SolarSystemRegistry.org — Documentation

⚠ In development — prototype · Project management and author: Adrien Normier (with grateful acknowledgment to Jonathan Justman and V. Nakache for help in the project's earliest steps) · Ontological development grounded in the works of the Cosmic Footprint Society (2024 ISSI Forum on Cosmic Footprint) · The databases belong to their owners · Licence: SSR-GPL v2.0 · CFS public-interest grant

Disclaimers. This is a non-normative effort: we do not claim that the effects shown are good or bad. We do not hold responsibility for the accuracy of the aggregated databases; despite our best efforts, some errors may remain. Not for use for celestial navigation.

What this is

A 4D visualization and data registry rendering humanity's presence in space within its natural setting — one continuous interactive scene spanning more than twenty orders of magnitude in space and the full depth of cosmic time. Its purpose is to provide contextual awareness of what humanity does, by placing it in the setting where it happens: the natural cosmos, rendered at the same scale, in the same frame, at the same time.

Every record carries two orthogonal tags: provenance (how the value came to be: RAW measured · PROCESSED derived · MODELLED · SYNTHETIC · VISUAL) and origin (NATURAL · ARTIFICIAL · MIXED). Estimated content always reads as estimated: synthetic and fitted renditions render dimmed/translucent, and every factory-estimated parameter is listed per object, with its method.

Classifying the footprint — a tentative working scheme

As an experiment, and with no claim to authority, every human-made object is given a provisional classification along two axes. We follow the ordinary causal-chain way of thinking used in environmental assessment on Earth (loosely: a source, a pathway, a receiver), applied here as a convenience for sorting and filtering — not as a settled or validated taxonomy. We may well have some of it wrong; corrections are welcome.

Containment class — roughly, how far a footprint can reach: mission (reach bounded to an object or body) · biotope (reach into one planetary environment, where planetary protection is the concern) · uncontained (reach outward as light/radio, where little governance exists). We offer this only to make the very uneven state of oversight easier to see, not to assert any conclusion about it. Filter with containment:mission, containment:biotope, containment:uncontained.
Vector — the carrier the object is: engineered-object · fragment (debris, spent stages) · effluent · biological · electromagnetic (signals) · energetic (nuclear) · kinetic (impacts). Filter with vector:fragment, etc.
Originator class — very roughly, who operated the object: governmental · commercial · scientific · military. This is a coarse join from each object's operating organisation as recorded in GCAT's orgs table (its own Class code), so it inherits that table's gaps and simplifications — an object whose operator is unlisted is simply left unclassified rather than guessed. Filter with originator:commercial, etc.

The tag is a derived, provisional heuristic, computed from each object's kind, origin, host body and emission — nothing is invented, and an object with insufficient data is left unclassified rather than guessed. It is deliberately non-normative: it tries to record a physical fact, never to judge whether a footprint is good or bad. Treat every label as a first attempt, open to revision. A further axis (a maturity flag) may be added later, and only where the underlying data is solid enough to support it.

Methodologically this borrows, humbly, from the standard environmental-assessment chain (Source–Pathway–Receptor, Holdgate 1979; the DPSIR indicator family, OECD/EEA; Elliott et al. 2017). The mapping to space objects is our own tentative interpretation and carries no external endorsement.

Meeting the footprint one category at a time

The registry sorts the human footprint into three Cosmic Footprint categories — material · electromagnetic · biological — and the opening view uses that sorting to decide what a first-time visitor meets. It shows the material footprint on its own: spacecraft, upper stages, components, exhaust and the marks a physical arrival leaves, the things one already knows how to recognise.

This is done with an ordinary badge, not a special control. Between 2026-07-30 and that same day the categories briefly had three dedicated buttons; they were removed as redundant, because the badge bar already is the one place where the scene is filtered. The landing view therefore carries a hide badge on the electromagnetic footprint, and that badge behaves like every other: it is visible in the badge bar with the count it is holding back, it is removed with its ✕, and it travels in a shared link through fx=. One control layer, learnable once.

The badge is a query, not a hand-written list. It selects on the footprint vector of the classification above (the carrier the object is), so a newly ingested catalogue is covered without anyone editing a list. The electromagnetic category is electromagnetic · energetic; material is engineered-object · fragment · effluent · physical · kinetic; biological is biological. The eight vectors are split across the three categories exactly once each, with none left over.

Two consequences worth stating plainly. A nuclear detonation is carried by the energetic vector — which our classification assigns for its electromagnetic pulse — so detonations are hidden by the landing view’s badge until it is removed. And the badge selects on origin:artificial: natural objects, and natural emission such as a pulsar’s beam, are the setting rather than a footprint, and are never hidden by it.

Nothing behind a badge is deleted. It stays in the registry, keeps its record, stays findable by name in search, and travels in a shared link.

Uncertainty & confidence — how error bars are tracked

Epistemic transparency (a guiding principle) means a value is only as good as its uncertainty, so every measured or derived parameter can carry its 1-sigma error bar. The method is deliberately simple and generic, following standard science-display practice:

Model. Each record holds an optional sigma side-map — 1-sigma absolute uncertainties keyed by the field they qualify (orbital.a, tagged.radius, …), in the value's own units. A side-map (rather than wrapping every value in a {value, sigma} pair) keeps the rendering hot path untouched: the error data is read only where confidence is shown.
Honesty (A5). A sigma is present only where a source measured it or the factory propagated it. Absent means unknown — never silently zero, never invented. So "± 20 m" is a real measurement and a missing bar is an honest "we don't know".
Display. The record card shows the value with its bar and, when large enough to matter, the relative size — e.g. a near-Earth asteroid's radius 170 m ± 20 m (12%) next to its orbital elements known to one part in a billion. Values are shown to a precision matched to their uncertainty.
The precision gate (I15). The relative uncertainty (sigma/value) is a first-class, queryable quantity; a 1% gate decides when a value is "precise enough" to rely on (it already governs which stars enter the catalogue).
Propagation. When the factory fits or estimates a value, it carries the uncertainty through by first-order propagation (sigma_f = root-sum-square of each input's contribution), so an estimate states its own confidence — and drops a term it cannot quantify rather than guessing.
Cost. Tracking is computed at ingest/bake and read only on demand (the inspected object), so it adds zero per-frame cost; it scales to every object in the registry. Heavier visual error bars (a position-error ellipsoid, a banded orbit) render only for the one selected object.

Today the near-Earth asteroid set carries full orbital + size error bars from JPL, the exoplanet layer carries the archive's published error columns (over 10,400 of the ~10,900 exoplanet records hold a 1-sigma bar on temperature, radius and/or mass), and the bright-star layers (naked-eye Gaia, Hipparcos, Local Volume galaxies) carry their per-star or per-galaxy distance uncertainty. More sources and the visual error bars follow on the same spine. Find carriers with the has:uncertainty search.

A standing provenance & uncertainty coverage audit (provenance-audit.json, regenerated by scripts/audit-provenance.ts) measures the baseline — audit snapshot of 2026-06-15 (130,479 nodes; the registry has since grown — a re-run is pending): every parameter's source is documented (100% of nodes, via the per-layer attribution), while the per-field error-bar coverage is reported honestly — the orbital + radius σ on the JPL asteroid set, the exoplanet error columns, the star-layer distance σ — with the remaining factory-propagation coverage stated as incomplete rather than assumed. A live per-quantity coverage table (value ⊇ σ ⊇ precision-gate, with the UNKNOWN remainder kept distinct) is available in the application (measurement-coverage panel, permalink cv=1). The same file carries the full per-level parameter inventory.

Guiding principles

No privileged center — Sol is a node like any other · time is a dimension (scrub 1940–2060, or far beyond) · continuity, no visual jumps · fluidity over detail · epistemic transparency · the natural setting contextualises the artificial — same scale, same frame, same time · zero server cost (fully static hosting) · extensibility without modification.

How it is built — the five blocks

Who does what, conceptually. Ingest gets the world's catalogs and normalizes them — it may reshape, never invent. The registry is the arbiter: it reconciles identities across sources (the same spacecraft in three catalogs becomes one object), layers trajectories by quality, validates the result, and bakes ONE deterministic document. The viz turns that document into pixels and gestures — it runs no new physics and asks no external service. The factory is the single licensed maker of values nobody measured: explicit models from published literature (every one listed in the library), deterministic under a seed, and every output labeled estimated where you read it — its functions are inventoried on the factory page. The archive remembers everything append-only, so any past bake can be audited or reproduced. The full engineering specification (coordinate conventions, frame-tree invariants, inter-block contracts) is access-controlled.

Data sources, citations & licenses

Citation policy. Every scientific reference the registry leans on hyperlinks to its DOI (or, where none exists, its authoritative stable archive URL) — a reader reaches the source in one click. The complete, DOI-linked bibliography is the library, organised by class of natural element and by class of anthropogenic effect; each entry notes which registry element it informs. New sources are added there in the same change that introduces them.

Currency. Living catalogs are re-fetched on a scheduled maintainer-side cadence tuned per source — weekly for the fast movers (GP element sets, the SATCAT, upcoming launch windows), monthly or quarterly for the slower catalogs (never by your browser — the site stays a static, deterministic bake): each refresh lands as a dated, append-only snapshot in a review proposal that headlines any rows removed upstream, and only the maintainer's review merges it — previous snapshots are never deleted. Deletion retention. For the catalog sources (GCAT object catalogs · CelesTrak SATCAT · UNOOSA index · exoplanets), a row deleted upstream does not silently vanish from the next bake: the bake unions prior snapshots' rows back in and labels each such record upstreamRemoved:<last-seen-date> (for a name-keyed catalog the label honestly reads "removed or renamed" — the data cannot distinguish). Point-in-time feeds (GP element sets, DSN fixes, DONKI event windows, rolling upcoming-launch lists) are deliberately excluded: absence from the next fetch is their nature, not a deletion.

1) Natural setting

The cosmos as it is without us — the stage every artificial object is measured against.

