ORBIS is polyglot on purpose. Each subsystem is written in the language whose ecosystem already contains the thing it needs — because rewriting a mature orbital propagator or RTI binding is a cost with no upside.
Every platform in the reference dataset carries its standards enumerations from the moment it is imported, so the interoperability layer never has to guess or translate through an internal identifier.
| Standard | Status | Use in ORBIS |
|---|---|---|
| DIS v7 — IEEE 1278.1-2012 | Built | Entity State, Fire, Detonation, Start/Resume and Stop/Freeze over UDP multicast, with dead-reckoning transmit gating and unmapped-entity diagnostics. |
| SISO-REF-010 | Built | Entity and munition enumerations carried on every platform class and every weapon in the dataset. |
| WGS84 / geocentric ECEF | Built | Positions, geodesics and the coordinate frame every DIS position is expressed in. Vincenty direct and inverse for motion and bearing. |
| MIL-STD-2525 / APP-6 | Built | Symbology in the Deck, driven by a symbol identification code held on each platform class. |
| HLA — IEEE 1516-2010/2023, RPR-FOM, NETN-FOM | Planned | The requirement that most often decides a coalition exercise procurement, and absent from the closest comparable product. |
| C2SIM and MSDL | Planned | Order and scenario exchange with command-and-control systems. |
| SGP4 / SDP4 | Planned | Space domain — orbital propagation and constellation modelling. |
Where a physical relationship exists, ORBIS implements the relationship and scales it from a published reference point — so a figure can be argued about on its merits instead of being an unexplained constant.
Reference-range scaling. Radar and active sonar are two-way, so detection range varies with the fourth root of cross-section; ESM and infrared are one-way, varying with the square root of emitted power. Terrain and radar horizon mask the result.
The dynamic launch zone is derived from closure rather than tabulated. Effective range
scales as R·(1 + closure/Vm) — a target running away shrinks the envelope,
and a head-on target extends it.
Components are individually damageable and a platform has a structural tolerance measured in effective hits. Never a hit-point pool: a mount, a radar and a propulsion plant fail in different ways and matter differently.
Proportions use the Wilson score interval, which behaves correctly near 0 and 1 where the normal approximation does not. Continuous measures use Student-t. The right interval per measure, because one batch usually reports both.
Atmosphere, sea state and light conditions feed sensor performance and motion. Altitude and throttle drive fuel burn through a per-platform table rather than a flat rate.
The sender runs the receiver's own extrapolation and transmits only when the picture has drifted past an agreed threshold. In a representative run this suppresses roughly four fifths of candidate updates.
A verification suite that reports a single pass rate hides the thing you need to know: how strong each check actually was. ORBIS grades every case and publishes the grade.
A published formula gives the answer and the model must match it. Radar horizon from mast height, geodesic distance between two known points, coordinate round-trips.
No absolute answer exists, but the shape of the response is fixed. Quadrupling radar cross-section must increase detection range by exactly the fourth root.
The weakest grade, and labelled as such. A figure that is merely in a defensible range is not evidence, and calling it evidence is how a model becomes trusted for the wrong reason.
A single-bit cosine difference between two platform maths libraries. A dataset field that no component was forwarding, which would have made a whole subsystem emit nothing while every unit test passed. A wire format that round-tripped perfectly because the encoder and decoder shared the same mistake.
None of these were visible to a suite of individually-correct components. That is the argument for the last item on the left.
The kernel cannot reach the network at all, no service phones home, and the client is served entirely from the deployment. The one remaining route out is geospatial: the map streams basemap and elevation tiles in the development posture and reads staged versioned bundles in the deployed one. Switching between the two is a single configuration value, deliberately kept visible — and until the bundle path is built, the client says LIVE TILES · DEV on screen rather than letting anyone assume otherwise.
Authentication, role-based access, classification filtering and audit live in the Gateway rather than in each service, so there is one place to review and one place that can be wrong.
The client is served with a restrictive CSP and the API is same-origin by reverse proxy, so connect-src stays 'self' — the development and air-gapped configurations are the same configuration rather than two to keep in step. Tile hosts are the single documented exception and are named explicitly rather than covered by a wildcard.
No copyleft in the runtime stack. Geospatial and symbology dependencies were licence-assessed before selection and the exclusions are documented.
Live kernel, real dataset, real decision traces.
ORBIS