A mission-control console that tracks a commercial LEO broadband constellation live, models the RF link down to the rain fade, and scores its own predictions against measured terminal telemetry. Every number on screen says where it came from.
The redesigned LEO Ops board, recreated. Every site, satellite, and reading here is fictional; the propagation, the RF math, and the provenance discipline are real. Scroll the panel - five working surfaces, one instrument.
One Earth, five operator layers that switch on the globe: the space segment - Starlink, OneWeb, Kuiper, SES O3b MEO, Iridium and Globalstar; the ground segment of teleports, gateway earth stations and ATG towers; a live weather overlay driving Ka/Ku rain fade; an RF / spectrum layer flagging interference and GNSS-denial; and the fused link view that tracks any tail - plus maritime AIS and ADS-B aircraft - and resolves the full PACE ladder: primary, alternate, contingency, emergency.
Slant-range geometry, free-space path loss, ITU-R rain attenuation, and the carrier-to-noise budget - every term shown, every constant cited to its recommendation. The elevation hero curve proves why PACE exists: as the bird sinks, margin falls and the console hands off before the link drops. Research-grade, not a coverage gradient.
A read-only collector polls a real terminal over its local API - latency, throughput, an SNR obstruction dome - and a model-vs-observed panel scores the console's own forecasts against what the hardware measured. Offline, it falls back to a labeled model.
Take the live constellation you're tracking and branch into a Walker what-if designer - total sats, planes, phasing, inclination, altitude - then compare real altitude shells (Starlink, OneWeb, SES O3b) and watch coverage, revisit, and margin answer against the live baseline.
A sensitivity tornado ranks what actually moves NPV - fill rate and ARPU dominate, not launch cost - and a 10,000-trial Monte Carlo with correlated, Latin-hypercube-sampled inputs returns the full outcome cone: P10, P50, P90, and the probability the program clears zero.
Link budget, margin, Doppler, GSO discrimination, frame timing, and spectrum are one analysis seen six ways. Tab between them; the geometry stays fixed underneath.
A program office asks what the link margin looks like in a Florida thunderstorm, how the terminal rides through a satellite handoff, or what happens to timing when GPS gets noisy - and the pitch deck answers with a coverage map that has no math behind it. The room can tell.
Most tools in this market make it worse. Tracker sites plot dots on a globe with no RF layer underneath. Analyst platforms sell authoritative-looking figures from models nobody outside can audit. Neither survives contact with an operator who has actually pointed an antenna.
The bearing: LEO Ops is the counterargument. It propagates more than ten thousand satellites from operator-published ephemerides, carries the RF math in the open - link budgets, Doppler, rain fade fed by live weather - and checks its own predictions against measured telemetry from a real terminal. Every value on screen is labeled. That labeling discipline is the product.
Tracks 10,000+ LEO broadband satellites on an interactive globe, propagated in the browser from operator-published state vectors. Pass prediction, ground tracks, a serving-satellite timeline, and a candidate set gated at the 25-degree user-beam floor - with accuracy tiers that narrate data age instead of hiding it.
Band plans, OFDM frame timing, and full link budgets with ITU-R P.618/P.838 rain attenuation driven by live precipitation at the site - no manual slider. Doppler profiles, opportunistic-PNT geometry with GDOP, TDOA/FDOA passive-geolocation loci, GSO-avoidance angles, and spectrum occupancy, each computation pure and separately tested.
A read-only collector polls a real LEO user terminal over its local gRPC API: latency, throughput, power draw, boresight pointing, outage and handoff logs, and a 123×123 SNR sky grid rendered as an obstruction dome. A model-vs-observed panel scores the console's own predictions against what the hardware measured.
Live ADS-B aircraft with a through-satcom connectivity view, maritime AIS, a GPS-interference heatmap computed from broadcast integrity fields, satellite thermal detections, and a space-weather strip - Kp, F10.7, X-ray flare class - that feeds directly into the ephemeris trust window.
Six labels - first-party-measured, first-party, third-party, self-computed, model, simulated - and every rendered value carries one. A plausible number with the wrong label is treated as a defect, not a detail. The map never blanks; it degrades with a dated chip that says exactly what you're looking at.
The console makes zero third-party API calls at runtime - a self-hosted basemap, an edge data plane that fails closed, and two propagators kept strictly apart. Measured terminal telemetry rides the top provenance tier; when the hardware is offline, panels fall back to labeled models automatically.
A Python pipeline pulls operator-published ephemerides on the publisher's own eight-hour cadence, plus third-party element sets, space weather, live precipitation, ADS-B, and AIS. Range-optimized fetching moves about 90 MB per refresh instead of 17 GB, and the pipeline fails closed - it refuses to publish a snapshot it can't parse.
Snapshots land in edge object storage; a single edge function serves them storage-first with a committed static fallback. Fresh data flows without a redeploy, and if the pipeline dies the console serves the last good snapshot with an honest staleness label rather than an empty map.
Two propagators, kept strictly apart: universal-variable Kepler for operator state vectors, SGP4 for third-party element sets - mean elements never enter the two-body engine. Observer geometry is WGS84-exact, pass prediction runs off the main thread in a Web Worker, and roughly 535 automated tests sit behind the math.
A small read-only process on the terminal's LAN polls its local gRPC telemetry - never a mutating call - and pushes measured data through the same edge path under the top provenance tier. When the terminal is offline, panels fall back to their labeled models automatically.
A single-page app on a MapLibre GL globe with a self-hosted basemap - zero third-party API calls at runtime. Mission-control HUD styling, drag-and-resize panel decks across four working views, keyboard-accessible controls, dark by default.
One conversation, one written summary, no commitment. The bearing comes first.
Schedule a call - 30 min