Layer	Content	Authority / citation	License / terms
Solar system	Sol, 8 planets, 7 dwarfs, 22 moons; IAU spin states; moon orbital planes/phases are real J2000 osculating elements. Thirteen more bodies now point where their kernel says (2026-08-30): Io, Europa, Ganymede, Callisto, Mimas, Enceladus, Tethys, Dione, Rhea, Titan, Iapetus, Vesta and Ceres carry their published pole, prime meridian and rate — every body the registry holds a real map or figure for and previously could not orient. Eleven are tidally locked, which supplies an anchor outside our own arithmetic: the kernel's rotation rate must reproduce the body's independently published orbital mean motion, and all eleven agree to better than 0.14 % (Vesta and Ceres, which are not locked, miss the same test by more than a hundredfold — so it measures the lock, not the algebra). Frozen at J2000; the excursion that costs is periodic, not secular, worst at 3.06° (Mimas) over 1950–2100. The Moon is tidally locked, and now says so in the geometry: it carries the IAU/NAIF Mean Earth/Polar Axis rotation (W = 38.3213° + 13.17635815°/day) and orbits at the measured sidereal month, so its near side faces Earth at every epoch — the sub-Earth longitude holds inside the ±8° optical-libration band from 1900 to 2150, where before it swept the whole ±180°. That frame is the one every lunar coordinate here is published in, so the correction rotates the 9,086 lunar Gazetteer features, the 20 impact sites, the 694 Apollo surface items and the 101 landed craft together, rigidly. Declined with measured magnitudes: the 13 nutation terms in W (≤0.053°, because the matching pole terms cancel them), the 18.6-year Cassini precession of the pole (≤3.1°), the Mean-Earth↔principal-axis difference (~0.03°)	Standish (1992) mean elements · JPL Solar System Dynamics · JPL Horizons (moon i/Ω/ω/M₀, J2000) · IAU rotation reports (Archinal et al. 2011/2018) via NASA/NAIF `pck00011.tpc`	Public-domain U.S. government data; NAIF kernels “may be downloaded and used by anyone”; cited per record
Local universe	CMB/Local Group/Milky Way anchors, M31 · M33 · LMC · SMC with observed disc orientations, 11 nearest/brightest named stars — each anchor star carries its MEASURED space velocity (proper motion + radial velocity read from the committed Gaia DR3 / Hipparcos / XHIP snapshots, converted through the one cited solar-motion convention; a star with no admissible RV keeps an honest tangential-only drift, stated per record)	Gaia DR3 / Hipparcos parallaxes & proper motions · XHIP radial velocities (Anderson & Francis 2012, q_RV 'D' excluded) · NED distance compilations · Planck 2018 dipole · RC3 (de Vaucouleurs+ 1991) disc PA+i · IAU galactic frame per the Hipparcos catalogue (ESA SP-1200, Vol 1 §1.5.3)	Public catalogs; per-record citations in each object's notes
IAU star names	451 official proper names (WGSN) reconciled onto the star layers by sky direction with a triple identity gate (6′ position · uniqueness · magnitude agreement), magnitude tie-breaking arbitrating close pairs and Bayer-designation rows renamed to their proper names: 419 of the 451 names attach (Sirius A, Canopus, Vega, Rigel, Aldebaran, Porrima, Alcor, Deneb Algedi…); the remaining 32 stay honestly unmatched or ambiguous rather than forced	IAU Division C Working Group on Star Names — IAU-CSN, snapshot 2026-06-11	IAU — CC BY 4.0
Exoplanets	4,685 host stars + 6,257 planets. Cross-joined to Gaia DR3 by sky position: a host coinciding with a Gaia star (within 0.3 pc) is deduplicated (the Gaia duplicate is dropped, not double-counted), and Gaia's independent effective temperature is reconciled onto the host — kept as a corroborating source when it agrees, a selectable alternate when it disagrees (ERASE NONE)	NASA Exoplanet Archive, `pscomppars` table (IPAC/Caltech, under contract with NASA)	Public; "This research has made use of the NASA Exoplanet Archive" — snapshot 2026-06-10
Hazardous asteroids (PHA)	2,547 Potentially Hazardous Asteroids incl. 99942 Apophis — heliocentric osculating elements at per-row epoch; radius only where a diameter is measured (340). Each object also carries the orbit solution behind those elements, on its record: who determined it (2,511 by JPL's automated pipeline, 32 by J. Giorgini, 3 by D. Farnocchia, 1 by S. Chesley — so 98.6% of these orbits are automated fits, which is worth knowing), how many observations were used and how many were radar, the arc they span with first and last observation dates, the normalised residual RMS, and the Minor Planet Center's uncertainty parameter U (0–9). All 2,547 rows carry all of it, from the same API response as the elements. U is shown with the in-orbit longitude runoff band it denotes, derived from the MPC's own published definition rather than transcribed — 1,751 objects sit at U=0 (under 1″ per decade) but 796 are at U≥1 and 52 at U=9 (≥41° per decade: (1979 XB), for instance, rests on 18 observations over a four-day arc in 1979). The MPC states that "the U value should not be used as a predictor for the uncertainty in the future motion of NEAs" — it classifies the solution, not the future, and the layer card repeats that. Search the text `MPC U=`	NASA/JPL Small-Body Database (SBDB) Query API — snapshot 2026-07-29; MPC U definition: minorplanetcenter.net/iau/info/UValue.html	Public domain (U.S. Government work, NASA/JPL-Caltech) — verified at fetch against ssd.jpl.nasa.gov/about, which asserts no terms and asks only to be cited; the fetcher refuses to save a snapshot if that statement has gone
Comets	1,417 bound periodic comets — 1P/Halley, 2P/Encke, Hale-Bopp and the rest of the well-constrained closed-orbit population (semi-major axis within the outer-solar-system/scattered-disc scale, a ≤ 500 AU) — each on its real heliocentric osculating orbit at the row's own epoch (Halley's retrograde 162° inclination preserved). Hyperbolic, parabolic, and near-parabolic single-apparition Oort-cloud comets (2,651 rows) are EXCLUDED, not forced onto an invented orbit (A5). Radius only where SBDB measures a nucleus diameter (104). Each comet also carries its JPL orbit solution (observations, arc, residual RMS, producer) exactly as the asteroid layer does — all 1,417 of them. Only 1,128 carry an MPC U value, and the gap is deliberate: long-period comets are classified by MPC orbit QUALITY codes (1A, 2B — Marsden et al. 1978) on an entirely different scale, so those cells are left empty and counted rather than mapped onto U. On 70 of the 1,417, JPL declares that it used a low-precision TWO-BODY dynamic model — the same approximation this registry makes, so there the record says so rather than implying more: 34D/Gale, for one, rests on 47 observations whose last is known only to the year, 1938. Where a source publishes a date to the year alone it is kept that way, never completed to a day. Search `kind:comet`	NASA/JPL Small-Body Database (SBDB) Query API, `sb-kind=c` — snapshot 2026-07-29	Public domain (U.S. Government work, NASA/JPL-Caltech)
Orbit covariance — what we deliberately do NOT carry	JPL publishes a full orbit covariance matrix per small body, and this registry does not ingest it. The reason is a measurement, not a preference, and it is recorded here so the decision can be checked rather than trusted. (a) It is not in the bulk response we snapshot: it lives on the single-object `sbdb.api` endpoint, one request per object — about 4,000 requests, not "the same response". (b) Its epoch is not our epoch. The covariance is stated at the epoch of the orbit SOLUTION while the elements are served at the current osculating epoch. Measured over a 45-object sample of this very PHA population, they differed on 43 of 45, by a median of 2,994 days — over eight years. Carrying a correlated uncertainty across that gap needs the state-transition matrix of JPL's own force model (DE441, SB441-N16, per-object non-gravitational terms), which this registry does not have; a correlated uncertainty transported wrongly is a numerical falsehood, so it is not transported at all. (c) It is expressed in a different element set (cometary e, q, t_p, node, peri, i against our Keplerian a, e, i, Ω, ω, M₀) and is of variable rank — 43 of the 45 sampled are 6×6, 2 are 7×7, and bodies with fitted non-gravitational terms run to 8×8 (Apophis, Bennu), 9×9 (1P/Halley) and 13×13 (C/2013 A1, a rotating-jet model). (d) Verbatim it would add 6.49 MiB to the download — about a tenth of the whole registry — for a quantity no part of the product could honestly evaluate at the epoch we render. The related `not_valid_before` / `not_valid_after` fields are skipped for a blunter reason: they are null on every object measured, Apophis included. The staleness question they would answer is answered instead by the observation arc and last-observation date, which are really there. What we carry today is the diagonal 1σ per element; expressing a correlated uncertainty remains an open limitation of this registry, stated rather than hidden	Probed 2026-07-29 against ssd-api.jpl.nasa.gov/doc/sbdb and the bulk field list (`sbdb_query.api?info=field`, 52 orbit fields, no covariance column)	Public domain (U.S. Government work, NASA/JPL-Caltech) — the licence is not the obstacle here; the epoch is
Dimorphos — the DART mutual orbit	ONE node, and the only object in this registry whose orbit humanity has deliberately changed. Dimorphos is absent from every catalogue we ingest because SBDB records the (65803) Didymos system as a single object, so it is placed from JPL's own DART reconstruction: ~126 mean-element phases along the `dimorphos_s547_long` kernel, stepping at the impact instant so the deflection is geometry rather than a sentence — the orbit visibly shrinks from 1189.2 m to 1152.0 m and speeds up from 11.92150 h to 11.36763 h. That change, −33.23 minutes, is measured here and lands inside the published −33.0 ± 1.0 (3σ) min, which is the only check that the deflection drawn is the deflection that happened. Horizons' own ELEMENTS are deliberately NOT used: they convert the state with the system's point-mass GM, and the resulting conic implies a 12.34 h period against a true 11.92 h — a 3.5 % error, the same size as the 4.6 % result being measured. Instead the period is measured from the trajectory and μ is set so the rendered mean motion equals it exactly (the same rule the 437 planetary satellites and 5,720 exoplanets already follow); that μ absorbs the primary's oblateness and is not the system's GM. Honesty: rotation is not modelled (tidally locked before the impact, left tumbling after it, no published post-impact pole), the 10.2 m primary–barycentre offset is unmodelled, and tracking against the kernel's own vectors over the two densely covered years is 0.6 m before the impact and 190 m after — that floor is solar perturbation swinging the eccentricity between 0.023 and 0.031, which no fixed conic follows. Search `Dimorphos`	NASA/JPL Horizons, target 120065803, kernel `dimorphos_s547_long`, ssd.jpl.nasa.gov/horizons — snapshot 2026-08-18; the published period checked against Naidu et al. 2020 (10.1016/j.icarus.2020.113777) and the deflection against Thomas et al. 2023, Nature 616, 448 (10.1038/s41586-023-05805-2)	Public domain (U.S. Government work, NASA/JPL-Caltech); cite JPL Horizons
Planetary satellites (moons & moonlets)	437 known moons of Mars→Pluto beyond the 22 curated majors — Jupiter's irregulars, Saturn's 280+ moonlets, the small moons of Uranus, Neptune and Pluto — each on its real mean-element orbit under its planet. JPL publishes each row's elements referred to the local Laplace plane (pole given), the planet's equator, or the ecliptic; all are rotated into the shared J2000 ecliptic frame (conversion verified against JPL Horizons osculating angles for Phobos, Io, Triton, Ariel and Charon). Mean elements at epoch: nodal/apsidal precession is not modelled (stated per node). Mean radius only where JPL publishes one (24); the rest render as points, no invented radii (A5). The 22 rows matching curated majors are not discarded: their a/e/i + radius reconcile onto the curated moons as corroborating/alternate sources (erase none — e.g. the Moon's JPL elements corroborate, Io's eccentricity is a kept alternate). Search `kind:moon`	JPL Solar System Dynamics — Planetary Satellite Mean Elements + Physical Parameters, ssd.jpl.nasa.gov/sats/elem — snapshot 2026-07-10; equatorial-frame reference poles per Archinal et al. 2018, IAU WGCCRE (10.1007/s10569-017-9805-5)	Public domain (U.S. Government work, NASA/JPL-Caltech)
NEOWISE diameters & albedos (join)	143,318 thermal-model (NEATM) diameter/albedo records joined onto the PHA layer by JPL designation: 295 matches — 1 fills a small body that had no measured radius yet, 258 CORROBORATE (agree with, within tolerance) SBDB's own measured diameter, kept alongside it as an additional confirming source, and 36 DISAGREE, kept as a selectable alternate (ERASE NONE — the registry never overwrites a value, every source is retained; see the object's provenance chain). An annotation/upgrade layer: adds no objects of its own	NASA/IPAC Infrared Science Archive (IRSA), `neowisesbpropv2` catalog — WISE/NEOWISE mission (JPL/Caltech); cite Wright et al. 2010 (10.1088/0004-6256/140/6/1868), Mainzer et al. 2011 (10.1088/0004-637X/731/1/53), Mainzer et al. 2014 (10.1088/0004-637X/792/1/30) — snapshot 2026-07-01	Public NASA/IPAC archive data; reuse with the acknowledgement above
Sentry impact risk (join)	The NASA/JPL Sentry-II impact-monitoring system's OWN published statements, joined onto the PHA layer by JPL designation (annotation-only, adds no objects, computes no probability of its own — held deliberately apart from the PHA orbit-class facet, which is a different catalogue's different criterion). THREE-VALUED: 15 PHAs are ON the current risk list and carry Sentry's cumulative impact probability, potential-impact count and year range, and Palermo/Torino scales, verbatim-close in Sentry's terms (101955 Bennu: 5.7×10⁻⁴ across 157 potential impacts 2178–2290; 29075/1950 DA); 870 PHAs are on the REMOVED table and carry the removal date with the source's own semantics — "objects are removed when all previous potential impacts are eliminated", typically by further observations (99942 Apophis, removed 2021-02-21: retired risk, stated); the 2,159 risk-list objects with no node here are counted (most sit below the PHA size cut), and a PHA absent from both tables carries NOTHING — absence is never presented as "probability zero"	NASA/JPL SSD/CNEOS Sentry Data API v2.0 (Sentry-II: Roa et al. 2021, 10.3847/1538-3881/ac193f) — snapshot 2026-07-28, risk list + removed table	Public domain (U.S. Government work, NASA/JPL-Caltech); SSD requests a download acknowledgement
Nuclear power sources carried on board (join)	Which space objects carry nuclear material, and of which of the three very different kinds — a fission REACTOR (tens of kilowatts thermal, with a fuel core to dispose of), a radioisotope THERMOELECTRIC GENERATOR (tens of watts electric), or a radioisotope HEATER UNIT (grams of decay heat). Nothing to do with the nuclear-detonations layer, which is terrestrial explosions. Annotation-only: adds no objects. TWO AUTHORITIES, kept apart. (1) The UN register, read from its own remarks — 57 objects: 29 reactors (the RORSAT class; each states its disposal, and the register's own estimate is that the bodies stay in orbit 400–2,310 years and the ejected fuel cores 2,020–3,340 years), 4 radioisotope heater units, 1 named RTG pair, and 24 declarations that say only “a nuclear power source”. THOSE 24 STAY UNSPECIFIED ON PURPOSE: in UN usage that phrase is a genus covering reactors and generators alike, so calling them RTGs would be inventing a species from a genus, however guessable the answer. Each declaration carries the rung of authority it rests on — a State's own notification under the NPS Principles (8), the UN register (33), or UNOOSA's secondary compilation (17) — and the sentence it was read from. (2) NASA's RPS programme, which names the DEVICE the register does not: 23 missions with model and unit counts (29 generators, 298 heater units). Where they overlap (19 objects) both statements are kept, never merged. Apollo 11 is recorded as a stated DISAGREEMENT between the two. Absence is UNKNOWN — no source lists the objects that carry nothing. Measured on the shipped bake (2026-08-13): the merged facet offers 61 objects (UN register 57 · NASA RPS 23 · both 19, so neither authority contains the other) and all 61 are annotated — 29 carry a reactor, 21 an RTG, 12 an RHU and 23 the unnamed genus (the four are not a partition: Voyager carries both an RTG and heater units). Search with `has:nuclear`, narrowed by `has:nuclear-reactor`, `has:nuclear-rtg`, `has:nuclear-rhu` or `has:nuclear-unspecified`; coverage is reported over the 21,261 UN-REGISTERED objects, never over every artificial object — a denominator holding 2,189 detonations and 687 launch sites would be a true number about nothing	UN Online Index of Objects Launched into Outer Space + the Principles Relevant to the Use of Nuclear Power Sources in Outer Space (UNGA res. 47/68, 1992); NASA/DOE Radioisotope Power Systems Program, snapshot 2026-07-30	© United Nations, reused under the UN terms of use; NASA Images and Media Usage Guidelines (factual use, NASA acknowledged — insignia and logotype excluded)
On-board autonomy — and why it is not a “has AI” flag	Whether a vehicle decides anything for itself, and WHAT KIND of decision. A boolean was refused: “AI” has meant symbolic reasoning, then expert systems, then machine learning, then large models, so one flag would mean different things at the two ends of a registry that spans 1957 to now. Instead the facet applies a written frontier — does the vehicle make, ON BOARD, without a human in the loop, a decision that COULD HAVE GONE OTHERWISE, from data it acquired itself? — and names the REGIME: navigating itself, re-planning its own activities, deciding which of its data matters, choosing its own targets, locating itself by matching the terrain it sees, picking its own landing point, driving itself, flying itself, or learning aboard. The frontier's cost is stated rather than hidden: attitude control, Kalman filtering, safe-mode entry, stored sequences and even neural image compression are all OUTSIDE it, because none of them chooses; a 1998 rule-based planner is INSIDE it. The line is about where and when the choice is made, not how sophisticated the method is. 14 records over 10 objects (Deep Space 1, Stardust, Deep Impact, Earth Observing 1, Opportunity, Curiosity, Perseverance, Ingenuity, OSIRIS-REx, Hayabusa2), each with a verified DOI. Two regimes are declared and deliberately EMPTY, and 6 further systems we believe flew are DECLINED for want of a citation that could be verified — both counts are published, because absence here means “nobody has documented it”, never “it decides nothing”. Coverage is biased toward missions whose autonomy is published in English, and any census drawn from this layer inherits that bias. Measured on the shipped bake (2026-08-13): 9 of the 10 keyed craft are annotated. The tenth is Ingenuity, and its absence is the rule working: the UN index holds it as “2020-052*”, international designator not assigned, so there is no key to join on — and pinning the record onto Perseverance’s designator would assert that THE ROVER FLIES ITSELF. It is counted and named in the bake log instead. Search with `has:autonomy`, narrowed by regime: `has:autonomous-navigation` · `has:autonomous-planning-scheduling` · `has:onboard-event-detection` · `has:onboard-target-selection` · `has:terrain-relative-navigation` · `has:hazard-detection-avoidance`. Coverage is reported over the 28,431 objects whose operational life GCAT has catalogued (`has:operated`) — the population in which the question is askable at all, since a launch site or a fragment of debris decides nothing by construction	Peer-reviewed and conference literature, one citation per record — Muscettola 1998; Bhaskaran 2000, 2012; Rayman 2000; Chien 2005; Estlin 2012; Lorenz 2017; Francis 2017; Balaram 2018; Nelessen 2019; Terui 2020 (DOIs in the library)	Facts cited from the published literature; the classification and its frontier are the registry's own, tentative and open to revision
Gaia DR3	Stars within 50 pc (distance precision < 1%), full 3D space velocities where radial velocity exists — expressed in the galactic rest frame via the cited solar motion (v_c 229 km/s, Eilers et al. 2019 + (U,V,W)_⊙, Schönrich et al. 2010). Deduplicated against BOTH the curated named stars and the exoplanet hosts (a Gaia row within 0.3 pc of either is dropped so a host is never double-counted); each such deduplicated star's effective temperature corroborates its matched exoplanet host (kept as corr/alt — erase none), and a row dropped near exactly ONE curated named star reconciles its Gaia distance + temperature onto that star the same way (Sirius A, Barnard's Star… — the α Cen/Proxima triple is honestly skipped as positionally ambiguous)	ESA / Gaia / DPAC — Gaia Data Release 3	ESA Gaia data credit policy (free use with credit) — snapshot 2026-06-10
Gaia DR3 — naked-eye bright sky	The familiar bright sky: every Gaia DR3 star of G ≤ 7 (the naked-eye limit with margin) at ALL distances — Orion's belt, Canopus, Betelgeuse and the rest that the 50 pc cut misses. The distance-precision gate is DELIBERATELY relaxed for this layer (a naked-eye star with a poor parallax still belongs in the sky): a star is placed at 1/parallax only when its parallax exceeds 3σ, and the per-star distance uncertainty is carried honestly (σ_d in the record's uncertainty side-map + a σ_d/d note); rows without a usable parallax — including some of the very brightest, saturated in Gaia — are EXCLUDED, never placed at an invented distance. Deduplicated by source_id against the ≤50 pc layer and positionally (0.3 pc) against the curated named stars and the exoplanet hosts (unambiguous matches corroborate the host's temperature, erase none). Its own `gaia-bright` layer, so each Gaia layer card states its own precision gate	ESA / Gaia / DPAC — Gaia Data Release 3	ESA Gaia data credit policy (free use with credit) — snapshot 2026-07-10
