Medallion Tiers
Bronze, Silver and Gold say how refined the data and models behind every forecast are. Bronze is on every plan. Silver and Gold, as the table below compares them, come with Constellation for Enterprise, but a model release can run at Silver on a self-serve plan too: Weather attenuation and Satellite demand are served at Silver on every plan, and Link SNR from Operator up.
Pass predictions from public orbital elements, derived weather at each station, and link candidates from geometry.
Maneuvers detected, weather gaps flagged, and link scores checked against your telemetry, with provenance on every record.
Orbits, weather and link scores fused across sources, with error measured against what actually happened.
What you get on Bronze vs with Enterprise
| BronzeEvery plan | SilverWith Enterprise | GoldWith Enterprise | |
|---|---|---|---|
| Pass prediction basis | Public orbital elements, propagated | Cleaned elements, maneuvers detected, propagation checked | Orbits fused across sources, checked against observed passes |
| Weather per pass | Derived public weather at the station | Quality-controlled weather, gaps filled or flagged | Premium weather fused per link and per pass, checked against observed fade |
| Link scores | Geometric candidates from pass geometry | Candidates with link budgets, checked against telemetry | Scored by evaluated SNR and weather models |
| Confidence | Method and limitations on each model page | Provenance on every record | Error measured against ground truth and documented |
| Your own data refined in your deployment | Not included | Yes | Yes, with models trained on it that serve only you |
| Tier labels and lineage on outputs | Each forecast names its model release | Tier, release and model version on every output Rolling out | Tier, release and model version on every output Rolling out |
| Dedicated deployment | Shared service | Your own API host and Platform site | Your own API host and Platform site |
Today the model catalog labels each model's served release with the tier it runs at, and every plan that asks a model calls that same release: Weather attenuation is served at Silver on every plan. Further Silver releases, and Gold, reach Enterprise deployments as they are released. Rolling out
Your data and any model trained on it stay inside your Enterprise deployment; see Tenant isolation.
When the gap matters
Bronze is enough for most days. These are the days a higher tier is built for. No model is served at Gold yet, so these say what each tier aims at, not what it does today:
- A marginal pass in heavy weather. A storm cell is moving over your primary station during a low-elevation pass. Weather attenuation, served at Silver on every plan, already forecasts the fade for that antenna at that time. Gold aims to fuse premium weather per link and check it against observed fade.
- A crowded orbit after a maneuver. A neighbor in your shell has just maneuvered. Public elements lag the maneuver, so predictions carry the old orbit until the catalog catches up. Silver aims to detect the maneuver; Gold to fuse fresher sources and say how far off the prediction might be.
- Choosing between ground stations. Two stations can both see the satellite, and geometry says they are equal. Gold link scores aim to say which one will deliver your data, from measured station performance and per-pass weather.
Talk to us about Constellation for Enterprise at sales@constellation.space.
FAQ
Is Bronze good enough to build on?
Yes. Bronze is real data and served models, with their method and limitations documented, and it is the foundation every higher tier is built from. You outgrow it when a wrong prediction starts to cost you: when a missed pass, an unexpected fade or a stale orbit changes what your team does that day, or when you need to show which data a decision rested on.
Who decides a model's tier?
Our product team. The ML team trains models on data refined through a medallion architecture, and the product team QAs each model version, reviews its offline and online evals, cross-references feedback from enterprise and academic partners, and promotes it at the tier the evidence supports. The data a model was trained on does not set its tier.