street_lamp), and a lamp every 35 m, on alternate sides, along each street tagged lit=yes. They are switched on by the dark (the sun below the horizon), and the nearest eight light the scene. | With the scenery of a survey (see above), in the same moment. | © OpenStreetMap contributors (ODbL), through the public Overpass instances, one request at a time like the roads. | Volunteers tag a street as lit far more than they place its lamps (three lamps and eight lit streets in the centre of Saint-Lô), so the lamps along lit streets are drawn, not surveyed. A street not tagged is not necessarily dark. Many French towns switch their lighting off at night: a recording of such a night should have its lamps removed or switched off. |
| 3D models (Kenney, Jagobo, Jungle Jim, Nicholas-3D, amvlab, Poly by Google, MakeHuman) | Cars, a house, trees, a street light, airliners (A320, 737, A350, A380, 787), people; and the craft and beings of the cases, built by the project from their accounts. | When a decor object or a body names a model; the catalogue itself is read once. | Re-hosted on ufoathome.org, each with its licence file: CC0 (Kenney, MakeHuman and the project's own), CC BY 4.0 (the trees, the airliners of amvlab), CC BY 3.0 (Poly by Google). The catalogue is read from /models/index.json beside the page first, then from ufoathome.org, so a site can serve its own copy. Each credit is shown in the player's credits. | Hosted here because free kits sit on mirrors of uncertain permanence. A model whose credit is unknown is not drawn; Google Earth models were refused for their licence, and the stylised broadleaf tree for not being realistic enough. People are built from the MakeHuman system assets alone, whose licence is CC0: community packs are not used, their licences differ. A craft or a being is made from the documents of its case by a script that says which of its details are assumed, and any better model replaces it by taking its id. Every model is a placeholder of the right kind, not of a particular make or year, and the picker says so. A model publishes its own settings (a costume, the leaves, a hull, retracted landing gear, which lamps and which noise suit it), which the editor offers and a recording states in modelParameters. Whenever a model cannot be had, and when the observer stands inside the object, the built-in shape is drawn instead. |
| Panoramax | Street-level photographs and panoramas near the place, each with its position and direction. | Only when the author searches for pictures in the editor. | Panoramax API, credited with each picture's author and licence. | Open and readable from any page. Google Street View was refused: its terms keep the pictures inside Google's viewers. |
| Weather sounds (OpenGameArt) | Rain, wind and thunder. | Loaded when the weather of the scene needs them. | Bundled. Rain and wind CC0; thunder by Jerimee, CC BY 3.0, credited in the player. | The phenomenon's own sound is synthesized from its description, not taken from a recording. |
Drafting and assessment
| Source | What it provides | When it is read | Hosting and credit | Choices and limits |
|---|---|---|---|---|
| Claude (Anthropic) | A first draft of a recording from a written account, each value marked as stated, derived or assumed. | Only when the author opens the panel and gives their own API key. | Called from the browser with the author's key; nothing passes through ufoathome.org. | No server of ours, so no account and no key held here. The draft is a proposal the author edits, never a result. |
| Hynek classification | The close-encounter category a recording supports, beside the tool's own coverage questions. | Computed in the page. | After J. Allen Hynek. | Categories the format cannot state (a physical trace, a radar return) are marked unsupported rather than guessed; tags are not taken as evidence. |
The detail of each choice, with the measurements that settled it, is in the code comments of each source in the repository, and the README says how to rebuild each catalogue.
What a looked-up value becomes in the file (weatherSource, the record-low, record-mid and record-high layers) is described on the sighting file's page.