Hipparcos — the brightest stars	the ~150 BRIGHTEST stars in the sky (Sirius, Canopus, Vega, Rigel, Betelgeuse…) that Gaia DR3 saturates on (no usable parallax at G ≲ 3, so the naked-eye layer honestly lacked them): the Hipparcos main catalogue (ESA 1997) subset at Vmag ≤ 4, placed at its measured trigonometric parallax (> 3σ gate, per-star σ_d carried). Effective temperature is DERIVED from B−V via the published Ballesteros 2012 relation and labelled per record — never passed off as a catalogued value. Each star’s drift carries a MEASURED radial velocity from the XHIP compilation (Anderson & Francis 2012, joined by exact HIP number, σ_RV noted) — full 3-D space velocities for all 148. Deduplicated against every star population already ingested by direction + distance + BRIGHTNESS (the magnitude condition keeps binaries honest — without it Rigel would be erased against its 12-mag companion Rigel B). Its own `hipparcos-bright` layer; IAU proper names attach through the same positional name join as every star layer	ESA — Hipparcos main catalogue (SP-1200, 1997), via CDS/VizieR I/239/hip_main; teff relation Ballesteros 2012 (10.1209/0295-5075/97/34008)	ESA mission data — free use with attribution — snapshot 2026-07-10
XHIP distances — naked-eye completeness	The V ≤ 6.5 naked-eye stars the registry’s parallax honesty gates refuse — the junk-ORIGINAL-parallax luminous supergiants (Deneb: I/239 Plx 1.01 ± 0.57 fails 3σ; Alnilam…) — placed at the XHIP compiled heliocentric distance: the HIP2 New-Reduction parallax (van Leeuwen 2007) with the catalogue’s own statistical correction where a σ is published (carried per star as σ_d), else a cluster-membership fit labelled MODELLED with NO invented σ; a star with no XHIP distance (Aludra) stays honestly absent. Effective temperature DERIVED from B−V (Ballesteros 2012, labelled); measured XHIP radial velocities joined by exact HIP number (quality grade D — “no error available or other serious problems” — excluded). Deduplicated by exact HIP + the 6′ direction+magnitude identity gate against every star layer. Its own `xhip` layer. PLUS the RADIAL-VELOCITY EXTENSION over the whole sky: the same 46k-row XHIP compilation joined POSITIONALLY (6′ + magnitude gate, ambiguity honestly skipped) onto every star still carrying a tangential-only drift — thousands more stars gain full 3-D space velocities, each note naming the mechanism (`[XHIP, positional]`)	Anderson & Francis 2012, AstL 38, 331 — XHIP, Extended Hipparcos Compilation (10.1134/S1063773712050015), via CDS/VizieR V/137D; teff relation Ballesteros 2012 (10.1209/0295-5075/97/34008)	Published compilation via CDS/VizieR — free use with attribution — snapshot 2026-07-20
Reference pulsars	3 pulsars (Crab, Vela, PSR B1919+21) as natural EM beacons, described in the same emission language as human transmissions — their natural context	ATNF pulsar catalogue	Factual public data; per-record notes
ATNF Pulsar Catalogue	4,178 rotation-powered neutron stars (pulsars) placed in the Milky Way galactic frame, distance-gated — only pulsars with a catalogued distance are shown; ~140 without one are excluded rather than guessed (A5). The full catalogue as its own `atnf` layer, distinct from the 3 curated reference pulsars. Search `kind:pulsar`	Manchester, Hobbs, Teoh & Hobbs 2005, AJ 129, 1993 (10.1086/428488) · ATNF Pulsar Catalogue, atnf.csiro.au/research/pulsar/psrcat — snapshot 2026-06-24	Public catalogue; attribution / acknowledgement requested
Local Volume galaxies (UNGC)	864 galaxies of the Local Volume (within ~11 Mpc) — the GALAXY rung above the stellar layers: every named nearby galaxy (Centaurus A, M81, M101, Sculptor, the Andromeda dwarfs…) placed at the catalog’s ADOPTED published distance, the distance METHOD named per record (TRGB · Cepheids · surface-brightness fluctuations · Tully–Fisher · brightest stars · group membership; rows whose only adopted distance is the Hubble flow are labelled redshift-derived). The published distance ERROR is carried as σ_d where the paper’s own distance-measurement table lists one (298 records). 4D honesty (A5): a galaxy’s RADIAL velocity is measured (heliocentric + Local-Group-frame values in the record) but its tangential motion is UNMEASURED — no drift is modelled, positions are epoch positions. Parentage is gravitational (A1): the 69 galaxies within 1.0 Mpc — the measured Local Group zero-velocity radius R0 = 0.96 ± 0.03 Mpc (Karachentsev et al. 2009), a declared heuristic — join the Local Group anchor and co-move with its CMB-dipole flow; the other 795 are field galaxies in the CMB rest frame. The curated M31/M33/LMC/SMC + the catalog’s own Milky Way row keep their curated identities (name + major-galaxy position dedup; M31’s dwarf companions like M32/NGC 205 are preserved). Search `kind:galaxy`	Karachentsev, Makarov & Kaisina 2013, AJ 145, 101 (10.1088/0004-6256/145/4/101), via CDS/VizieR J/AJ/145/101; LG zero-velocity radius Karachentsev et al. 2009 (10.1111/j.1365-2966.2008.14300.x) — snapshot 2026-07-14	CDS/VizieR published-compilation data — free use with attribution (cite the paper + CDS)
Galaxy clusters (Planck PSZ2)	1,094 galaxy clusters from the second Planck Sunyaev-Zel'dovich source catalogue — every PSZ2 detection with a measured redshift, placed in the CMB rest frame at its line-of-sight comoving distance. Placement honesty (A5): the catalogue measures sky positions and REDSHIFTS, never distances — the z → distance conversion goes through one cited flat-ΛCDM door (Planck 2018 parameters) and is labelled MODELLED per record; the 559 detections without a redshift are EXCLUDED, never placed at an invented distance. The SZ-proxy mass M500 is carried with its published 1σ where given; no radius is invented and no drift is attached (cosmological recession is the expansion of space, not motion through it). Search `kind:cluster`	Planck Collaboration 2016, A&A 594, A27 (10.1051/0004-6361/201525823), via CDS/VizieR J/A+A/594/A27 — snapshot 2026-07-14	ESA Planck mission data via CDS/VizieR — free use with attribution (cite the paper + CDS)
Largest structures (curated)	14 cited landmarks at the scales ABOVE galaxy clusters — the Virgo (Local), Laniakea, Perseus-Pisces and Shapley superclusters, the CfA2 / Sloan / South Pole / Hercules-Corona Borealis great walls, the Boötes / Local / KBC voids, the Great Attractor region, the Pisces-Cetus supercluster complex and the Huge-LQG quasar group. A small, strictly-cited set: each structure is placed at a CATALOGUED centroid/anchor (each record says which); published redshifts / recession velocities convert through the same cited ΛCDM door and are labelled MODELLED, while a directly cited distance is used verbatim; walls and voids get NO invented shape or radius (sizes stay in the notes in the papers' own units); a structure whose published localization is a region carries an explicitly INDICATIVE direction, and the one contested structure is shown WITH its published rebuttal. Search `kind:structure`	Published astronomy, one citation per structure (e.g. Tully et al. 2014 — Laniakea, 10.1038/nature13674; Geller & Huchra 1989 — the CfA2 Great Wall, 10.1126/science.246.4932.897) — the full per-row DOI list is in the library	Public astronomical facts (positions/redshifts/classes); per-record citation
Gravitational-wave events	2 confirmed compact-binary mergers — GW150914 (the first direct detection, a binary black hole) and GW170817 (the first binary neutron-star merger, with an electromagnetic counterpart) — each rendered as a disturbance expanding outward at the speed of light from its detection instant, the natural-event analogue of the EM emission shell. A deliberately small, strictly-cited set; sky-localization centroids are representative directions within large credible regions, not precise source points (declared per record). Search `kind:gw-event`	Abbott et al. 2016 (GW150914, 10.1103/PhysRevLett.116.061102) · Abbott et al. 2017 (GW170817, 10.1103/PhysRevLett.119.161101) · the GWTC catalogue via the Gravitational Wave Open Science Center (GWOSC)	Public record; GWOSC strain data CC0
Gravitational-wave events (GWOSC full catalogue)	394 confirmed compact-binary mergers (the full GWTC-1 through latest-O4 CONFIDENT list, beyond the two curated ones) as its own `gwosc` layer, each an expanding shell at c from its detection epoch. Only confident-catalogue events are kept — marginal / independent-analysis / external-trigger / hardware-injection entries are dropped (A5). GWOSC publishes no point sky position, so each event's DIRECTION is a deterministic seeded placeholder (labelled display-only), placed at the real published median luminosity distance (285 events; the rest, mostly newest O4, carry no published distance and sit at a flagged nominal distance). Search `kind:gw-event`	Gravitational-Wave Open Science Center (LIGO/Virgo/KAGRA), event API — GWTC-1 (10.1103/PhysRevX.9.031040), GWTC-2 (10.1103/PhysRevX.11.021053), GWTC-3 (10.1103/PhysRevX.13.041039), GWOSC software (10.1016/j.softx.2021.100658) — snapshot 2026-07-01	CC0 / public domain; acknowledgement requested
Supernovae (historical + landmark)	9 supernovae humanity has actually witnessed — the naked-eye Galactic events across a millennium (SN 185 · SN 1006, the brightest recorded · SN 1054, the Crab · SN 1181 · SN 1572, Tycho's · SN 1604, Kepler's · Cassiopeia A) plus the two extragalactic landmarks (SN 1885A in Andromeda, the first seen in another galaxy; SN 1987A in the LMC, the first with a neutrino detection). Each is a point placed in the galactic frame at its catalogued distance, appearing in TIME at its observed peak (A2 — rewind before 1054 and the Crab has not yet exploded). A deliberately small, strictly-cited, fact-only set; SN 386/393 and the large modern extragalactic catalogues are omitted (A5). An optional LIGHT-SHELL overlay (kill switch `sn=1`) draws each event's photon front — at Earth's distance at the observed peak, expanding at c thereafter (a geometric c·Δt construction from the cited date + distance; scrub to 1054 and the Crab's front reaches Earth). Search `kind:supernova`	Published astronomy — Williams et al. 2011 (10.1088/2041-8205/732/1/L11), Winkler et al. 2003 (10.1086/375314), Hester 2008 (10.1146/annurev.astro.45.051806.110608), Ritter et al. 2021 (10.3847/2041-8213/ac2edf), Krause et al. 2008 (10.1038/nature07608), Reynolds et al. 2007 (10.1086/510515), Perets et al. 2011 (10.1086/504413), Arnett et al. 1989 (10.1146/annurev.aa.27.090189.003213)	Public astronomical facts (positions/dates/types/distances)
Naked-eye deep-sky objects (curated)	31 classical deep-sky objects visible to the unaided eye — the Messier-class set: 20 open clusters (Pleiades M45 · Hyades · Beehive M44 · Ptolemy M7 · the Perseus Double Cluster · NGC 6231 · the Jewel Box …), 9 globular clusters (ω Centauri · 47 Tucanae · M13 · M22 · M4 …) and 2 great nebulae (Orion M42 · Carina NGC 3372). Each is a point under the Milky Way in the galactic frame at its PUBLISHED distance, verbatim from the cited catalogue; open clusters carry the published Gaia DR2 half-members radius (r50) converted exactly to a physical extent at that distance, globulars carry the Harris integrated V but NO size (the queried catalogue publishes none — absent stays absent, A5). Node names carry the M/NGC/IC designation, so the strict sky-culture deep-sky join attaches the published Chinese/Korean names for M 7 · M 44 · M 45 · ω Cen · Carina · NGC 6231 with zero join change. M 31/M 33/LMC/SMC live in the local-universe layer, not here. Search `kind:cluster` · `kind:nebula`	Cantat-Gaudin et al. 2020 (10.1051/0004-6361/202038192, Gaia DR2 open-cluster census), Harris 1996 (10.1086/118116, globular-cluster catalogue), Melis et al. 2014 (10.1126/science.1256101, Pleiades VLBI), Menten et al. 2007 (10.1051/0004-6361:20078247, Orion VLBI), Göppl & Preibisch 2022 (10.1051/0004-6361/202142576, Carina Gaia EDR3), Smith 2006 (10.1111/j.1365-2966.2006.10007.x)	Public astronomical facts (positions/distances/sizes/magnitudes), cited per row
Constellations (native, per sky culture)	each PUBLISHED sky culture's constellation figures — the 88 IAU western stick figures, the classical Chinese sky (318 figures), Korean (271), H. A. Rey's redrawn western set, Indian, Norse, ancient Egyptian, Japanese moon stations, Sky & Telescope and Hlad western variants — drawn as lines BETWEEN the registry's own real 3-D stars (resolved by the same 6-arcmin + magnitude identity gate as the IAU name join): from Earth the familiar shapes appear; travel anywhere else and they honestly distort, because a constellation is an Earth-perspective overlay on stars at wildly different distances. ANY SUBSET of cultures renders SIMULTANEOUSLY (multi-culture overlay) — a compact Visualisation chips row (off · western · chinese · …; only cultures whose figures actually resolve are offered), each enabled culture drawn in its own deterministic hue (western keeps the classic steel-blue; hues chosen for colour-blind-reasonable spacing over the common overlays — with many simultaneous cultures some hues necessarily approach each other), still ONE pooled draw call; permalink `cn=<culture>[+<culture>…]` (e.g. `cn=western+chinese`; legacy `cn=1` still reads as western; an unknown member is dropped, an all-unknown value reads as off). OPTIONAL NAMES LEGEND (`cnl=1`, default OFF): each enabled culture's own figure NAME labels the figure at its on-screen centre (native form where the culture publishes one, the english translation as a fainter suffix; ≤24 labels — the figures nearest the centre of view win, re-elected on a ~2 s heartbeat), and a selected member star shows a "constellations" row naming its figures per enabled culture (a hue dot ties row, label and lines). A figure star with no registry star passing the honesty gates (e.g. a junk-parallax supergiant) is DROPPED with its segments and counted, never faked (A5 — no figure is completed by invention). OFF by default; the line sets lazy-load only when enabled (zero boot cost)	Stellarium sky cultures (stellarium-skycultures), through the registry's default-deny sky-culture governance — Tier-1 historical/scholarly traditions only, each culture's text/data licence re-verified at fetch AND at bake (living-community traditions and no-derivatives cultures are never published as line data); figure stars: ESA Hipparcos/Gaia via CDS/VizieR — snapshot 2026-07-20	Per-culture (CC BY-SA / GPL, copyleft honoured per slice) — a separable COLLECTION artifact (`atlantis-constellations.json`, each culture's table under its own licence; supersedes `atlantis-constellations-western.json`), never part of the SSR-GPL bake
Named surface features (IAU Gazetteer of Planetary Nomenclature)	15,870 officially approved names of natural surface features across 39 bodies the registry carries — every crater, mons, vallis, planitia… the IAU has named, from the Moon's 9,086 (Tycho, Einstein, Sinus Lunicus…) through Olympus Mons on Mars and Maxwell Montes on Venus down to the named boulders of Bennu — each a surface-fixed, co-rotating mark at its published center lat/lon (planetocentric, +East — the export's one declared convention, cross-checked per row), carrying the published diameter as its size and the IAU origin/etymology text. TWO NAME CLASSES, the source's own: 8,807 deliberate proper names and 7,063 systematic lettered designations ("Einstein A" — the Gazetteer's "Satellite Feature" class, kind `satellite-feature`), kept distinct so approach context can reveal proper names first. Honesty: 216 features of bodies the registry does not carry (Eros, Ida, Gaspra, Dimorphos…) are held out, never given a fabricated host; 312 features without a published diameter carry no size; nothing is generated. Search `kind:crater` · `kind:mons` · `under:MARS kind:mons`	Gazetteer of Planetary Nomenclature — USGS Astrogeology Science Center for the IAU Working Group for Planetary System Nomenclature (planetarynames.wr.usgs.gov), the site's own nightly per-target KML bulk exports — snapshot 2026-07-29	Public domain — the site's own FAQ, verbatim: "Everything in the Gazetteer of Planetary Nomenclature is in the public domain", with the requested citation form (IAU WGPSN, access date, site URL)
Sky-culture star names (Tier-1 cultures)	each PUBLISHED sky culture's own star names joined onto the registry's real stars by the SAME 6-arcmin + magnitude identity gate as the constellation figures — 2,813 registry stars carry cultural names (chinese 3,021 HIP names → 2,812 stars · chinese_contemporary 3,022 → 2,813 · korean 69 → 67 · indian 31 · norse 3 · egyptian 2 · western/Rey 2; unresolved names dropped + counted, never guessed, A5). A selected star shows a compact "sky-culture names" row with the source community + licence attribution per culture. DEEP-SKY NAMES (increment 2): the cultures' non-star name keys join onto the registry's own galaxy/nebula/cluster nodes by EXACT catalogue designation only — M 31, the Andromeda Galaxy, shows its classical Chinese name (奎宿增廿一, Legs Mansion XXI) exactly like a named star. PLANET NAMES (increment 3): the 5 ancient-Egyptian PLANET names join onto the solar-system planets by EXACT planet-name identity ("NAME Jupiter" ↔ the node named exactly Jupiter — identity certain by construction; ≥2 exact carriers would refuse) — Jupiter shows Horus Who Limits the Two Lands, Mars Horus of the Horizon, Mercury Associated with Set (Sebegu), Saturn Horus the Bull of the Sky, Venus God of the Morning. Of the 12 non-star keys in the snapshot, 1 joins as deep-sky + 5 as planets; the rest are counted and never guessed (no registry object carries M 7 / M 44 / M 45 / NGC 5139 / NGC 6231 / NGC 3372 — A5, ambiguity refused). GOVERNANCE (default-deny, `registry/skyculture.ts`): ONLY Tier-1 historical/scholarly traditions are published; living-community traditions appear as LINK-OUTS only (culture + community + upstream URL, zero name data — the full linked list is on the library page) pending documented community consent; no-derivatives cultures (incl. al-Sufi — its text/data licence is CC BY-ND 4.0) are excluded entirely. Loads lazily on the first star/galaxy/planet selection (zero boot cost)	Stellarium sky cultures (stellarium-skycultures), per-culture licences re-verified at fetch AND at bake — snapshot 2026-07-20; stars: ESA Hipparcos/Gaia via CDS/VizieR	Per-culture (CC BY-SA / GPL, copyleft honoured per slice) — a separable COLLECTION artifact (`atlantis-skynames.json`), each culture's table under its own licence, never part of the SSR-GPL bake
Eclipse shadows (geometric)	the geometric umbra/penumbra of a moon crossing its host body (Moon→Earth solar eclipses, Io’s shadow transiting Jupiter, Earth→Moon during a lunar eclipse), constructed purely from catalogued positions and radii — nothing invented; annular (antumbral) geometry handled by the cone sign. Bodies on fixed-epoch orbital elements have precession-limited positions, so eclipse timing is approximate far from the element epoch (stated on the toggle). On by default; Visualisation toggle; permalink `es=0` hides	derived at render time from the registry’s own catalogued positions + radii (no external source)	—
Coronal mass ejections (NASA/CCMC DONKI)	8,435 analysed coronal mass ejections over the FULL DONKI ARCHIVE — 2010-04 → 2026-07 (the catalog's whole record; DONKI's first analysis is 2010-04-03), tracing both solar cycles (Cycle-24 maximum 2014, minimum 2018, Cycle-25 maximum 2025, including the May-2024 Gannon storm) — refreshed by dated snapshot, not a rolling/live feed: a sim clock after the snapshot's end simply has no CME data yet, and none exists before 2010-04 in this catalog. Shown as its own `donki` layer — transient events at the Sun (existence-windowed), each carrying its cited directed-cone parameters (apex heliographic latitude/longitude, cone half-angle, bulk speed, time the leading edge reached 21.5 R☉) in its inspector notes. An event missing cone geometry is skipped, never completed with a guess (A5). A Visualisation toggle (“CME cones (DONKI)”, permalink `dc=1`) renders each event’s measured cone expanding ballistically from the Sun near its epoch — the orientation uses a double-verified Stonyhurst→ecliptic solar frame (Fränz & Harper 2002; IAU pole cross-check), and the extent beyond the 21.5 R☉ fit is a constant-speed extrapolation, labelled MODELLED (A5). Search `kind:cme`	NASA Community Coordinated Modeling Center — DONKI CMEAnalysis web service (kauai.ccmc.gsfc.nasa.gov/DONKI; the `api.nasa.gov/DONKI/CMEAnalysis` DEMO_KEY mirror serves identical data) — snapshot 2026-07-21	US-government work (NASA/CCMC); public domain — cite NASA/CCMC DONKI (no DOI)
Textures	11 planetary maps + the Milky Way structure map	Solar System Scope (planets) · NASA/JPL-Caltech R. Hurt (galaxy concept)	CC BY 4.0 / NASA public domain — see `/textures/LICENSE.json`
Synthetic backdrop	598 procedural galaxies/stars + 30,000 Milky Way field stars sampled from the NASA structure map — an artist-informed density proxy, NOT a stellar census (real stars are the Gaia layer); all tagged SYNTHETIC, rendered dimmed	Internal factory, seeded & reproducible; map credit NASA/JPL-Caltech R. Hurt	© Adrien Normier · SSR-GPL v2.0

2) Artificial elements

Humanity's footprint — every object, signal and scar we have put into space or sent through it.

Layer	Content	Authority / citation	License / terms
GCAT	47,675 artificial objects: payloads, stages, components; multisegment trajectories; landings. Small-body encounter phases joined 2026-08-13: proximity, orbit and surface rows whose GCAT Primary/World is a catalogued small body (Itokawa · Ryugu · Bennu · 67P/Churyumov-Gerasimenko · Didymos · Tempel 1 · Wild 2 · Toutatis) now parent on the SBDB/comet node itself — 80 phases across the Rosetta/Philae, Hayabusa, Hayabusa2, OSIRIS-REx, DART/LICIACube, Stardust, Deep Impact and Chang'e 2 families — and the letterless-Piece identity collision that folded a mission's deployed family into one node is un-folded for those launches (+25 nodes: MASCOT, the MINERVA/MINERVA-II rovers and target markers, both Hayabusa return capsules, the Deep Impact Impactor). A bound orbit is derived only where the primary's mass is MEASURED (Itokawa · Ryugu · Bennu · 67P); an estimated-mass primary or a hyperbolic flyby row yields a held fix at the catalogued instant, never an invented orbit (A5); primaries with no registry node (Eros, Gaspra, planetary moons — the Cassini class) stay excluded and counted. Includes (2026-07-20 sub-catalog completion) 1,797 failed-to-orbit objects shown at their launch site (existence-windowed to their fall — no fabricated suborbital orbit), 363 uncatalogued analyst objects (csocat, e.g. WT1190F) on their catalogued orbits, auxcat payload detail (pauxcat: category, activity), and hardware identity (alternate names · manufacturer · bus · dry/total mass where differing · operational-orbit class) in the searchable notes. Marked-erroneous (ERR) phantom entries and 0/0 undetermined-orbit rows are excluded rather than fabricated. The worlds.tsv 292-world reference table joins as a CORROBORATION source only (never a node producer): its independent radii/masses reconcile onto 52 matching solar-system bodies — 56 corroborating, 33 kept as alternates (erase none; GCAT quotes equatorial radii where the registry carries mean radii, an honestly-kept difference class). Two sub-catalogs stay deferred with stated reasons: ecat (attached-phase/EVA history parented by OBJECTS, which the body-parented phase model cannot yet render) and tmpcat (provisional scratch rows that graduate into satcat/auxcat at the next GCAT release)	J. McDowell, General Catalog of Artificial Space Objects (planet4589.org)	CC BY 4.0 — snapshot 2026-06-09 (worlds.tsv 2026-07-20)
Celestrak GP element sets (orbital-phase join)	28,390 general-perturbations element sets joined onto GCAT's currently-tracked Earth-orbit objects by SATCAT/NORAD number — the join upgrades every currently-tracked Earth-orbit object the snapshot covers: the orbital-plane orientation (Ω, right ascension of the ascending node), the argument of periapsis (ω) and the orbital-phase angle (M₀, mean anomaly) go from a deterministic placeholder spread to MEASURED values at the element set's own epoch. GCAT's own Perigee/Apogee/Inclination columns (semi-major axis, eccentricity, inclination) are unaffected — this join fills only the three angles GCAT does not carry. ω joined 2026-07-29: it was being parsed and then dropped, so all 19,989 joined orbits were drawn on a MEASURED plane at an UNMEASURED point — displaced along their own track by ω, whose magnitude across the snapshot runs 30° at the 10th percentile and 90° at the median. Carrying it moved those objects a median of 10,146 km (p10 4,807 km, p90 24,489 km), and it is ROTATED, not copied: ω is measured from the ascending node, which moves when the plane is converted from GCAT's body-equatorial frame into the ecliptic. Honest remainder, measured: 110 of the 19,989 (0.55%) still show ω = 0 on their earliest orbital phase, because that phase comes from a row processed before the object's catalogue number was known or from a sub-catalogue the join deliberately does not trust for identity; their current phase carries the measured value. An annotation/upgrade layer: it adds no objects of its own; unmatched objects (deep-space, decayed, surface-landed) keep the honest placeholder-spread note. Search/inspect: the per-object provenance note reads "Ω,ω,M₀ MEASURED (Celestrak GP)" when joined	Celestrak (Dr T.S. Kelso), General Perturbations (GP) element sets — public US Space Surveillance Network data	Public US Government data redistributed by Celestrak; reuse permitted with attribution to Celestrak (Space-Track deliberately not used — its terms forbid automatic redistribution) — snapshot 2026-07-09
CelesTrak SATCAT (independent-catalog reconciliation)	69,870 satellite-catalog rows (the full US SSN catalog incl. decayed objects) reconciled onto GCAT objects by NORAD number (COSPAR fallback) — an INDEPENDENT catalog of the same population, installed to cross-check GCAT's launch/decay history: 42,156 objects matched (re-measured 2026-08-13; the delta from earlier figures is dominated by later re-ingests, plus one deliberate correction — the M-V-5 third stage no longer wears Hayabusa's NORAD number, which the old letterless-Piece fold had mis-attached); launch dates 41,467 corroborated to the UTC day / 689 kept as disagreeing alternates; end/decay dates 16,009 corroborated / 2,890 kept as alternates (erase none — the canonical is never rewritten, every disagreement stays inspectable); 141 objects GCAT left open-ended gained a labelled Earth-reentry end date from the SATCAT; 28 operational-status annotations where GCAT carries no activity window. An annotation/reconciliation layer: it adds no objects of its own; surface-landed hardware is never given an end date (the footprint persists)	CelesTrak (Dr T.S. Kelso), SATCAT satellite catalog — public US Space Surveillance Network data	Public US Government data redistributed by CelesTrak; reuse permitted with attribution to CelesTrak (Space-Track deliberately not used — its terms forbid automatic redistribution) — snapshot 2026-07-14
GCAT launch log (site / vehicle / outcome join)	The GCAT launch log (launch.tsv) joined onto catalogued objects by Launch_Tag (47,319 of 47,675 objects; the honest remainder is the csocat analyst objects, whose launch is by nature unidentified) → three new search facets: `site:` (launch site code), `lv:` (launch vehicle, e.g. `lv:soyuz`) and `outcome:` (success / failure / partial, read from the LaunchCode result letter; 'U' unknown is left unset, never guessed — A5). It also upgrades the launch time to minute precision (47,274 objects) where the log carries a time-of-day, replacing the vaguer satcat launch date. An annotation/upgrade join: it adds no objects of its own. Search e.g. `outcome:failure`, `site:CC lv:atlas`	J. McDowell, GCAT launch log (planet4589.org/space/gcat)	CC BY 4.0 — snapshot 2026-06-09
GCAT launch sites	687 launch sites on Earth's surface (their catalogued longitude/latitude) as their own `gcat-sites` node layer (kind:launchsite), each existence-windowed from its operational span (TStart/TStop). Coordinate-less generic entries (e.g. "Ocean", air-launch) are skipped rather than placed (A5). The site code is the same vocabulary as the object `site:` facet, so `site:<code>` returns the site node and every object launched from it. Search `kind:launchsite`	J. McDowell, GCAT sites catalogue (planet4589.org/space/gcat)	CC BY 4.0 — snapshot 2026-06-09
Upcoming launches (Launch Library 2)	94 announced upcoming launches as PLANNED markers on their launch pads (the pad's catalogued latitude/longitude), each carrying its agency, rocket, mission, target orbit and announced launch window; the marker EXISTS from the window's start and wears an amber "PLANNED — window" label — an announced launch has no measured trajectory, so NO orbit is drawn (A5). A launch the GCAT catalogue already logs (by launch designator or object name) is NOT re-emitted, so no doubles appear when reality catches up. Recent PAST launches are cross-checked against the GCAT launch log (annotation-only, erase none): 150 of 183 recent launch designators matched GCAT Launch_Tags, 148/150 minute-precise launch times agree to the minute (2 genuine source disagreements kept visible, max ~24 h; 33 launches newer than the GCAT snapshot). Image assets referenced by the API carry per-image licences and are deliberately not used. Search `kind:mission`	Launch Library 2 API — The Space Devs (ll.thespacedevs.com) — snapshot 2026-07-14	Free use per The Space Devs terms ("use the data in any way, shape, or form, and share what you create with it"; no raw forwarding without added value — this is a value-added integration); attribution encouraged and given
Deep-space trajectories	24 probes (Voyagers, Pioneers, Cassini, Dawn, DART, Rosetta, Galileo…) — heliocentric cruise arcs refined by piecewise osculating elements from the missions' own navigation solutions; planetary tours and flybys keep their catalog representation	NASA/JPL Horizons API (mission navigation teams' solutions)	Public domain (NASA/JPL); cite JPL Horizons — snapshot 2026-06-10
Predicted trajectories (Horizons)	Published PREDICTED ephemerides for in-flight future missions, rendered on the flown spacecraft's own node — the showcase is Europa Clipper's cruise to Jupiter (monthly osculating arcs through the announced 2030-Apr-11 arrival, bracketing the 2026-12-03 Earth gravity assist). Horizons' own object header states the honesty boundary verbatim ("based on tracking data through 2026-Jun-30. Trajectory thereafter is predicted"): arcs at/before it carry the navigation-solution label, arcs after it are labelled `horizons-predicted` — a published projection, never presented as flown history (A5). Scrub the clock past today and the craft rides the published prediction. Quarterly manual refresh	NASA/JPL Horizons API — JPL mission-design / navigation trajectory files (the mission team's own published prediction)	Public domain (NASA/JPL); cite JPL Horizons — snapshot 2026-07-21
Interplanetary cruise arcs (Horizons, second batch)	1,530 heliocentric osculating arcs on 49 more deep-space craft and upper stages — the Mars and Venus orbiters (Mars Express · MRO · MAVEN · Mars Odyssey · Tianwen-1 · ExoMars TGO · Venus Express · Akatsuki), the landers' cruise legs (Perseverance · Curiosity · InSight · Phoenix · Spirit · Opportunity · Mars Pathfinder), the L1/L2 observatories, and the Centaur/S-IVB/Delta stages that carried them. This is a gap in the catalogue, not a want of precision. GCAT publishes a, e, i for its heliocentric rows and NOT Ω, ω, M₀, so the ingest fills those with a documented placeholder spread: the ellipse is roughly right and the position on it is filler. Measured through the engine's own compose on the previous bake, Perseverance stood 466,032,238 km from Mars four days before it landed — on very nearly the opposite side of the Sun — and was already 316,515,757 km from Earth at the epoch of its own cruise element set, four days after launch. The anchor the arcs cannot force: InSight and the two MarCO cubesats flew on one launch and are held here in three SEPARATE JPL kernels, fetched separately and joined through three different nodes; the placeholder spread put them 96 / 323 / 343 million km apart at cruise midpoint, the arcs put them 4,489 / 6,981 / 8,157 km apart. Where the arcs win, they win by RULE: an integrated ephemeris outranks a catalogue mean element (rank 3 vs 1) over the window both cover, and the superseded set is kept on the record with its epoch, its a/e/i and the reason — 66 of them, arbitrated per window, with zero coin tosses. Stated limits: JPL's heliocentric coverage ends at planetary capture, so the residual at the catalogued arrival instant is ~1.2 million km rather than zero (2–3 orders better, and the capture window itself needs body-centred arcs); 15 of 91 windows have no JPL kernel at all (the Apollo S-IVBs past translunar injection, Clementine past its lunar phase, Hayabusa before its return leg), recorded by quoting JPL's own refusal; and two objects are REFUSED because Horizons and GCAT disagree about which piece letter carries which object of a shared launch (2024-179A · 2025-260C)	NASA/JPL Horizons API — `EPHEM_TYPE=ELEMENTS`, heliocentric, ecliptic, over the windows GCAT itself labels heliocentric	Public domain (NASA/JPL); cite JPL Horizons — snapshot 2026-07-30
Body-centred orbiter arcs (Horizons, third batch)	2,826 body-centred osculating arcs on 24 lunar, Mars, Venus and cislunar orbiters (lots 1+2) — LRO · GRAIL Ebb/Flow · LADEE · Lunar Prospector · Chandrayaan-1 · Danuri · CAPSTONE · Artemis I · SLIM and kin at the Moon, MRO · Mars Express · MAVEN · ExoMars TGO · Mars Odyssey · MOM at Mars, Venus Express · Akatsuki at Venus, THEMIS-B/C and the Chang'e boosters on Earth-centred deep ellipses. The gap this closes is the at-body phase, not the cruise: GCAT's lprcat holds ONE element set for a whole orbital phase — Mars Odyssey rolled a single set for 9,066 days at Mars, while a sun-synchronous orbit precesses its node ~0.5°/day, so after twenty years that is not "less precise", it is a different orbital plane. Each arc's µ is read from the response's own "Keplerian GM" line (pinned against the published lunar GM, 4,902.80 vs 4,902.800 km³/s² DE440), never assumed; epochs are converted from TDB with a declared leap-second table. Honest edges kept and counted: 6 hyperbolic approach epochs (the day of orbit insertion, before capture) are kept as approach phases; Chandrayaan-1's JPL kernel is the 2016 radar-recovery ephemeris, so 2008–2016 honestly keeps the GCAT set; 14 superseded catalogue sets are retained on the record with the verdict naming the loser in km. Four designations are REFUSED because Horizons carries two spacecraft ids for one COSPAR (Chandrayaan-2/3, LCROSS, Clementine) — choosing by name would be the Mariner-10 trap one notch over	NASA/JPL Horizons API — `EPHEM_TYPE=ELEMENTS`, body-centred (500@301/499/299), ecliptic, over the windows GCAT itself labels as the at-body phase	Public domain (NASA/JPL); cite JPL Horizons — snapshots 2026-08-17 + 2026-08-18
Horizons spacecraft roster (coverage measurement — adds no object)	The JPL Horizons major-body index (825 rows, 256 of them flown spacecraft) snapshotted as the AUTHORITY on which artificial objects a trajectory could be sourced for — so a claim about somebody else's archive becomes a dated measurement instead of a one-time assertion. Two numbers come out of it, both re-computed from the committed snapshot. Coverage: of the 247 Horizons spacecraft carrying a COSPAR designation, this registry holds 237 by designation and 5 more by name alone (98.0%); the 5 absent are named. The designation match and the weaker name match are reported as SEPARATE figures and never summed into one. The orphan roll: 29 GCAT deep-space objects that no catalogue places, because GCAT lists them but publishes no geometry for any phase — the Soviet planetary programme (Mars-4/6/7 · Venera-12/13/14 · Vega-1/2 · Fobos-1), Mariner 10, the upper stages that carried them, and three recent objects. Horizons carries exactly ONE of the 29 (New Horizons), and the registry already draws it. Two more have since been placed, and not by Horizons: Vega 1 and Vega 2 are drawn from NASA/NAIF SPICE kernels (row below), which leaves 26 unplaced — and the correction is worth stating in the same breath as the claim, because "not in JPL's archive" was true of the Horizons INDEX and quietly false of JPL's archive as a whole. The remaining 26 stay unplaced and counted rather than invented (A5). Resolution against the index is by COSPAR designation ONLY: measured live, a NAME query returns MARINER 2 for "Mariner 10" and Tianwen-1 for "Tianwen 2" — confident answers for the wrong spacecraft — and that fact is archived per refused object in the snapshot manifest rather than left as a warning	NASA/JPL Horizons API — the major-body index (`COMMAND='MB'`)	Public domain (NASA/JPL); cite JPL Horizons — snapshot 2026-07-30
NAIF/SPICE kernels (Vega 1 & Vega 2)	Two Soviet deep-space buses placed for the first time, from NASA/NAIF binary SPK kernels converted to sampled osculating conics at ingest — a browser never parses a kernel (§3.1). They were on the orphan roll above: GCAT lists them and publishes no geometry, and Horizons has no SPK for them. Anchored on published geometry, not on our own arithmetic: Vega 1 passes Halley at 8,885.5 km against 8,889 km published (−0.04%), Vega 2 at 8,064.7 km against ~8,030 km (+0.43%); the conic chain itself departs from the underlying integration by at most 123 m and 144 m, falling as step² — the planetary perturbation a two-body fit cannot carry. Coverage is the honest headline: 15.75 and 16.25 days, i.e. 3.4% and 3.6% of each mission, covering the March 1986 Halley encounter only. Existence is the ARC window, not the mission, so the record says nothing about the June 1985 Venus flyby fifteen months earlier — which is also why both read as "no catalogued arrival at a classified body" in the planetary-protection facet, a true statement about the window and a false one about Vega 1, which landed on Venus. Mariner 10 is REFUSED by name (`frame-not-measured`): its states sit in a NAIF frame nobody has measured against an external ephemeris, and Horizons cannot settle it because that NAIF id has since been REUSED for MSL/Curiosity — so resolving by id would have placed 1970s hardware on a 2010s trajectory, silently. Identity is the pair (kernel, id), never the id alone	NASA/NAIF (Navigation and Ancillary Information Facility) — PDS-published SPK kernels, naif.jpl.nasa.gov; encounter distances cross-checked against the published Vega/Halley flyby geometry — snapshot 2026-08-11	Public domain (U.S. Government work, NASA/JPL-Caltech). NAIF states redistribution in writing and reaches foreign-produced kernels explicitly ("All SPICE data appearing on this server, whether produced at JPL, at another U.S. entity, or by a foreign person/entity, are published by NASA's Planetary Data System"; clearance CL#05-2438). The permission sentences are archived beside the kernels and re-read on every fetch — the fetcher REFUSES to write a snapshot if they have gone
Nuclear detonations	16 curated detonations, Trinity → DPRK 2017, with existence windows	Public historical records, curated	Factual public data; per-record notes
Nuclear explosions	~2,170 detonations 1945–1998 (SIPRI + CMR merged), 19 test sites, EMP emissions modeled per atmospheric/space shot	SIPRI/FOA "Nuclear Explosions 1945–1998" (2000) + OGSO/CMR "Catalog of Nuclear Explosions" (rev. 3, 2006), both from the CMR database (Yang, North & Romney 2000, LDEO) — via data-is-plural/nuclear-explosions	Public domain. Import specification document © Adrien Normier. EMP powers are MODELLED (yield-scaled, Starfish Prime-anchored) — snapshot 2026-06-10
Nuclear weapons tests (OWID aggregate cross-check)	A bake-time CROSS-CHECK only — adds no nodes. Our per-country detonation tally (the SIPRI+CMR layer + curated nukes) is diffed against Our World in Data's annual per-country nuclear-weapons-TEST counts, and the agreement/discrepancy is written onto the nuclear-tests layer card. A5: OWID counts "tests" (a salvo may detonate several devices) and runs the eight declared states to the present, while our count is of individual detonations and its base ends ~1998 — a labelled systematic offset is expected, never reported as silent agreement	Our World in Data, "Nuclear weapons tests" (ourworldindata.org), compiled from the Federation of American Scientists / Bulletin of the Atomic Scientists (drawing on SIPRI) — snapshot 2026-07-01	CC BY (Our World in Data); attribution to OWID + FAS/SIPRI
EM signals (METI)	9 intentional interstellar transmissions, 5 emitter sites incl. Arecibo (with collapse date)	Zaitsev (2006) and public records	Factual public data; per-record notes
ASAT tests	The 5 destructive anti-satellite intercepts 1985–2021, as events on the targets' documented orbits (intercept point along the orbit not catalogued — placeholder anomaly, declared per record)	Public intercept records (tracking data, official statements)	Public record
Maneuver exhaust (derived)	118 propellant puffs located at navigation-arc Δv discontinuities, each carrying a Tsiolkovsky propellant mass — upper-bound proxies, declared limits in the method article	Derived from the JPL Horizons layer (factory Footprint)	Dataset © Adrien Normier; underlying arcs public domain (NASA/JPL)
Habitability map (habitability facet)	the cited limits of life (replication −15→122 °C, a_w floor 0.585/0.611, Deinococcus radiation, SR-SAG2 255 K/0.60/500 yr) + a per-environment habitability table (habitable / marginal / uninhabitable / control / survivable), surfaced in the inspector "habitability" block; "has water ≠ habitable" controls included. The survivable verdict is a DIFFERENT claim, not a weaker one: a dormant cell persisting, never anything growing. It carries a derived, flat-ground airless-body law (published microbial UV lethal doses and maximum growth temperatures against a blackbody stellar ultraviolet and the body's own geometry) reporting how much SHADOW a polar site would need — a niche is made by topography, which this law cannot find	Curated data-as-code from the limits-of-life & ocean-worlds literature (Takai 2008; Stevenson 2015/2017; MEPAG SR-SAG2 2014; NRC Ocean Worlds 2012; Waite 2017; Hand 2007; Hallsworth 2021; Saxena 2026)	Dataset © Adrien Normier; underlying studies public (per-row citation)
Solar-system water map (water facet)	per-body inventory of every water phase — ice · subsurface ocean · brine · vapour · hydrated minerals · adsorbed · deep supercritical (Titan's hydrocarbon seas flagged non-water); cited & confidence-tagged, surfaced in the inspector "water & ice" block	Curated data-as-code from the primary mission literature (MESSENGER, LCROSS/LRO, M³, MRO, Mars Express, SWIM, Dawn, OSIRIS-REx, Hayabusa2, Galileo, Cassini, Juno, Voyager 2, New Horizons, Rosetta, Herschel)	Dataset © Adrien Normier; underlying observations public (per-row citation)
Biomass map (biomass facet)	the modelled off-Earth biomass ceilings (Enceladus ~10³–10⁴ kg C, Europa ~10⁶ kg C subsurface, Mars density-only — no invented planet total) expressed as % of Earth's measured 5.5×10¹⁷ kg C baseline, in the unified Life & Water panel; MODELLED ceilings, clearly tagged	Curated data-as-code; Earth baseline Bar-On, Phillips & Milo 2018, PNAS (10.1073/pnas.1711842115)	Dataset © Adrien Normier; underlying study public
Exobiology-status ladder (exobiology facet)	a curated, cited, per-body ladder of what can honestly and tentatively be said about the possibility of life — ranked by DEFENSIBILITY (life-detected · habitable-measured · habitable-inferred · candidate-biosignature-contested · prebiotic-chemistry · agnostic), surfaced in the unified Life & Water panel. Nowhere off Earth has confirmed life; every rung is a claim about EVIDENCE, and contested signals (Venus phosphine, Mars methane) are shown WITH their published rebuttals. HYPER-HUMBLE / NON-NORMATIVE / A5	Curated data-as-code from the primary literature (Waite 2017; Postberg 2018/2023; Hand 2007; Tarnas 2018; Webster 2018 & Korablev 2019 & Eigenbrode 2018; Greaves 2021 & Villanueva 2021; Hörst 2017; Iess 2012; Raymond 2020 & De Sanctis 2017; Saur 2015; Glavin 2025; Oba 2023; Colaprete 2010) — per-row DOI in the library	Dataset © Adrien Normier; underlying studies public (per-row DOI)
Icy-world interior structure	cited interior layering of 9 candidate ocean worlds (Europa, Ganymede, Callisto, Enceladus, Titan, Mimas, Triton, Pluto, Ceres) — mean radius · ice-I shell · subsurface-ocean thickness · rock-core radius, confidence-tagged; the inspector "interior" line and the data spine for the subsurface cutaway	Curated data-as-code from the primary literature (Čadek 2016; Hemingway & Mittal 2019; Lainey 2024; Steinbrügge 2024; Hemingway 2013; Vance 2018; Park 2016; Ermakov 2017; Nimmo & Pappalardo 2016) — per-row DOI in the library	Dataset © Adrien Normier; underlying studies public (per-row DOI)
Surface imagery (sourced terrain)	real body imagery streamed on close approach (kill switch `terr=1`): a global equirectangular mosaic draped on the body plus a high-resolution local patch ring as you zoom in. Mars wears the Viking MDIM 2.1 colour mosaic (232 m), the Moon the LRO WAC mosaic, Earth the GIBS Blue Marble (+ optional daily true-colour cloud cover). The body stays a smooth sphere (no global DEM); tiles cover the global mosaic, never suppress it. Streamed from the pinned source, never hosted by us	NASA Trek (Mars/Moon WMTS) · NASA EOSDIS GIBS (Earth) — open plate-carrée tile services	Public domain (NASA/JPL/USGS/GSFC); attribution shown in-app with the imagery
Moon & small-body surface imagery (sourced terrain)	real global mosaics for 15 more bodies, draped on close approach exactly like Mars/Moon: Titan (Cassini ISS 938 nm albedo mosaic), Enceladus (Cassini ISS 100 m control mosaic), Mimas · Rhea (Cassini ISS, DLR/Roatsch et al.), Tethys (293 m) · Dione (154 m) · Iapetus (783 m) (Cassini/Voyager), Io · Europa · Ganymede (Galileo SSI/Voyager), Phobos (Viking VIS, DLR-controlled), Ceres & Vesta (Dawn Framing Camera), Mercury (MESSENGER MDIS BDR 166 m), Venus (Magellan C3-MDIR SAR). Bodies with no real published mosaic (Deimos, Callisto, Triton, Charon, the Uranian moons…) deliberately keep the labelled generic texture — never a wrong map. Streamed from the pinned source, never hosted by us	NASA Solar System Treks (per-body EQ WMTS) — every layer id + tile + CORS header confirmed live 2026-07-10	Public domain (NASA/JPL/USGS/JHUAPL; Cassini mosaics credit SSI/DLR, Dawn mosaics credit UCLA/MPS/DLR/IDA); attribution shown in-app with the imagery
Apollo landing-site super-detail (LRO NAC)	at close zoom over the six Apollo sites — Apollo 11 (Tranquility Base), 12, 14, 15 (Hadley Rille), 16 (Descartes), 17 (Taurus-Littrow) — the Moon's patch ring switches to the NASA Trek LRO Narrow Angle Camera landing-site mosaic (to z14–16), the lunar analogue of the Mars rover HiRISE orthos: a transparent PNG outside its small box, so the WAC global mosaic shows through at the edges. Coverage boxes taken from the Moon Trek catalog records, never guessed	NASA/GSFC/ASU LRO LROC NAC, via NASA Trek (Moon EQ WMTS) — layer ids + coverage + CORS confirmed live 2026-07-10	Public domain (U.S. Government work); LROC credit ASU
Your own KML (USER overlay)	load any terrestrial .kml (a Google Earth export…) from the Visualisation stage — its points, paths and polygon outlines drape on Earth's rotating ground as a USER-provenance overlay. Strictly local (§ zero-server): the file is read in your browser and never transmitted anywhere; altitudes are draped at ground level (never invented); capped counts are reported honestly. Display-only — never enters the registry	the user's own file (USER provenance)	the user's own data — no licence claimed by us
Compare your database (local comparison tool)	from the Search & AI palette ("Compare your database (local)" — no AI key needed): load your own CSV or JSON (an array of objects) and compare it against the loaded registry. Matching runs a four-rung LADDER, matches counted per rung: rung 0 EXACT catalog identifiers (COSPAR with its piece letter · NORAD/SATCAT number · HIP number); rung 1 DESIGNATIONS (a launch-level COSPAR without the piece letter — expected AMBIGUOUS on any multi-object launch, stated · minor-planet catalogue number / IAU provisional designation · Gaia DR3 source_id); rung 2 POSITION for star-like rows (RA/Dec — decimal degrees or sexagesimal, RA hours convention — within 6′ plus \|Δmag\| ≤ 1.5 magnitude arbitration, the registry's own star-identity gate; SYNTHETIC backdrop stars are never matchable); rung 3 CORROBORATED NAME — a bare name alone NEVER matches: a same-named registry object is admitted only when a second mapped field corroborates it (launch/decay on the same UTC day · \|Δmag\| ≤ 0.5 · radius/mass within the census near band); same-named rows nothing corroborates are counted NAME-ONLY, never common, and the rung honestly DECLINES (stated) when the mapping carries no corroborable field. Reports common / theirs-only / ours-only / ambiguous / name-only counts, plus a per-field census on launch & decay dates (same-UTC-calendar-day tolerance, stated on the row), magnitude (\|Δmag\| bands, stated), and the measured radius (m/km, user-confirmed unit) and mass (kg) — σ-AWARE where both sides carry a 1σ (the 2σ criterion, verdicts counted per regime; else the flagged σ-blind relative spread), with named worst-disagreement examples. The row classification runs in a background Web Worker where the browser provides one (the page stays fluid on a 250,000-row file), with an equivalent chunked main-thread fallback — which engine ran is stated. After a run the results are live as SESSION facets — cmp:matched · cmp:disagree · cmp:ours-only · cmp:compared — drivable as badges or by the AI through the ordinary query grammar; session-only (replaced by the next run, never in a permalink — a shared cmp: badge resolves to zero elsewhere). Three-valued honesty: a field absent on either side is UNASSESSED, never a disagreement; a key matching several of our objects is counted AMBIGUOUS, never guessed. Strictly local (§ zero-server): the file is read in your browser and never transmitted; there is no URL import by design; caps 50 MB / 250,000 rows with truncation stated; the result has no share or permalink path (honestly non-portable) and is for your own use. Display-only — never enters the registry. The compare panel also carries the PAIRWISE SOURCE-OVERLAP TABLE: for every cross-source identity join the bake itself performs (Celestrak SATCAT ↔ GCAT · Launch Library ↔ the GCAT launch log · Gaia ↔ exoplanet hosts · NEOWISE ↔ SBDB · JPL satellites ↔ the curated moons · Gaia ↔ curated stars · GCAT worlds ↔ bodies · the Exoplanet Archive ↔ curated stars), how many objects are doubly-sourced and how many values were kept as corroborations vs disagreeing alternates (erase none) — persisted at bake time in the small `atlantis-joins.json` sidecar (censuses only, never row dumps) and never recomputed in the browser	the user's own file (USER provenance) compared against the loaded bake; the overlap table: the bake's own join reports over the committed snapshots	the user's own data — no licence claimed by us; catalogs disagree legitimately, nothing here is a verdict
Sun imagery by date (GOES-SUVI)	with the toggle on (kill switch `si=1`, off by default) the Sun's procedural surface is replaced by a REAL full-disc solar image nearest the sim clock, projected onto the sphere. The earth-facing hemisphere shows the observed disc; the rest is the SAME disc replicated at 90/180/270° about the polar axis — a disclosed hypothesis, never claimed as observed on the far side. Honest scope: SUVI is a rolling recent buffer (~2017-present, GOES only), so a sim clock outside the window degrades to the generic Sun texture. Streamed from the pinned source, never hosted by us	NOAA SWPC (Space Weather Prediction Center) GOES-SUVI 195 Å full-disc PNGs — services.swpc.noaa.gov (CORS-open animation index, timestamped frames)	Public domain (US Government — NOAA / GOES-R); credit shown in-app
Rover/lander super-detail (HiRISE)	at rover/helicopter/lander zoom — where the 232 m global mosaic shows a single texel — the patch ring switches to the NASA Trek HiRISE site ortho covering that landing site, the sanctioned exception to the smooth-sphere "no-DEM" doctrine (which still stands for the generic body). Installed sites: Perseverance + Ingenuity (Jezero, 25 cm controlled ortho), Curiosity (Gale), Opportunity (Meridiani), Spirit (Columbia Hills), InSight (Elysium). The ortho is a transparent PNG outside its small box, so the global mosaic shows through at the edges; the HiRISE credit replaces the base credit while it is on screen	NASA/JPL · University of Arizona HiRISE (MRO), via NASA Trek (Mars EQ WMTS) — layer ids + coverage confirmed live 2026-06-17	Public domain (NASA/JPL); HiRISE credit University of Arizona
Rover tracks from orbit (HiRISE)	the published orbital observations of the Curiosity (Gale), Perseverance (Jezero) and Opportunity (Meridiani) wheel tracks, as surface-fixed marks linked to MARS (`kind:site`) at each worksite — "traces are objects linked to the planet, not the rover"	NASA/JPL · University of Arizona HiRISE (MRO) — uahirise.org/releases/msl-tracks	Public domain (NASA/JPL); HiRISE credit University of Arizona
Planetary protection (pp facet, derived)	per-object COSPAR classification (mission class · physical contact · contemporary & present-day category) and a modelled upper-bound bioburden under era-compliance, with the per-body "maximal compliant burden" metric — method & declared limits in the method article; never a compliance verdict. COVERAGE — 100% of the launched population. COSPAR's obligation attaches to a MISSION and its DESTINATION, not only to the objects that touched the ground, so the classification is derived for every one of the 51,799 artificial objects from the catalogued TRAJECTORY: the destination is not a column (GCAT's `Dest` holds dispositions — LEO, GEO, landing strips; `Primary` reads "Earth" on 69,360 of satcat's 69,420 rows) but the STRUCTURE, since GCAT carries it through its sub-catalogues and the bake materialises those as phases under their own parent body. Measured on the shipped bake (re-measured 2026-08-13 after the small-body encounter phases landed): 1,254 objects reached a body beyond Earth — Moon 1,017 · Mars 94 · Venus 90 · Jupiter 21 · Ryugu 17 · Saturn 8 · Itokawa 8 · Mercury 6 · Tempel 1 5 · 67P, Didymos and Wild 2 three each · Bennu 2 · Europa, Ganymede, Titan, Uranus, Neptune, Vesta, Ceres and Toutatis one each — and 50,545 reached none (50,279 never left the Earth system, 266 heliocentric with no catalogued arrival — eleven former no-arrival cruisers, Hayabusa and Rosetta among them, now carry their catalogued small-body arrival; the Vega 1/Vega 2 buses placed from SPICE kernels remain in this class, their ingested arcs covering the Halley encounter and not the Venus flyby), which is itself a classification: no target body means no forward-contamination category. WHAT THE COVERAGE PERCENTAGE IS COUNTED OVER, and why that was corrected (2026-08-13). The facet used to report its coverage over every artificial object, and that was a true number about nothing: an Earth-orbit satellite, a launch pad or a nuclear detonation cannot reach a body, so it can never carry a category and does not belong in the denominator of one — the reading was 1,497 / 51,774 = 2.89 %. The eligible population is the objects that LEFT THE EARTH SYSTEM, and it is now a search facet in its own right: `has:beyond-earth` — 1,520 objects (re-measured 2026-08-13), everything humanity has sent past Earth's own frame (the Moon counts, and so does heliocentric space). It is derived from the bake's frame structure, not from a list of names: an object's frames are its static parent plus every phase parent, and a frame that is itself artificial contributes ITS frames, so a spacesuit hanging off a lunar module inherits the Moon. Counted that way the figure reads 100 % — every object that could carry a category does carry one — which is what this row has said in words since it was written. The agreement is by construction (the facet and the population read the same frame closure), so the percentage's job is to REDDEN if the two ever diverge rather than to carry news; absence of a category inside that population would mean none is classified here, never that no obligation was owed. Jettisoned hardware is in on its own merits — 88 rocket stages and 55 components carry a category of their own. THE PHASE SHAPE IS THE EVIDENCE, at the resolution the category needs and no finer: a fixed offset in the body's frame is surface contact, orbital elements are an orbiter, a frame the object leaves again is a passage — a flyby, which COSPAR states in the SAME clause as an orbiter ("Category III … mostly flyby and orbiter") — and a terminal frame assignment with neither is an arrival the catalogue does not resolve. Where that leaves orbiter and lander both open AND their categories differ at that body, no category is claimed (2 objects); a body COSPAR's tables do not classify gets none either (36 since 2026-08-13 — the mission-target small bodies Itokawa, Ryugu, Bennu, 67P, Didymos, Tempel 1, Wild 2 and Toutatis are not in COSPAR's body listing, so the craft, rovers and impact scars at them are counted rather than shoehorned into a category another body's clause would supply). Two refusals more: the `atmos` refinement (destroyed in descent vs reached the surface) is invisible to geometry and stays curated, and a flyby the catalogue kept in heliocentric elements is not seen at all — so the flyby population is undercounted, never overcounted. THE RESULT, AND THE ONLY SHAPE IT MAY BE PUBLISHED IN: 96 objects carried a decontamination obligation under the policy in force at their launch (Mars 92 · Europa, Ganymede, Titan, Ceres one each) — an obligation being a category that CAPS the delivered bioburden, which Categories I and II never do; the documentation is publicly traceable for 0 of them. The two counts are asymmetric by nature: the obligation is exhaustively computable (body × mission class × epoch, all three in the bake), the conformity is not (it depends on public documentation, the 17-row curated column — whose objects are all Moon and Venus, i.e. category II, which is why not one of them is among the 96). The difference is NEVER a non-compliance count: undocumented is not non-compliant. A pre-1964 launch carries no obligation at all, because no policy existed to carry — the regime table dates it.	Derived from the bake's own phase structure (registry/ppframes.ts) + the published COSPAR policy tables (registry/pp.ts); the documentation column curated per object. Category-specific body listing and the Mars numbers read from the COSPAR Policy on Planetary Protection, with the 2024 Icy-Worlds restructure (Space Research Today n°224)	Dataset © Adrien Normier; GCAT public (CC BY); COSPAR policy public
Footprint classification (`containment` & `vector` facets)	a tentative, provisional sorting of human-made objects along two axes: a containment class (`mission` — reach bounded to an object/body · `biotope` — reach into one planetary environment · `uncontained` — reach outward as light/radio) and a vector (engineered-object · fragment · effluent · biological · electromagnetic · energetic · kinetic). A derived heuristic (from kind/origin/host/emission), offered for filtering only — non-normative, never a value judgement, open to revision; shown in the record's "footprint" row. Search with `containment:biotope`, `vector:fragment`	Our own tentative interpretation, borrowing the standard environmental-assessment chain (Source–Pathway–Receptor, Holdgate 1979; DPSIR, OECD/EEA). No external endorsement is claimed	Derived classification © Adrien Normier
3D models (index)	185 spacecraft/instrument GLB models indexed by URL (binaries stream from the pinned source commit, never stored here); 256 reconciled onto registry objects by conservative name match + curated aliases — including the ISS, Skylab Orbital Workshop, the Space Shuttle orbiter fleet (STS-qualified, so the SpaceX "Dragon Endeavour" Crew flights are not dressed as shuttles), the Apollo Lunar Module at the Apollo landing sites, and Deep Impact (whose GCAT node is named by its terse designator "DIF", so only a measured-safe curated alias could bridge it)	NASA 3D Resources (github.com/nasa/NASA-3D-Resources)	NASA media guidelines; per-model credits in the repo
Observed small-body and moon shapes	23 real (observed, never procedural) 3-D shape models joined onto registry nodes — 6 radar models streamed from the NASA PDS Small Bodies Node's own open mirror, plus 17 vendored: Apophis (radar; the original §3.1 no-CORS exception), the six MISSION TARGETS of 2026-08-13 — Didymos (DART/DRACO stereophotoclinometry v003), 67P/Churyumov-Gerasimenko (Rosetta SPC), Bennu (OSIRIS-REx SPC), Itokawa (Hayabusa/AMICA Gaskell), 81P/Wild 2 (Stardust NavCam plate model) and 9P/Tempel 1 (Deep Impact + Stardust-NExT 2012 plate model) — and the TEN MOONS AND MINOR PLANETS of 2026-08-30. The mission-target science meshes are 3.6–118 MB, phone-hostile at any origin, so each is DECIMATED at ingest (uniform vertex clustering to ~8k vertices, 460–486 KiB each, 2,903,403 bytes for all six — measured, not estimated) and the record credit says "decimated": the global figure is preserved, the science mesh it is not. Wild 2's un-imaged side is the model authors' own extrapolation, stated. THE TEN ARCHIVED FIGURES THE JOIN COULD NEVER REACH. The designation walk is SBDB-shaped and gated to asteroid/comet kinds, so Phobos and Deimos sat as smooth spheres although their shape models have been archived for decades. The join now also keys on the EXACT node id — an id and never a name, because the registry holds a 30 m nuclear detonation whose name is “DIONE” and another named “VESTA”, and a name join would hang a moon’s figure on a bomb. The ten, 3,321,583 bytes in all: Phobos, Mimas, Dione, Tethys and Phoebe from R. Gaskell’s stereophotoclinometry vertex/facet tables (decimated at ingest from 25,350 to 7,755–7,993 vertices, the decimation stated on each record), and Deimos, Vesta, Hyperion, Janus and Epimetheus from P. C. Thomas’s 5° latitude/longitude radius grids. A grid file states no topology at all, so it is DERIVED from the grid shape, and a table that is not a complete rectangle — or that has the right row count with a cell missing — is refused rather than guessed at. A prerequisite for the Martian pair was the rotation: a mesh drawn in a frame that is not the body’s own is a claim about which side faces where. TWO WAYS A GRID FILE PRODUCES A CONFIDENT WRONG BODY, both found by measurement rather than by review — and this is the part worth reading. (1) THE SIGN. These grids are WEST-positive where a body-fixed IAU frame is EAST-positive, and the archive label does not say so. Read with the wrong sign a body comes out MIRRORED — the same craters, the same limb, the wrong face toward its planet — and nothing complains; ESA’s own Deimos kernel carries exactly that correction in its version history (V01 → V02, “changed treatment of input longitudes to West-positive”). The sign used here is ANCHORED rather than assumed: converted west-positive, our Deimos agrees with ESA’s independently built DEIMOS_K005_THO_V02 kernel to a mean 69.5 m and a worst 495.6 m on a 15 km body, where the east reading sits a mean 310 m away. The four sibling grids inherit that sign from the same archive bundle, which is an inference and is written here as one. (2) THE SEAM. The files write BOTH 0° and 360° and repeat each pole across a whole ring, so a reader that takes the rows as written emits 179 duplicate vertices and leaves the mesh OPEN — measured at 356 boundary edges and V−E+F = 3 on all five. It renders perfectly and it is not a solid, so anything that needs a closed surface rather than a picture — a CAD kernel, a slicer, any fabrication file — would inherit a torn body. Welding is a reading of the archive rather than a repair, since the label itself declares the duplication; the five now carry 2,522 vertices and 5,040 facets each and are closed, and a test asserts that on the shipped files. The four decimated Gaskell meshes are watertight too, but uniform clustering can weld two surface sheets onto one edge, so their non-manifold edge counts (Phoebe 211 · Mimas 45 · Dione and Tethys 2 · Phobos 0) are a declared one-way ledger rather than a silence. A third trap the bundle sets and the numbers settle: its top-level description says “a diameter listed for each grid point” while each file’s own label says radius — reading the looser one would have produced a half-size Deimos, and the converted extents (16.0 × 12.7 × 11.1 km) match the published 15.0 × 12.2 × 11.0 triaxial figure. Vesta’s is the pre-Dawn HST figure and its credit says so. And the archive’s own ready-made Phobos OBJ is deliberately NOT the one used: it drops the first of the table’s 49,152 facets, leaving one 12,953 m² hole — 7.9 parts per million of the surface, at latitude 33.5° longitude 136.9° — the signature of a converter that skipped a line past the header; reading the table ourselves does not inherit it. Deimos’s anti-Mars hemisphere is constrained by limb data only, which no conversion can fix. REFUSED, with reasons on the exception list: Ryugu (PDS SBN archives no Hayabusa2 shape product; the JAXA-hosted sources need their own licence read); Dimorphos (the registry node now exists, so what is left is a curated key and a published post-impact pole); and the Ernst et al. 2023 models of both Martian moons, which are the best that exist — Phobos at 18 m ground sample and 36 m accuracy — and whose data pages carry “all rights reserved”: the paper is CC BY, the files are not, and the promised PDS deposit has not happened yet	NASA PDS Small Bodies Node (sbn-psi shape-models mirror + sbnarchive.psi.edu + pds-smallbodies.astro.umd.edu holdings) — Gaskell 2020 (10.26033/xzv5-bw95) · Thomas et al. 2021 (10.26033/g5e0-kh52) · Thomas 1993 (10.1006/icar.1993.1130) · Daly/Barnouin et al. 2023 (10.1038/s41586-023-05810-5) · Duxbury et al. 2004 (10.1029/2004JE002316) · Thomas et al. 2013 (10.1016/j.icarus.2012.02.037)	Public domain (NASA PDS archived data; the 67P model is ESA/Rosetta data the SBN itself openly redistributes — credited ESA/Rosetta on the record); vendored derivatives carry source + decimation counts in their credits
DSN Now	Measured deep-space network fixes (validation layer)	NASA/JPL Deep Space Network public feed	Public — snapshot 2026-06-10
Live pipes (watch-live)	For a handful of still-streaming objects, a prominent "▶ watch live" that fetches the object's OWN current published output on demand — the Sun now (SDO AIA 193 Å, SOHO LASCO C3), Earth now (GOES-19 ABI GeoColor), the latest Mars raw camera image (Perseverance, Curiosity) and the Deep Space Network antennas talking right now. Endpoint facts only are baked; NOTHING is stored or proxied — your browser fetches the provider directly when you click (§3.1)	NASA/SDO · SOHO (ESA & NASA) · NOAA STAR/GOES · NASA/JPL Mars raw images · NASA/JPL DSN Now	US-gov public domain (SOHO joint ESA/NASA, attribution shown); fetched live, per provider terms
In the news (live rail)	from the Search & AI palette ("🗞 in the news"): a compact rail of current spaceflight headlines — METADATA ONLY (title · news site · date · summary snippet); article images are deliberately not shown and full text stays at the source (every item is an external link to its original site). Where an article names a Launch Library 2 launch the registry’s own upcoming-launch marker carries (an EXACT shared LL2 id — no guessing, no language processing), the row gains a "voir dans le registre" teleport chip: the sim clock jumps into the announced launch window and the camera flies to the pad marker. Articles matching nothing stay honest link-outs, never force-joined; the joined census is stated (N articles · M joined). Fetched by YOUR browser on the explicit open gesture only — one modest page per session, never at boot, never polled; ⟳ is the only re-fetch; offline or unreachable degrades to an honest note (nothing is cached or proxied, § zero-server). Live content — no permalink key by design	Spaceflight News API v4 — The Space Devs (spaceflightnewsapi.net; article metadata + Launch Library 2 launch cross-references; each story credited and linked to its original news site)	The Space Devs free-use terms (the same footing as the Launch Library 2 ingest: "use the data in any way, shape, or form"; no raw forwarding without added value — the registry join is the added value; attribution encouraged and given); per-article text/images remain the original site’s — linked out, not reproduced
UN Register (UNOOSA)	24,866 index entries of the Online Index of Objects Launched into Outer Space — joined by COSPAR designator onto the objects above, yielding the three-way registration facet: declared (registered with the UN, document symbol kept per object) · undeclared (in the index, no registration submitted) · unknown (payloads the index does not list). An annotation layer: it adds no objects of its own. Search/filter with `un:declared`, `un:undeclared`, `un:unknown`	United Nations Office for Outer Space Affairs, Online Index of Objects Launched into Outer Space (Registration Convention / GA res. 1721B) — snapshot 2026-06-11 via the index's own search service	© United Nations — UN terms of use
Manmade Material on the Moon (NASA)	694 catalogued line items left at the six Apollo landing sites (tools, experiment hardware, flags, symbolic objects), each a child of its lunar module — deployed at the landing instant, positioned at the site, co-rotating with the Moon. GCAT-tracked hardware (descent/ascent stages, S-IVB, subsatellites) is never duplicated; the catalogue’s Soviet/robotic pages list whole spacecraft GCAT already tracks. Search `db:nasa-mmm`	NASA History Program Office, "Catalogue of Manmade Material on the Moon" (2012-07-05) — snapshot 2026-06-11	US government work (public domain)
A Profile of Humanity	~90 interstellar radio messages + moonbounce transmissions, 110 cultural artifacts & intended messages (83 matched onto their carrier spacecraft), bio-footprints on Moon/Mars	Paul E. Quast — "A Profile of Humanity: the cultural signature of Earth's inhabitants beyond the atmosphere" (Int. J. Astrobiology, 10.1017/S1473550418000290) and the author's living workbook (ed. 2024-01-16)	Courtesy of the author — prototype import from the provided workbook; final license terms at publication. Cite the article.
UCS Satellite Database	~7,560 operational satellites — joined by COSPAR designator onto the catalogue objects above, annotating each with its operational purpose (Communications · Earth Observation · Navigation · Space Science · Technology Development…) and user class (Commercial · Government · Military · Civil). An annotation layer: it adds no objects. Search/filter with `purpose:observation`, `users:commercial`, etc. 7,506 of the catalogue's objects matched. A second, independent view of the operator (cf. the GCAT-derived `originator` facet — where the two sources differ, both are kept). Data paused at version 5-1-2023 (current to 2023-05-01)	Union of Concerned Scientists (UCS) Satellite Database, version 5-1-2023 — snapshot of the published tab-delimited file	Free, use unrestricted; UCS requests acknowledgement
International Space Station (curated)	One persistent navigable ISS on its operational orbit (~420 km, 51.64°, 1998→present) + a timeline of 33 events (16 assembly milestones, 11 dockings/undockings, 6 EVA milestones) as time-located `kind:iss-event` pulses parented to the complex. The modules are individually in GCAT; this adds the assembled, time-progressive complex. Search "ISS"	NASA ISS assembly elements · NASA spacewalking history · NASA EVA chronology · GCAT	Public record (US government work) · GCAT CC BY 4.0
Artificial impact sites (curated)	26 marks: 25 located human-made impact craters/scars, each a surface-fixed mark at its measured lat/lon, appearing at the impact instant: 19 on the Moon (Rangers, the Apollo S-IVB seismic impacts, SMART-1, LCROSS, GRAIL, LADEE, Beresheet, Vikram, Luna 25…), 3 on Mars (Curiosity/Perseverance descent stages, Schiaparelli), 1 on Mercury (MESSENGER), DART→Dimorphos, Hayabusa2 SCI→Ryugu. Plus one FORECAST, flagged as such and drawn in a different colour: the Falcon 9 upper stage 2025-010D, due to strike the Moon near crater Einstein on 2026 August 5 at 06:34:32.9 UTC — nothing there has been seen yet, and when it has, the predicted coordinates are replaced by measured ones. The flown object itself (GCAT_S62719) carries the same solution as its terminal fix — at the predicted point at the predicted instant, its end-of-existence corrected to 06:34:32.9 UTC with GCAT's own 06:44 decay epoch kept as an alternate (erase none). The DESCENT is now drawn, and from a different source: Project Pluto publishes no arrival azimuth, so its approach curve would have been invented and was refused — JPL Horizons carries the object (`spkid -162719`, trajectory solution `2025-010D_GA1A2_15`) and serves the selenocentric ephemeris, so the stage rides 184 sourced osculating arcs in the Moon's frame from 2026-08-04 00:00 UTC (80,005 km out) to impact instead of vanishing a quarter of a million km short. The two determinations DISAGREE and both are kept: at the Project Pluto instant the Horizons solution still has the stage 28.5 km up, reaching the ground 23.3 s later and 14.8 km away — the impact point shown is the chosen one, the Horizons instant is an `existsUntil` alternate. One limit stays on the record: the whole arc is a PREDICTION (that solution is fit to data through 2026-Jul-07). The other has been CLOSED and measured — the arc is inertial while the impact point is surface-fixed, and until the Moon was given its real synchronous rotation the two could not meet: the descent missed its own crater by 1,108 km. With the IAU/NAIF lunar rotation in place the miss is 40.4 km: 32.1 km of it is simply the altitude the Horizons solution still carries at the Project Pluto instant (the two determinations' own 28.5 km disagreement, not a model error) and 24.3 km is sideways, essentially all of it the deliberately unmodelled 0.76° Cassini precession of the lunar pole at that date. The DART row says what the impact THREW, and says it as bounds: 104 boulders were tracked in LICIACube/LUKE images over the 214 seconds from 29 to 243 s after the strike — radii up to 3.6 m, speeds up to 52 m/s, thrown into a roughly 135°×100° cone directed south, and carrying more than three times the spacecraft's own momentum, so humanity's first act of planetary defence sent away more momentum in rock than it brought in. Those are the published bounds and not a table of 104 objects: what the archive holds is image-plane PIXEL positions, while the individual 3-D states are the authors' derivation given in figures, and digitising a figure would be our measurement wearing their authority. A representative swarm is therefore a Factory product (SYNTHETIC), never one of the 104. `kind:impact-site`; each note links the high-res LROC/HiRISE image	NASA NSSDCA · LROC (Wagner et al. 2017, Icarus 283) · HiRISE · DART (Nature 2023) · DART ejecta (Farnham et al. 2025, PSJ 6, 155) · Hayabusa2 (Science 2020) · Bill Gray / Project Pluto (2025-010D impact solution) · JPL Horizons (2025-010D terminal approach)	Public record (NASA/LROC/HiRISE/JPL public domain); the DART ejecta figures under CC BY 4.0, credited on the record itself as the licence requires
Rover tracks from orbit (curated)	Orbital HiRISE (MRO) observations of the Curiosity (Gale), Perseverance (Jezero) and Opportunity (Meridiani) wheel tracks — surface-fixed marks on Mars linking the published image; the trace as a feature of the planet, not a property of the rover (per the owner). `kind:site`	NASA/JPL · University of Arizona HiRISE	Public domain (NASA/UA imagery)
Mars surface traverses & Ingenuity flights (MMGIS)	The published waypoint logs of the Perseverance and Curiosity traverses, joined onto the rover objects as dated surface positions (each kept waypoint places the vehicle on the rotating Martian ground at the sol it was there, so playing time drives it along its real path; the logs are distance-decimated to a declared cap, shape-preserving). The Ingenuity helicopter is its own `roverpaths` node: each of its flights is expanded from the log's start time, duration and maximum altitude into a simple MODELLED climb–cruise–descent profile so the hops read as flight — the profile is declared modelled, and a log row whose own timing fields contradict each other is kept as a single ground-to-ground move rather than trusted (A5). One thing a traverse carries that the launch catalogue's landing record does not is an elevation: a catalogued landing is placed at the body's mean radius, because that is all its source gives. Where the object's own traverse then measures the same place with a real elevation, the datum placement is superseded — dropped, with the traverse's first waypoint re-anchored to the catalogued landing instant, so the rover stands on the ground from the moment it lands. Until 2026-07-29 the two sat side by side and were interpolated between: Perseverance descended 2,570 m over 13.4 days and Curiosity 4,501 m over 3.1 days, through altitudes nobody measured. A placement carrying a real elevation is a measurement and is never dropped	NASA/JPL-Caltech — Mars MMGIS waypoint feeds (mars.nasa.gov/mmgis-maps, M20 + MSL) — snapshot 2026-06-12	Public domain (NASA/JPL-Caltech)
Object class & nationality	The generic GCAT 'payload' kind is split by psatcat Category into `kind:spacestation` (760, Mir/Salyut/Tiangong/ISS…), `kind:probe` (planetary/deep-space, Category PLAN) and `kind:telescope` (Category AST). Each object's GCAT State code is mapped to a human nation name on a `nation` field — the `nation:france` / `nation:usa` search facet	GCAT psatcat Category / State codes (J. McDowell)	CC BY 4.0 — snapshot 2026-06-09
Concept missions (projected)	A gated, PROJECTED scenario overlay (never baked into truth) drawing a dashed leg-chain through real catalogue stars. First instance: the owner's photogravitational-lightsail concept (Sol → Sirius A → Procyon A; laserless launch 4.6%c · Sirius-A hub ≤12%c · afterburner ≤9%c · chained to 17%c), shown humbly as "a concept"	A. Normier (Paris-Sorbonne MII, 2020), after Heller & Hippke 2017 (10.3847/2041-8213/835/2/L32) & Heller, Hippke & Kervella 2017 (10.3847/1538-3881/aa813f)	Author's concept; underlying papers public (DOI). Modelled/projected, origin ARTIFICIAL
Future missions (curated)	A small set of announced / en-route deep-space & crewed missions — Europa Clipper, JUICE, Dragonfly, the Artemis crewed lunar landing, the Mars Sample Return concept — shown as PLANNED-MARKER annotations placed at the centre of their target body (Moon · Mars · Europa · Ganymede · Titan). No trajectory is invented: an unflown mission has no real position, so it is a labelled marker on its destination, not a faked path. `existsFrom` starts at the announced launch window (or year) only where one is publicly stated; an unfixed date is omitted, never guessed (A5). Each record carries its cited status/window, the Launch Library 2 launch UUID where one exists (the dedup key: a curated mission the LL2 upcoming manifest also lists renders ONCE — the curated marker wins, the LL2 pad marker is suppressed at bake, census on the layer card), and — for launched missions — the flown spacecraft's COSPAR cross-reference. Published predicted trajectories ride the flown object (see "Predicted trajectories (Horizons)" above), never this marker. Quarterly manual curation. Search `kind:mission`	Public NASA / ESA mission-announcement pages — Europa Clipper (science.nasa.gov/mission/europa-clipper) · JUICE (esa.int/juice) · Dragonfly (science.nasa.gov/mission/dragonfly) · Artemis (nasa.gov/artemis) · Mars Sample Return (nasa.gov/mars-sample-return) · launch cross-refs: Launch Library 2 (The Space Devs, checked 2026-07-21)	Public record (mission announcements — no DOI; the stable mission page is the citation); LL2 ids under The Space Devs free-use terms

Refresh discipline: sources are re-fetched into new dated snapshots, reviewed, and re-baked offline — the application itself never calls external services. Snapshots are never deleted or overwritten.

3D / DEM inventory — what surface is actually shown, body by body

Asked “what do you show of this world’s surface?”, the honest answer is a number, and it is not written anywhere: it falls out of the tile pyramid a map is served on, crossed with the body’s own published radius. So the table below is derived from the registry’s catalogues rather than typed — a new map, a deeper zoom level or a re-measured radius moves it without anyone editing this page, and a guard fails the build if this page and the catalogues stop agreeing.

The inventory — 1,066 natural bodies the registry can draw as a sphere: 13 with a real shape · 3 with measured relief · 11 with real imagery on a smooth sphere · 6 with a generic texture only · 1,033 featureless

Five rungs, in order of how much of the real body they carry. shape — a body-fixed mesh, the true figure rather than an ellipsoid. relief — a global elevation model, so the sphere is displaced by measured heights. map — real imagery on a smooth sphere. texture — a generic full-body image. sphere — nothing but a radius.

Two resolutions are reported and they answer different questions. The ground sample is how finely the surface is sampled, at the deepest level the service is observed to serve; the height step is how finely elevation is resolved, which for these products is coarse because NASA Trek serves them as 8-bit greyscale images with no elevation metadata — the grey→metres stretch is derived per body and validated against that body’s real numeric DEM (Mars MOLA MEGDR, Moon LOLA LDEM, Venus Magellan GTDR), leaving a step of tens of metres.

body	what is shown	ground sample	relief	notes
Mars	relief	325 m/px	1.30 km/px, 55.93 m per grey level	5 landing-site overlays down to 0.317 m/px
Moon	relief	167 m/px	666 m/px, 77.42 m per grey level	6 landing-site overlays down to 0.325 m/px
Phobos	shape	4.32 m/px	—	Gaskell stereophotoclinometry figure and the sharpest global map in the registry — and see the orientation note below
Vesta	shape	25.2 m/px	—	pre-Dawn Thomas figure, stated on the record; orientation note
Ceres	map	45.0 m/px	—	orientation note
Enceladus	map	96.7 m/px	—	orientation note
Mimas	shape	152 m/px	—	orientation note
Dione	shape	215 m/px	—	orientation note
Mercury	map	234 m/px	—	—
Earth	map	244 m/px	—	daily true-colour available as a dated layer
Tethys	shape	407 m/px	—	orientation note
Rhea	map	586 m/px	—	orientation note
Europa	map	599 m/px	—	orientation note
Callisto	map	924 m/px	—	added 2026-08-30 with its pole — it was the one Galilean drawn as a bare sphere while the other three carried mosaics
Ganymede	map	1.01 km/px	—	orientation note
Iapetus	map	1.13 km/px	—	orientation note
Io	map	1.40 km/px	—	orientation note
Venus	relief	2.32 km/px	4.64 km/px, 57.679 m per grey level	Magellan radar mosaic over Magellan topography — the one DEM here served at its source’s NATIVE raster size, so nothing is decimated and nothing saturates
Titan	map	3.95 km/px	—	—
9P/Tempel 1 · 67P/Churyumov-Gerasimenko · 81P/Wild 2	shape	—	—	mission shape models; a decimated display figure, not the science mesh
Deimos · Hyperion · Phoebe · Janus · Epimetheus	shape	—	—	a measured FIGURE and no map at all: an outline with nothing to drape on it. Four of the five come from 5° radius grids, Phoebe from a stereophotoclinometry table
Sol · Jupiter · Saturn · Uranus · Neptune · the Milky Way disc	texture	—	—	a generic full-body image; no positioned mosaic exists here (the gas giants have no surface to map)
everything else — 1,033 bodies, Triton, Charon, Pluto and the Uranian moons among them	sphere	—	—	a radius and nothing more. Deliberate: where no real published mosaic exists, the body gets NOTHING rather than a plausible wrong texture

The orientation note, and it is the important one — the count that stood here is now zero, and the sentence says what that cost and bought. A streamed map is draped on a group oriented by the body’s chain rotation times its spin. Thirteen of the nineteen mapped bodies used to publish no rotation at all, so that spin was the identity and their imagery was pinned to the parent’s axes rather than to the ground it depicts — not merely un-animated, in the wrong frame. 0 of the 19 mapped bodies are in that state on the shipped registry. Phobos was the sharpest case in both senses and is the sharpest measurement of the repair: it is tidally locked, so its sub-Mars point sits at longitude 0 by the definition of its body-fixed frame, and before the ingest landed that point swept 310.3° of the map in one 7.65-hour orbit. Measured on the shipped registry it now holds to 3.30° across the same orbit — and that residue is not slack: it is the optical libration in longitude that Phobos’s own eccentricity demands, 2e = 3.46° peak-to-peak at e = 0.0151, so the map is not merely stationary but wobbles by the amount the orbit says it must. Deimos, at a hundredth the eccentricity, holds to 0.15°. The published IAU rotational elements were the fix, and they are a prerequisite for any real shape model: a mesh hung on a frame that is not the body’s own would be a claim about which side faces where. All thirteen carry those elements — pole, rate and prime-meridian phase, reduced from the NAIF kernel with the declined terms measured rather than assumed. Eleven of the thirteen are tidally locked, and that gives the reduction an external check no fitting can fake: a synchronous moon’s published rotation rate must reproduce its published orbital mean motion, two numbers measured by different means and never adjusted against each other — all eleven agree to better than 0.14 %. Titan is a thirteenth case and a different one: it was already oriented, on a placeholder axis, and drawn 28.05° from its published pole.

What the frozen axis actually costs was the opposite of the guess, and the measurement is the reason it is stated here. These poles carry secular rates like the planets’, so a fixed axis was expected to drift steadily — but the published elements also carry periodic terms, and those dominate. Held frozen at J2000 over 1950–2100 the worst excursions are Mimas 3.06°, Tethys 2.19° and Iapetus 1.56°, and the two kinds are distinguishable by where they end up: Iapetus finishes at its worst (a one-way secular drift), while Tethys comes back to 22 % of its own maximum — it oscillates. A single sample date therefore ranks these bodies wrongly, which is exactly how the first version of this check got Iapetus first instead of third. One trap is worth naming for whoever reads the kernel next: a rate quoted in right ascension is not a rate on the sky — near a pole at declination 75° a degree of RA subtends only 0.26°.

And the same reading found a larger one, in the planets — now repaired. Every spin axis used to be built from the body’s obliquity alone, tilted about a fixed direction, which pins every pole’s node longitude to Earth’s. Earth therefore came out exact and nothing else did: measured against the published IAU poles, the Moon was 1.6° off, Mars 36.8°, Titan 28.1°, Neptune 50.8°, and Venus and Uranus were drawn very nearly upside down at 176.6° and 167.8°. Nothing on a body moves relative to anything else on it — the whole surface turns together — but it turned wrong against the sky, which is where the Sun is: on Mars the sub-solar latitude, the season, read +0.2° at J2000 where the published pole gives −26.5°, the southern summer. Ten bodies now carry their published pole and prime meridian from the NAIF kernel, and their axes precess where the kernel publishes a rate — a cone whose period is derived from those rates rather than looked up, which for Mars gives −179,100 years against the ~171,000 the literature quotes, and which removes 99 % of the error a frozen axis accumulates over a millennium.

Two things caught along the way are worth stating, because both are about how a check can fail to be one. The pole guard compared the drawn axis against a table of published numbers — and the ingest read the same numbers, so when both dropped the same term from Mars’s series the guard read 0.000° while the axis was 1.60° wrong. What caught it was the body’s published obliquity, which comes from a different measurement entirely and refused to agree. And the dropped term was not really periodic: its argument advances 0.5° per century, a 714,000-year period, so over any span this registry is scrubbed across it is a constant wearing a periodic shape. Reading it as something safe to decline was a transcription error, not a declared approximation.

Ten bodies stopped being ellipsoids on 2026-08-30, and the boundary that remains is a real one. The shipped registry now draws a measured figure for 13 bodies: three comets, Vesta, and nine moons — Phobos, Deimos, Mimas, Tethys, Dione, Hyperion, Phoebe, Janus and Epimetheus. Every one of them is under 600 km in radius, and nothing larger has one. That is not our threshold: it is where the literature stops publishing shape models, because above it a body is round enough that an ellipsoid is a fair description and below it the figure IS the science. Phobos makes the point — its long axis is 1.16× its mean radius, so the ellipsoid it used to be drawn as was not a small simplification. The larger bodies’ poles are fixed now too, so what stands between a planet and a mesh is no longer the frame; it is the literature.

The batch created a contradiction, it was counted rather than left to be noticed, and the count is now empty. A mesh is drawn in the body’s own frame; an unoriented mosaic is pinned to its parent’s axes. Five bodies briefly had both — Phobos, Vesta, Mimas, Tethys and Dione — so on those the figure and the map disagreed about which way the body faced. The contradiction did not exist while they were bare spheres, because a sphere has no face to contradict, and it was written down as a ceiling the next rotation ingest could only lower. That ingest has landed: no body carries a shape and an unoriented map together. The number is worth keeping in view rather than deleting, because the same collision returns the day a figure arrives for a body whose rotation is still unpublished.

Reading the scene

Default style by tags: NATURAL bodies in white/grey · ARTIFICIAL objects in magenta · SYNTHETIC content dim and translucent. Trails are analytical orbit evaluations (not recorded histories), cast in a frame you can elect: a multiphase spacecraft can draw its full history since launch, data seams between navigation arcs are reconnected where the arcs genuinely converge (a seam that never converges stays an honest break), and an object resting on a surface draws no orbit-frame trail — its honest trace is the co-rotating ground track (toggleable; permalink tss=0 restores the legacy behaviour). Trails can be graduated like school TIME rulers — major tick strokes at round calendar steps (a day, a month, a year, a century…) plus one finer minor subdivision (a year with quarters, a month with 5-day blocks), both picked from the visible span; an optional date legend labels the selected trail's major ticks ("1981 · Jan · Feb … 1982" — a year tick prints its year, sub-day ticks the time, deep-time ticks a ±ky/My/Gy offset from J2000) — and a trail-length slider can impose how much time every trail draws (1 hour → 5 Gy — deep enough to watch Andromeda's approach; automatic by default). The selected object's trail can also be composed in MORE THAN ONE frame at once — the "trail frames" row (Visualisation) names the elected composing frame and a (+) adds up to 3 of the object's own ancestor frames (its real gravitational chain, never an arbitrary frame): each added frame draws one more, dimmer trail of the same object re-expressed in that frame (e.g. Earth's heliocentric orbit ring alongside the same year of motion dragged along the Sun's galactic drift in the Milky Way frame); every honesty guard applies per frame, and the set is session-only, clearing when the selection changes. Once a length is imposed, a slim slider band also surfaces on the LEFT edge of the main screen (mirroring the right-hand zoom band, and reaching all the way to ~5 Gy — long enough to watch Andromeda drift in) so the span can be adjusted without opening a panel; its "auto" control releases the band and restores the automatic election. At a scene anchored on a periodless node (a star system view, a galaxy view) each closed orbit draws its one nearly-closed loop — a full clean ring — rather than an arbitrary one-year stub; at a planet-anchored scene the planet's own period stays the shared time window. The trail budget (an FPS-governed cap, as are the ≤128-per-badge trail flags) is spent only on objects that actually draw — surface-resting items never consume a slot. When the selected object draws no trail, the trail settings (Visualisation) state the honest reason — resting on a surface, no measured velocity in the registry (no trail is invented), outside its existence window, or hidden by a filter badge. The dynamic legend picks the key objects at your current scale agnostically — no hardcoded favorites; landing on a planetary system labels its star and planets. The 🔗 button saves the entire view state into a URL you can share.

Provenance is visible in the scene itself. Every scene label of a non-observed element carries an amber sub-line naming WHY it is not an observation (GENERIC FIT · SYNTHETIC · MODELLED · PROCEDURAL SHAPE · MODELLED ORBIT · ± ALT SOURCES · GENERIC TEXTURE · MANUAL FIT), on by default (vp=0 hides it). A 3D model on stage carries its own provenance card — the model's source, plus a caution where its attitude or texture is procedural or hand-fitted (mil=0 hides it). Where a mission's model is known to change over time, the model follows curated time windows (the launch stack at launch, the rover only after touchdown); a window for which no honest mesh exists shows nothing rather than a wrong mesh (mp=0 disables the swap). Solar panels on a small set of reconciled models track the Sun and dish antennas aim at Earth — a geometric articulation, declared on the model card, not telemetry.

Search grammar. Search accepts free text plus facet filters — kind: · origin: · prov: · db: · has: (incl. has:landed / has:orbiting resting-state tags, has:operating — the object's catalogued operating window is open at the date on screen, the only activity tag that reads the clock; the older has:active and has:silent are a catalogue status instead (whether the record was ever closed), so they answer a different question and at a past date can name a completely different set; an object with no operating record is unassessed, never reported as switched off — has:nakedeye — visible to the naked eye in Earth's sky: stars by measured magnitude ≤ 6.5 plus curated facts such as the planets to Uranus, M31/LMC/SMC, witnessed supernovae at peak and the ISS; absence = unassessed, not invisible — and has:name — carries a deliberate human-given name rather than only an automatic catalogue designation (HIP 12345, Gaia DR3 …, PSZ2 G…, a 1996 JG provisional, hardware serials); a serialized program name such as Starlink 2113 counts as named, a Bayer/Flamsteed form such as alf Ari does not, and factory-generated labels never qualify) · un: · nation: · originator: · purpose: · users: · site: · lv: · outcome: · band: and profile: (the EM band(s) and emission profile of a human transmission) · containment: · vector: · under: (everything inside a named gravity well). Facet keys are case-insensitive; filters AND-compose with each other and with the free text. Numeric RANGE/COMPARATOR facets over a declared whitelist of measured fields — radius: · mass: · temperature: · distance: · launched: · mag: — accept > < >= <= and the inclusive a..b range, with unit suffixes (km/m · kg · K · pc/ly/au; no suffix = the SI base unit). Examples: kind:star mag:<=6.5 distance:>100pc — naked-eye-bright stars catalogued beyond 100 parsecs; launched:<1970 — hardware with a catalogued launch epoch before 1970. A record that never measured the field matches no comparator (unassessed — absence is not a value); an unlisted numeric facet is not grammar and degrades to free text.

Real position, or the sky as seen? — what "at this instant" means

The scene shows every object where it is at the simulated instant, in one shared slice of time. It is not a picture of the sky as seen from Earth: nothing here is delayed by the travel time of its light. Look at a star 300 light-years away and you are looking at a modelled present position, while the light actually reaching a telescope tonight left it three centuries ago.

That choice has a consequence at ingest, and the registry acts on it. An astrometric catalogue records where a star appeared — its position is already old by the light-travel time when it is measured. So a star's placement is displaced along its own measured velocity by two terms: the offset between the catalogue's reference epoch and J2000, and the light-travel time d/c. On the shipped data the second term is the larger of the two — a median of about 20 arc-seconds against about 3. A star with no measured velocity is not displaced at all; nothing is invented to make the arithmetic work.

Because different catalogues answer the question differently, every object's record carries a line saying which regime it is in — computed from a dynamical model at that instant · measured then propagated · measured and held because no motion was ever measured (which is an absence of data, not a claim that it is at rest) · placed at a modelled comoving distance from redshift (a present-day distance in a cosmological model, not a light-travel distance) · or an arrival epoch, where the date on the record is when the light or signal reached Earth rather than when the event happened.

Transients used to be in that last class and no longer are. A supernova or a gravitational-wave merger is now placed when it happened: the travel time is removed so the event sits in the same absolute slice as everything else. SN 1054 explodes around 5470 BCE, and 1054 is when its light reached Earth — a measurement that is preserved, not discarded. For a supernova the removed span is simply distance/c. For a merger billions of light-years away it is the modelled lookback time, which is not the luminosity distance over c: for GW190521 that naive figure gives 17.3 billion years, older than the universe itself, against a real 7.1. An event with no published distance is not re-dated at all, and says so.

One consequence is worth stating plainly: a gravitational-wave marker used to appear only from the date it was detected, and now exists throughout human history — because the merger does.

An optional observer view — showing the scene as one chosen observer sees it at a chosen moment, every object pushed back to where it was when the light now arriving left it — is designed and not yet built. One thing it will make checkable: seen from Earth, the corrected star positions must reproduce the raw catalogue sky exactly.

The convention, stated formally

The plain-language paragraphs above describe a choice that has a precise technical statement, and the registry's data rests on it, so it is set out here in full with its sources.

Two time arguments. Let t_c be coordinate time in the BCRS (Barycentric Celestial Reference System, IAU 2000 Resolution B1.3), realised with TCB, and let t_a be the time of light arrival at the solar-system barycentre. A statement "the source is at b at t_c" picks a point on a simultaneity hypersurface of the BCRS and is not directly observable. A statement "the source is seen in direction u at t_a" is an observable, and it refers to the source's coordinate position at the retarded instant t_c = t_a − |b|/c.

Which one the catalogues use. Gaia adopts the standard model of stellar motion — uniform rectilinear barycentric motion, b(t) = b_ep + (t − t_ep)v — after Lindegren et al. (2012), on the relativistic observation model of Klioner (2003). The Gaia Data Release documentation, Chapter 4 Astrometry, §4.1.4 Standard model of stellar motion, states its time argument explicitly:

"Equation [4.2] ignores the finite speed of light. In principle, the barycentric coordinate direction measured at time t corresponds to the barycentric position of the source at the time t − |b|c⁻¹, several (or many) years earlier."

"the standard model is therefore parametrised by quantities representing the position and motion of the source as they appear from the SSB at a given time. Thus, the time argument in ū_B(t) must always be interpreted as the time of light arrival at the SSB, not as the time of light emission from the source."

Gaia's reference epoch is therefore t_a, not t_c; its published position is the apparent, retarded direction. The same standard model governs Hipparcos and the XHIP compilation derived from it (ESA SP-1200, Vol. 1, §1.2.8).

Why that is not a defect in the catalogue. For uniform rectilinear motion the light-time offset is exactly degenerate with the position parameter: b(t_a − d/c) = [b_ep − v(d/c)] + v(t_a − t_ep), which is again uniform rectilinear motion with the same velocity. No astrometric fit can separate the two terms, so no catalogue removes it — and removing it is not a correction of the catalogue but a change of convention applied to it. The rigorous transformation between apparent and kinematic parameters, light-time included, is given by Butkevich & Lindegren (2014).

What this registry does. A node's stored position is declared to be its position at coordinate epoch J2000, and the scene is a simultaneity slice in t_c. Converting a catalogue row into that convention takes both terms, pos₀ = b_cat + v_hel[(J2000 − t_ep) + d/c], the first reconciling the epochs and the second removing the light-travel delay the standard model deliberately leaves in. Measured over the 56,399 stars carrying a measured heliocentric velocity, the second term subtends a median 19.6″ (p90 46″, p99 84″, max 353″) against the first term's median 2.9″; its angular size is v_⊥/c independently of distance. Using the catalogue's apparent velocity in place of the kinematic one introduces a relative error v_r/c ≲ 2×10⁻⁴, i.e. ≲ 0.004″ on a ~20″ term, so the first-order form is used and its omission is stated rather than assumed.

Epistemic status. b(t_c) is not observable. The default view is therefore a model — the apparent path re-expressed under a declared simultaneity convention — while an observer view is nearer to the measurement, since from Earth it returns the catalogue's own directions. Neither is privileged; each is a stated convention, and every record says which one it is in.

Sources: Lindegren et al. 2012, A&A 538, A78 · Klioner 2003, AJ 125, 1580 · Butkevich & Lindegren 2014, A&A 570, A62 · Lindegren et al. 2021, A&A 649, A2 · Gaia Data Release documentation, Ch. 4 §4.1.4 · ESA SP-1200 (Hipparcos), Vol. 1 §1.2.8.

When the drawn orbit is not a prediction

Most orbits here are propagated as a single fixed two-body ellipse from the catalogued elements. For the great majority of objects that is an excellent model. For a minority it is silently worthless, and until now nothing on the record said which was which.

The case that made it concrete: the Falcon 9 upper stage that is due to strike the Moon on 5 August 2026. On its stored ellipse it never comes within 39,000 km of the Moon — its apogee stops 15,500 km short of the lunar orbit. What drives it into the Moon is the repeated lunar tug that a two-body model has no term for. The drawn path and the real path are different curves.

So every orbit is now screened against the massive natural bodies sharing its frame: if it reaches inside a few Hill radii of one of them, the record says so, in the honest form — this is where a Kepler orbit would go, not a prediction of where the object will be. Across the shipped registry that flags 4,558 of 52,847 screened orbits (8.6%), of which 184 are Earth-orbiting objects that cross the Moon's reach. Reassuringly, the class it flags hardest is the near-Earth asteroid and comet population, which is defined by crossing — the screen agreeing with the catalogue.

Two limits are stated rather than discovered later. Crossing makes an encounter possible, not certain: inclination and phase can keep a crossing orbit clear for a long time. And the screen is gravitational only — solar radiation pressure on a light, large object matters enormously (it does for this very stage) and nothing here can see it, because the registry does not carry the area and mass it would need.

Two search facets follow from this, and they are deliberately kept apart. has:crossing is our modelled geometry — the object's orbit reaches inside a massive body's grip on some leg of its trajectory. has:hazard is a catalogue's own classification, reached with far more than we hold: NASA/JPL calls an asteroid potentially hazardous when its minimum orbital intersection distance with Earth's orbit is under 0.05 au and it is bright enough to be large. Merging the two would launder someone else's careful ruling into our arithmetic, so they stay separate rows with separate words.

Potential impactors

Every object with an orbit now carries a third neighbourhood group, beside its children and its siblings: what could one day run into it — natural and artificial, counted, each name a click away.

The test is the same geometry read symmetrically. Two bodies sharing a primary can meet only if the bands of radii they sweep overlap, each widened by its own reach — a massive body's gravitational grip, or failing that its physical size, or failing that nothing at all rather than a guessed size. Over the long run that is the right test and a near-term one would be the wrong test: nodes and apsides precess at different rates, so the angular geometry eventually sweeps through every configuration, while the range of radii a body visits does not change.

What the list is. It is what is not excluded. The exclusion is the strong half — two bodies whose bands are disjoint cannot collide while these elements hold — and the membership is the weak half. The registry holds two-body orbits, so it can rule an encounter out and never rule one in; there is no probability here and none is offered. Where a catalogue has already ruled, that ruling appears as its own line, in its own words.

Where the criterion stops, and why it stops there. The natural question is whether passing into a body's Hill sphere is the gate. It is the gate for one thing and the wrong gate for another, and the distance between the two is the whole argument. The Hill radius is where the target's gravity beats its primary's tidal field — crossing it is the gate for the two-body model has stopped being predictive, a statement about our model. An impact needs the target's physical radius, the disc that must actually be struck, enlarged a little by gravitational focusing. For the Earth those two are about 1.5 million km and 6,371 km: the region where the model already fails is some two hundred times wider than the target, forty thousand times its area. So crossing the Hill sphere tells you the answer is no longer computable from these elements, and almost nothing about whether the object hits.

The verdict therefore has two values and no third. Excluded: the bands do not overlap, so no collision is possible while these elements hold — provable, and the strong half. Open: the bands overlap. Where the object reaches inside the target's gravitational grip the record marks it ⌖ and says the sharper thing — the answer is not merely unknown, it provably is not in this data. There is deliberately no "likely". Settling it needs an n-body integration with a covariance, which is what NASA/JPL's Sentry publishes; until such a source is ingested, no number is offered.

The criterion is sharp rather than permissive, which is what makes it useful: two exactly circular orbits five kilometres apart are excluded, whatever their phasing. A relation that answered "possible" there would put every satellite in every other satellite's list.

Measured on the shipped registry: Earth has 2,788 natural candidates (the near-Earth asteroid population) and 61 artificial; the Moon has 205 artificial — objects in Earth orbit whose paths reach lunar distance; the ISS shares its band with 5,345 catalogued objects; Phobos has none, and says so. Names are ranked by closeness in orbital size, because two bodies on nearly the same semi-major axis share a shell — that is a proximity order, not a risk order.

Asking about impacts, and about when

Two search facets and a range make the impact questions askable, and they compose with everything else. has:impact is the set of objects whose catalogued life ENDS by arriving at a body — a recorded impact, an orbit that decays into re-entry or onto an airless surface, or a launch that fell straight back. ends: is the calendar year that happens, the counterpart of launched:. So ends:<2027 has:impact is "about to hit something", and has:impact under:Moon is everything whose life ends at the Moon.

The distinction that makes this honest: a record whose existence merely STOPS is not in either set. Measured on the shipped registry (2026-07-28): of the 33,808 records whose catalogued life has an end, 8,904 merely stop being recorded — 8,435 of them modelled solar-wind cones, plus launch sites, detonation marks and other dated windows. Reading the raw end date would have made "objects about to impact" mostly a list of expiring CME markers. The classifier reads the last leg of the trajectory instead: 89 catalogued impacts and 23,018 orbits that end at a body (together 22,978 at Earth, 77 at the Moon, 25 at Venus, 20 at Mars), plus 1,797 launches that fell back without reaching orbit. An ending ORBIT is an arrival; a bare window is not.

Tours can walk time, not just space. Ask the AI for a tour and it can anchor each stop to that object's OWN end date, with an offset — "walk me through the objects ten days before their impact on the Moon" stands the clock ten days before each one in turn, in date order. An object with no catalogued end date is still toured; the clock is simply left alone rather than moved on a guess.

The moment is marked. A catalogued impact or re-entry now draws a brief flash where and when it happens (permalink if=0 turns it off). What is real is the WHEN and the WHERE; the size and duration are chosen to be visible — an actual impact flash would be far too small and far too brief to see at any scale this scene is viewed at, and the page says so rather than letting the mark imply a measurement.

Moving through time and lives

Time is a dimension, and so is an object's life. Any phase chip or history moment on an object's record is a teleport: the clock jumps into that leg's time window and the scene re-parents to the era — a Mars rover sits under its launcher at launch week, under the Sun in cruise, on Mars after landing. The record marks WHERE in the life you are ("now k/N" on the phase row), and 35,036 catalogued objects carry a "separated from" lineage row — the parent object at the separation epoch, itself a teleport.

Guided tours. Five curated walks (a grand tour of the solar system · humanity's footprint · Apollo end-to-end · the Mars robots · the deep sky) step through registry objects at documented mission dates, with short captions; every stop is pinned against the shipped data.

Close-approach watch. A panel (permalink apw=1) lists the catalogued close passes of Potentially Hazardous Asteroids nearest the current clock, each row a teleport to the pass. The list is of catalogued passes — a lower bound, never a completeness claim.

Passes overhead. A panel (permalink pass=1) predicts when a satellite crosses the sky above a ground location YOU provide — the ISS as the curated default, plus the selected object when it is an Earth orbiter whose baked elements carry a current (Celestrak GP) measured phase; an orbiter without current elements says so honestly instead of guessing. Each pass row (rise · culmination · set, local + UTC, elevation + compass azimuth) is a teleport to the culmination. Your location is requested only when you press the button (or typed manually), is stored on this device only, and by design never enters a share link — the pass key carries panel visibility alone. Predictions are geometric horizon passes computed from the baked mean orbital elements — a dated snapshot, not live tracking: approximate (minutes-level near the element epoch, degrading over weeks), with no atmospheric refraction and no visibility/brightness claim.

Measuring distances. A 📏 measure control (Visualisation stage) lets you pick any two objects — the next two selections — and draws a ruler line between them with a live readout of their straight-line separation at the sim clock (auto unit + light-travel time), computed through the same position compose that renders the scene. Honesty rules: an object that does not exist at the date is refused, never placed; a spacecraft outside its dated trajectory data is flagged as riding its last held data arc; a carried line-of-sight position uncertainty (the distance-gated star layers' σ_d) is stated per endpoint, never blended into a false combined precision. A connected AI answers the same question by name — "what is the distance between Mars and Pluto in 1981?" — via its measure_distance tool (a bare year measures at mid-year and reports the min–max range over that year). The ruler is a live gesture tool and deliberately has no URL parameter.

Audience personas. The menu (the brand dropdown, or the phone ☰ sheet) carries a compact persona row — kid · classroom · researcher · vr-demo — each a named bundle of existing display switches (UI level, labels, trails, factory mode, star colours, twinkle, cinematic grade, eclipse shadows, ±1σ and provenance sub-lines, constellation figures, analytical panels). Applying one works exactly like loading a shared link: the view settings are first restored to their defaults, then the bundle's parameters are applied — nothing new is stored, and your selection, camera and clock stay where they are. The ⟲ reset button restores the default configuration. A connected AI can do the same by voice via its persona option.

Sharing & URL parameters

The 🔗 permalink serializes the whole view into the URL hash — defaults are omitted, unknown keys are ignored, and each parameter is applied best-effort so an old link keeps working. For reference, the complete key registry (defaults in parentheses):

Keys	What they carry
`org` · `sel` · `fr` · `cd` · `cy` · `cp`	identity + camera: active frame, selection, trail frame, camera distance/yaw/pitch (applied last)
`t` · `rate` · `play` · `vpx` · `aw` · `win`	time: sim clock (JD), manual rate, playing, auto-speed target, adaptive time window on/off (`aw=0`), manual window bounds
`tn`	“Now” as an INTENTION rather than a value. A shared link normally carries an absolute instant (`t`), which is the only correct encoding for a link that points at a dated event — an impact, an eclipse, a mission window — and the wrong one for a view that meant the present: opened tomorrow it shows yesterday’s sky, and the objects whose whole point is being on time (the station, a pass prediction, a live feed) go stale in silence. So `tn=1` travels beside `t` and, applied after it, re-reads the clock at the READER’s present — paused, it rests there; playing, it starts there. Absolute stays the DEFAULT: `tn` is emitted only when the clock both was PUT at the present by an affordance that means it (the timeline’s NOW, go live, a “today” scene) and still IS the present when the link is written, so a clock scrubbed to a chosen date never travels as “now” however close that date is to today, and a view that has since played away from the present travels as the instant it actually shows. `t` is still written underneath, so a link minted before this key existed — or read by anything that ignores it — lands on exactly the same instant it always did. The timeline’s NOW control lights up precisely while a link would travel relative, so which clock is being shared is visible before it is copied; a live-share stream strips the mark, because a stream mirrors the lead’s clock rather than each follower’s.
`tr` · `tfh` · `tsp` · `trc` · `tss` · `tef`	trails: mode (off/auto/max), full-history, traverse spline, seam reconnection, surface-trail suppression (each `=0` to disable), and multi-frame trails `tef` (the selected object's extra ancestor frames, `~`-joined ids — its motion re-composed in several frames at once; bound to the selection, so it reproduces alongside `sel`)
`fr2`	second trail frame (a node id; default none) — the ＋ after the tree's frame grip elects one, by default one level up the gravitational tree, and the whole drawn trail population is re-composed in BOTH frames at once (the two line sets share one governed ceiling, so frame 2 never doubles the budget). The frame-2 grip drags exactly like the first; an id the bake cannot resolve, or one equal to the primary frame, is refused and the key degrades silently.
`to`	orbites (`to=1`) — each drawn trail is cast instead as its CLOSED FORM: the orbit in its own phase parent's frame, the one frame where a bound trajectory closes. Objects with no closed form are refused by name, never faked — hyperbolic arcs, surface-resting hardware, drift-only stars and geometry-less records — and the trail menu's counter states how many drew and how many were refused, class by class.
`tlj`	Retired. The launch-pad trail join was removed on 2026-07-30 — it drew a straight dashed segment from the pad to an object’s first orbital elements, and the interval it spanned was measured at a median of 25 minutes but a maximum of 7.7 years, so for hardware jettisoned in orbit long after launch the line asserted an ascent that never happened. The key is still accepted and ignored so that links written while the feature existed keep opening the view they were made for; it is never emitted. The case it covered — two ends in spacetime with no known law between them — belongs to the general absence rule.
`tg` · `tln`	trail time-ruler graduations (`tg=1` — major + minor tick strokes at round calendar steps, a day/month/year/century… picked from the visible trail span; `tg=2` — the same plus a date legend on the selected trail's major ticks; `tg=3` — the same ticks labelled instead with the distance light covers in the look-back from the trail head, c·Δt, a relabelling and never a second ruler) and an imposed trail length in days (`tln`; absent = the automatic span election)
`tfa`	Field trails — in trail mode AUTO at stellar/galactic scales, the visible star field (up to 48 stars with measured velocities, nearest the view) draws trails of one shared duration, so stroke length reads as speed. Stars without a measured velocity draw nothing, and the count of such refusals is reported. Kill switch `tfa=0`.
`tdw`	Trail date window (`tdw=<from>..<to>`, JDs) — draw a trail from one chosen date to another, the FUTURE included: samples past the sim clock render at 50% opacity, a navigation aid saying where you are along the path (the dashed style stays the separate epistemic claim about a modelled shape — the two channels are never mixed). The window stops where declared data ends, never extrapolated: asking past an object's declared end draws exactly to that end and counts the refusal. Set from the trail menu (both scopes — the per-object window stays in the menu and deliberately has no key, since ids belong to a bake); a reversed or junk range is refused at the engine door and the key is ignored.
`fl`	Named surface features — the IAU-approved names of natural features (craters, montes, valles, planitiae…) from the Gazetteer of Planetary Nomenclature, revealed as you approach a body. A name appears when the feature’s OWN diameter reaches the size at which this renderer stops treating a thing as a dot, so a 200 km mare is named from far out and a 5 km crater only from close in; proper names are read from further out than the source’s lettered “satellite feature” designations, which is why Einstein appears long before Einstein A. The features carry NO marker on purpose: a feature is a REGION tens or hundreds of kilometres across and the coordinate published for it is its centre by mapping convention, not a measured point, so a dot would overstate what was mapped — which is a different statement from a position we cannot recover, and reads differently. The Visualisation control states, from the layer’s own counts, how many of the body’s names are being shown, how many are lettered designations held back until close in, and how many carry no published diameter and so can never be size-ranked; a body with no names is a gap in the nomenclature, not in the ground. Kill switch `fl=0`.
`lg` · `ls` · `pin` · `vp` · `mil` · `mp` · `ma`	labels + models: legend mode, label size, pinned labels, visual-provenance sub-lines, model card, model time-window swap, model attitude
`fa` · `unc` · `ms`	factory mode (off/light/fill), ±1σ display, minor-body procedural shapes
`hz` · `hzm` · `hza` · `gs` · `rl` · `gw` · `sn` · `cn` · `dc` · `es` · `lit` · `dsh` · `snp` · `if`	modelled/derived overlays: habitable-zone rings + method + overlay-all, Hill spheres, Roche limits, gravitational-wave shells, supernova light shells, native constellations per sky culture — one or several overlaid (`cn=western` · `cn=chinese` · `cn=western+chinese` …; legacy `cn=1` = western, an unknown member is dropped, all-unknown = off) with the optional figure-names legend `cnl` (`cnl=1` shows on-sky figure names + the selected star's membership row), CME cones, eclipse shadows (`es=0` hides), real lighting (`lit=1` — non-emitting bodies lit from their host star's actual direction: a day/night terminator with correct phase; the star is the nearest bound ancestor star, stars themselves stay emitters, and a faint fixed ambient keeps the night side readable — a display aid, not physics), terrain cast shadows (`dsh=1` — with real lighting and terrain on, each vertex of the loaded height field marches toward the star and keeps the fraction of its disc that clears the horizon, so a crater floor is dark because of its RIM and not because of a cosine; the answer is at the DEM's own scale, measured at 666 m per post on the Moon and 1,300 m on Mars, so it is real geometry from real data and never a metre-scale illumination survey), and shadow opacity as a percentage of NATURAL (`snp`, 0–100 — 0 keeps the ambient floor that keeps a night side readable, 100 removes it entirely so an unlit face is exactly as dark as the geometry says; the scene ambient and the terrain ring's floor move by the same factor, so they never disagree)
`dr` · `drs` · `drp` · `dru` · `drv`	Gravitational drape (`dr=1`, off by default) — the Newtonian gravitational potential Φ = −Σ G·m/r of the framed gravity well, in J/kg (m²/s²), sampled on a bounded grid from the registry’s own MEASURED masses and drawn as a sagging sheet. This is not a general-relativity embedding diagram: it is a scalar field, not a curved geometry — the physics-class rubber sheet is the route we took to the idea, never the claim we make. Bounded and stated: at most 24 sources, ranked by how much each one VARIES across the sheet (ranking by total \|Φ\| instead keeps two dozen galaxies and not one planet), with both omitted remainders reported on screen; a body with no published mass is refused and counted, never treated as massless. Only the sheet’s RELIEF is readable — measured at the Sun, about 99.86 % of the absolute level comes from the Galaxy and the Local Group, so the zero level carries no local information and the drape says so. `drs=log` selects a DECLARED, monotone logarithmic height law (depth ORDER exact, ratios not readable) and is never the default; `drp=view` puts the sheet in the view plane instead of the anchor’s own reference plane (both are declared conventions); `dru`/`drv` set the grid pitch along each axis independently — a grid reads where a dense surface goes flat, so the pitch and its anisotropy are the legibility control. Nothing is moved onto the sheet: bodies stay at their composed positions and a plumb line joins each to the sheet beneath it. MODELLED (factory `natural/potentialfield.ts`); governed, capped and kill-switched.
`terr` · `si` · `cld` · `ll` · `cm` · `va` · `cl` · `sb`	layers & dressing: sourced terrain (`terr=0` hides), Sun imagery by date (`si=1`), Earth clouds, Life & Water, concept missions, velocity arrows, comm links (ON by default since 2026-07-30 — `cl=0` hides; older links carrying `cl=1` still read as on), spectrum band
`cf`	Retired. The three Cosmic-Footprint gate buttons were removed on 2026-07-30 as redundant with the badge bar: the landing view now hides the electromagnetic footprint with an ordinary `fx=` badge, which is visible in the badge bar, removable with its ✕, and travels in every shared link like any other badge. The key is still accepted and ignored so links written while the buttons existed keep opening the view they were made for; it is never emitted.
`stc` · `cc` · `tw` · `vg`	display grade: measured star colors — star points in the black-body hue of their catalogued temperature (`stc=0` restores the coarser banded tints; stars without a published teff always keep the default look), cinematic ACES color grade (`cc=1`, an optional look), star twinkle (`tw=1`; pauses itself under reduced-motion or a slipping frame rate), vignette frame (`vg=1`)
`so`	star trajectory law for EVERY trailed star (one channel — the checked model owns it; since v2.162 a star the model cannot honestly draw shows NO line, never the other law's, and `so=off` draws no star trajectory line at all): `so=drift` = its MEASURED space velocity extrapolated linearly (the registry's own position law); `so=gal` = a MODELLED test-particle orbit integrated in a published Milky Way potential (Irrgang et al. 2013 Model I, DOI — not a Kepler ellipse: the Galaxy's extended mass keeps the rotation curve near-flat, Eilers et al. 2019, DOI). DEFAULT since v2.106 (owner): the galactic orbit is ON — `so=off` silences star trails, `so=drift` selects the linear law; stars without a measured full 3-D velocity get no line, with the reason stated (A5). Velocity-frame convention: a star's drift = its measured heliocentric velocity + the Sun's full galactocentric motion — v_c = 229 km/s (Eilers et al. 2019) + the solar peculiar motion (U,V,W)_⊙ = (11.10, 12.24, 7.25) km/s (Schönrich et al. 2010, DOI)
`vch`	vector chain (`vch=1` position · `vch=2` velocity) — the gravitational nesting drawn in the scene as a tip-to-tail VECTOR SUM along the selected object's time-aware ancestor chain (CMB → Local Group → … → the object). `vch=1`: one labelled arrow per frame-to-frame displacement, each leg from the simulation's own position compose, so the arrows visibly sum to where the object is (a link minted before the mode existed decodes byte-identically). `vch=2`: each leg's VELOCITY (m/s), summing to the object's motion against the CMB — every arrow under ONE shared scale (the largest spans a fixed fraction of the view, declared on the note) so the drawn sum stays exact; a chain edge with no declared velocity law refuses by name and is counted, never drawn as zero (a place is not a rest state — A5). A selector, never both at once: the two sums share neither unit nor scale. Legs too small to see at the current zoom are skipped and counted on an honest note; display-only. Default off; the selector sits in the Object panel directly below the gravitational tree
`lgd`	Local Group dynamics: by default (omitted) the MODELLED few-body motion of the massive Local Group members is ON — Andromeda (M31) and Triangulum (M33) MOVE along the mutual gravitational orbit integrated from their MEASURED present-day velocities (van der Marel et al. 2012, HST proper motions, DOI), so scrubbing the clock into deep time shows Andromeda approaching the Milky Way (the timing-argument first passage in ~4 Gyr — Kahn & Woltjer 1959, DOI) and its Gy-scale trail draws the approach. `lgd=static` freezes them at their present positions (the byte-identical baseline). The Milky Way is the fixed frame anchor; the LMC/SMC and dwarf galaxies are not moved; softened point masses with no dynamical friction, so the first approach is representative but the eventual merger is not modelled (A5)
`obs` · `ov`	observer view: `obs=<object>` redraws the whole scene as that object's APPARENT sky — every position light-time retarded by its own τ, and an event existing only once its light has reached the observer (from Earth, SN 1054 appears in 1054, not at its −5470 explosion). Absent = the default true-simultaneity view (the kill switch); an observer unknown to the current bake is refused by the engine and the link degrades silently to the default (A5 — never a fabricated observer). `ov=1` adds the SUPERPOSITION: the global-time copy, tinted cold blue, drawn only where it visibly differs from the apparent position (the observer's sky keeps natural colours); meaningful only with `obs=` and it travels only while actually on. This makes an observer view SHAREABLE — a link to the TRAPPIST sky as seen from Earth reproduces it
`lock` · `nu` · `ol` · `arf` · `zc` · `zr` · `ze` · `zf` · `psc`	camera behaviour: surface lock, north-up, orbit follow, auto-reframe, zoom target cascade, zoom relevance floor (`zr=0` restores the legacy 100 m floor — by default a childless star stops at the scale of its stellar neighbourhood, with an honest "nothing closer in the registry" note; a childless galaxy-cluster marker stops at its own MODELLED R500 extent, derived from its published SZ mass by the overdensity definition M₅₀₀ = (4/3)π·500·ρ_crit(z)·R₅₀₀³, stated as modelled in the note), local-environment zoom (`ze=1`, experimental, off by default: arriving at a childless star parks the view at its stellar neighbourhood — the gap to its nearest catalogued neighbour fills half the view — and zooming toward the relevance floor decelerates smoothly on the constant-feel law instead of hard-stopping), constant-feel zoom, play-stride cull
`sl`	lock zoom level — the locked on-screen scale as metres per CSS pixel (Visualisation → “lock zoom level”; the preset shows the selected system’s host star at ≈ 4 cm on screen, ping-pong size). While locked, navigation, origin re-homes and window resizes all keep the same 1:N and zoom gestures are inert; where the scale cannot be honoured (a body’s keep-out floor, the zoom ceiling) the row’s note SAYS so and the lock resumes when the view allows. Device-portable by construction — the same link shows the same on-screen size at any viewport, which a stored camera distance alone does not. Absent = unlocked (the default).
`sf`	FACTORY full-system fill overlay: `sf=1` draws, for the selected host star (or a planet in its system), the statistically-EXPECTED unseen planetary members as SYNTHETIC orbit ellipses — a deterministic seeded draw from published Kepler occurrence grids (Fressin et al. 2013, DOI · Dressing & Charbonneau 2015, DOI), constrained by the known planets (occupied class×period cells never re-drawn) and a Δ ≥ 10 mutual-Hill spacing criterion (Pu & Wu 2015, DOI). Never a claim about THIS system. Default off; the dice re-rolls the seed (session-side — a shared link reproduces the canonical draw) and the paper pack `sfp` swaps the cited coefficient table (a chosen non-default literature pack travels so a saved view reproduces it); a host without a measured mass+temperature gets nothing, with the reason stated (A5)
`ur` · `lt` · `eb` · `cv` · `ce` · `apw` · `pass` · `ds`	analytical panels: UN registration, launches over time, EM-band census, measurement coverage, presence census, close-approach watch (`apw=1`; legacy `aw=1` links still open it), passes overhead (`pass=1` — visibility only; the observer's ground location itself never travels in a link, by design), data sources
`hr` · `sc`	`sc=1` shows the orbital-shell congestion panel (Earth orbit by altitude band — LEO/MEO/GEO/HEO; a catalogued lower bound, with the excluded populations counted on the panel); `hr` is accepted for link compatibility only — the HR-diagram panel is not mounted, so that flag has no visible effect
`fx` · `hl`	the filter-badge stack (each badge's action/colour/trail/label/unknown/blink/size-boost markers + query) and hidden source layers
`tier` · `theme` · `hi` · `ui`	interface: UI level (0 ultra-low … high), theme profile, humanity-ink colour override, `ui=0` chrome-less render (never serialized into a shared link)
`tq` · `kft`	tours: a query-driven tour itinerary, and a shared keyframe-tour token (apply-only)

Embed mode. Adding ?embed=1 to the simulation URL (a query parameter, deliberately not part of the shareable hash) renders the scene alone with a small corner wordmark — for embedding the view in another page.

Working offline — an optional, explicit copy

The Share stage carries a “keep available offline” toggle. ON, it saves — into this browser only — the application shell, the full registry bake (roughly 62 MB raw — about 15 MB compressed in transfer) and these documentation pages, with a visible progress and size readout; a partial save is reported as incomplete, never as ready. The copy is a dated snapshot: pages refresh themselves when you are online, and a “↻ refresh copy” button re-snapshots the data files after a new release. OFF removes the worker and deletes everything saved. What is honestly not saved: third-party streams — NASA/USGS surface tiles, live solar/rover imagery, map tiles and CDN-hosted libraries, and cloud AI providers — which degrade offline exactly as they do on any network failure. The toggle is a device-local preference and deliberately has no URL parameter (it is meaningless in a shared link). The local ranked search works fully offline, with a small typo tolerance (“did you mean …” on a zero-hit query); best supported on Chrome-class browsers — iOS Safari may evict saved copies under storage pressure (the prepared date shows so a stale copy is visible).

Answer pages — the registry without the application

The scene computes everything in your browser. That is what makes it free to run, and it is also why a program that merely fetches /sim.html learns nothing: it receives the loading placeholder, 6,480 bytes, and no facts at all. Measured against the built product on a throttled link, a browser driving the full scene resolves an object outside the curated core after about 147 seconds, having transferred roughly 70 MB (about 16 seconds over a local network — a floor, not a promise). Most software, and many people, will not wait.

So the build also emits static answer pages — 4,362 of the registry's 185,053 nodes, 2.4 %, one HTML page each, about 8 KB — generated from the same bake the application runs on. Which objects get one is a rule, not a list: a node earns a page when its provenance is neither synthetic nor modelled (a Factory product is an estimate, not an answer about the world) and it falls in a declared class — natural bodies and frame anchors · dated events · stations and flagship structures · deliberate transmissions · named artefacts left on another body · named stars, galaxies, comets and exoplanets · asteroids that are individually named or on the NASA/JPL Sentry risk list · spacecraft that reached a body or carry a published 3-D model. Everything else is counted and named as refused rather than quietly dropped, and the build prints that census.

Each page is built so that quoting it honestly is easier than quoting it badly. A value, its 1σ and the database that measured it sit in the same table row, so even a crude text extraction keeps them together; a value with no published uncertainty says so in its own cell, because a blank cell reads as “exact”; where catalogues disagree, every source's figure is printed above the facts rather than resolved away; and each page carries a section naming what the registry does not know about that object. No page prints a position — a position is a function of time, and a fixed one dated to the bake would be stale the moment it was read — so a page publishes the sourced orbital elements and their epoch and links the scene, which composes a position for any date.

The pages are a derived artefact, regenerated by every build from atlantis-core.json and the fragments, and are not committed; a sitemap lists them, and /llms.txt carries the same capability manifest for clients that read only that. Quoting is welcome; please carry the source and the uncertainty with the value, and name the database that measured it alongside this registry.

Reporting back to a contributor — what the registry holds from you

Every database this registry aggregates belongs to somebody, and aggregating several of them produces something none of them can see alone: the places where two catalogues read the same object differently. Since the registry keeps every competing value rather than picking one, that comparison already exists as data. The Connect stage folds it the other way round — per contributing entity instead of per object — into a plain-text report the contributor can be sent.

It carries what the registry holds from them, where a second source offers a different value (with the widest disagreements named object by object), how completely each field arrives compared with the same field across the comparable population held from other contributors, and where consolidation would be cheapest. Disagreements about dates are reported in DAYS: a Julian Date is about 2.46 million, so a month’s disagreement about a re-entry is a relative difference of one part in a hundred thousand, and any tolerance meant for masses would file it as agreement.

Two things it deliberately does not say. It never reports a disagreement as the contributor’s error — the registry keeps both values precisely because it cannot say which is right, so the count means two readings exist and nothing more. And a thin field describes what reaches us, which may be a field they publish and we failed to ingest; where the counterpart is one of the registry’s own curated tables, the report says the reconciliation is ours rather than theirs. It is built locally and transmitted by nobody: the mail composes in the reader’s own client.

Counting visitors — what is stored, and what never is

The registry is free and needs no account. So that the project can know how many distinct people it reaches, a visit counter runs on page load, and after about ten visits a dismissible invitation offers to create an account with an email address. It is deliberately a gentle wall: “continue without an account” always works, nothing is ever withheld, and every failure path is silent — if the counter cannot be reached, the page simply carries on.

Your IP address is never stored and never logged. When the counting endpoint receives a request it converts the address, one way, into a pseudonym — HMAC-SHA256 keyed with a secret held only on the server, truncated to 16 hexadecimal characters — and stores that; the address itself is never written to a database or a log, and the pseudonym cannot be reversed without the secret. Alongside it: a coarse two-letter country code, and an email address only if you type one and tick the consent box. No cookies for tracking, no advertising, no analytics broker, no device fingerprint, no user agent, no referrer. The honest trade-off — the secret is stable rather than rotated daily, because “your tenth visit” cannot be counted across days otherwise — is stated in full, with the GDPR basis and your rights, in the privacy notice.

Off unless a backend is configured. This is the project's one explicit, owner-authorised exception to its otherwise zero-backend architecture, and it is deliberately bounded: the registry itself holds no part of it — no registry data lives in the service, and the simulation runs unchanged without it. The page reads its endpoint from a single ssr-api-base meta tag; when that is empty — as it is on any build with no service deployed — nothing is sent anywhere at all and the counter degrades to a purely local tally in your own browser's storage. The counter has no URL parameter by design (it is meaningless in a shared link).

The statistics page (/admin.html) is for the project maintainer. Its short reveal code only un-hides the login form and is explicitly not a security boundary — the real gate is a server-side login returning a short-lived signed token, and no statistic reaches a browser without it.

Connecting an AI — bring your own key

Every conversational and voice feature runs against your own model credentials (an Anthropic, OpenAI or Mistral API key), entered and stored locally in your browser and sent only to the provider you chose — never to us; we ship no model, pay for no inference and hold no key. A running session shows an estimated token cost. Note that a conversation here is not private to this site: your prompts go to your chosen provider under that provider's terms.

On-device AI is the default. Where the browser ships a built-in on-device model (Chrome 138+ Prompt API, after its one-time model download), the assistant answers with THAT model first — no key, no cost per question, nothing leaving the device, working on first load. A key is an upgrade, not the price of entry: enter one and you can prefer it, and if the on-device model cannot answer, the keyed provider is tried next. Either way the reply says which model produced it, because a weaker answer that does not admit being one would be worse than no answer.

The honest trade: the built-in model is nano-class — good for spoken commands (“go to Mars”, “show trails”, “tour the objects about to hit something”) and markedly weaker than the cloud models at multi-step tool use. It drives the exact same firewalled tool surface. Where the API is absent, a key is needed and the empty state says so.

The assistant can READ, not just act. It can ask the registry what is known about a named object — its source catalogue, its measured values with their uncertainties, which clock its position is on, when and how its life ends — and it is instructed to pass on what is not known too, because "no measured mass" is an answer and dropping it would teach you that everything is measured. It can also ask what dated events sit near the current moment, so it can offer you something you had no way to know to ask for.

The AI also understands the PROJECT itself: a self-description tool returns what the registry is, the epistemic rules it must respect (provenance, absence-of-data ≠ absence), and a live capability map — every installed data module with its identity card, and the current tool/option/facet tallies — all composed from the running engine at call time, so a database that connects is understood the moment it connects.

The AI discovers the search vocabulary LIVE: its category tool returns the real facet grammar with plain-language notes, live object counts and French/English synonyms ("œil nu" → has:nakedeye), derived at call time from the same declared surfaces the human search palette indexes — so a newly added facet is immediately usable by voice, and a failed query returns corrective suggestions instead of silence.

The AI drives only the existing public control surface through structured tools — navigate, time, speed, camera, filter badges, tours, time-aware distance measurement (measure_distance), and a generic options table — and, by architecture, never sees or emits internal identifiers and has no path to source code or the registry's truth. The full option vocabulary it (and you, by voice) can set:

trails full-history-trails smooth-rover-trails reconnect-trails surface-trail-suppress trail-rulers trail-length trail-frames field-trails feature-names play-stride-cull model-attitude visual-provenance-labels model-info-label model-phase-swap labels factory uncertainty hz-rings hz-method hz-all-methods hill-spheres roche-limits gravitational-drape drape-height-law drape-plane gravitational-wave-shells supernova-shells impact-flash constellations constellation-names cme-cones eclipse-shadows real-lighting dem-shadow terrain sun-imagery star-colors cinematic-color twinkle vignette earth-clouds comm-links velocity-arrows star-orbit footprint-material footprint-electromagnetic footprint-biological observer vector-chain vector-chain-compose system-fill lg-dynamics concept-missions life-water surface-lock north-up orbit-follow auto-frame zoom-floor zoom-env auto-speed zoom-cascade zoom-feel scale-lock spectrum-band spectrum-strip slideshow-loop census-panel approach-panel passes-panel coverage-panel sources-panel un-registration-panel launches-panel em-band-panel shell-congestion-panel help-panel search-panel connect-ai-panel bring-your-data-panel ui-level persona theme humanity-color clear-labels clear-badges tour

What is physical, what is visual. Positions, orbits, time windows, emission kinematics and detection ranges come from data or declared models. The DRESSING — animated sun surfaces, atmosphere rims, departure animations (smoke streaks, mushroom/fireball stagings), deposition ellipses — is VISUAL: physically scaled where a published fit exists (each one referenced in the library), but rendered to be read, not to simulate. Estimated values appear in gold with their method.

EM emissions render as expanding shells at the speed of light (radius = c·Δt since transmission; a continuous broadcast is a filled sphere, a radio-telescope message a needle-thin cone along its real beam). One record shape covers a METI carrier, a nuclear EMP, a pulsar and the quiet Sun — natural and artificial in one visual language, the natural beacons giving the human signals their context. Detection ranges (humanity's best instruments vs a parameterized ETI photon limit) are computed in the registry at bake and shown per emission in the inspector — estimates labeled MODELLED.

This page documents sources, licenses and intent. It is deliberately insufficient to rebuild the system: the authoritative engineering specification is access-controlled in the application.