FAQ

Why did my satellites vanish when I scrubbed the timeline?

Almost certainly epoch validity, and it is a feature. A TLE describes an orbit at a specific moment; propagating far from that epoch degrades into fiction, so each satellite renders only within a validity window of about 14 days around its own epoch, and catalog entries more than 90 days behind the newest epoch are dropped as decayed. The satellite card shows the ephemeris epoch and its age. Fresher TLEs fix it. See Uploads and TLE ingestion.

What is the difference between toggling a constellation and filtering?

Visibility toggles (constellation groups, asset groups) are rendering decisions: they declutter the globe. Filters define analytics scope: what fleet statistics and assistant answers compute over. Hiding a group does not exclude it from a statistic unless a filter does, and answers always state exactly which assets they matched. See Search and filters.

What can I do as a guest, and what are the limits?

Guest mode is a real evaluation surface: the globe, timeline, imports, simulations, and deterministic analytics all work, and you get 10 assistant queries on the basic model. No prediction backends and no live platform API access. Signing in to a paid plan removes the query cap; see Plans and billing.

If I run the same analysis twice, do I get the same answer?

Yes, if the inputs are the same. Every analysis engine is deterministic: fixed steps, no randomness, stable ordering. Replay mode and saved scenarios pin the data side down, so a saved scenario reloaded at its epoch reproduces its numbers to the digit. In live mode, new telemetry legitimately changes the inputs. See Determinism.

Why can't I call the platform API from my web app's frontend?

CORS is allowlisted per environment, and when no origins are configured it is off entirely, so browsers block cross-origin calls by design. More fundamentally, shipping a tenant API key into a browser is the wrong trust model. Build server-side; see API overview.

How do I get an API key?

Keys are provisioned per tenant, per environment, by the platform team; there is no self-serve purchase today, and keys generated locally in the console browser are not registered until provisioned. Start from your tenant engagement or contact@constellation.space. See Authentication.

Why didn't re-importing my TLE file create duplicates?

The asset registry keeps one asset per physical orbit definition. Re-importing an identical TLE updates the existing asset in place, preserving its id, selection state, and visibility (including a deliberate hide); genuinely new records are added. A weekly TLE drop is a refresh, not a multiplication. See Assets.

Why isn't "unserved demand" the same as throughput?

Because one is modeled and one is measured, and the console refuses to blur them. Capacity analyses model demand against capability under stated assumptions; unserved demand is the modeled shortfall of that comparison. Throughput is a telemetry fact from your fleet. Answers name which kind of number they contain, so a modeled shortfall never masquerades as a measured rate.

How does PDF export actually work?

It is print-based: the console opens a print-optimized rendering of the report and hands off to your browser's native print-to-PDF path. Charts render as data tables in that path deliberately, so the archived numbers are exact rather than rasterized pixels. See Reports.

What is the difference between replay and live mode?

One simulation clock, two bindings. Live pins the playhead to the newest stored sample and tracks incoming data, with an explicit live/stale/disconnected indicator. Replay samples a stored window at wherever you scrub, and every metric re-evaluates at that instant. In both, numbers refer to the timeline instant, not wall time. See Timeline and simulation.

Which prediction models are real today?

One production model family: SNR, served per tenant through the Predictions API with p50/p10/p90 values. In the console, guests and keyless setups get a clearly labeled physics-anchored baseline for SNR, and prediction domains without a production backend render from labeled illustrative stubs. Every forecast names its source; nothing pretends.

What do bronze, silver, and gold actually change?

They are console plan concepts, fidelity tiers mapped from your subscription (bronze on Standard; silver and gold on Pro; custom models on Enterprise). The platform API has no tier parameter; the console's entitlements decide what it requests. See Plans and billing.

Why did I get the identical forecast twice in a row?

Predictions refresh on 60-second buckets with stale-while-revalidate semantics: identical questions in the same minute get identical answers, and an expiring forecast keeps rendering with its age shown while a fresh one is fetched. The age label tells you exactly how old a number is. See Telemetry and predictions.

Is my telemetry isolated from other tenants?

Physically, yes: each tenant gets a dedicated time-series database instance and a dedicated KMS encryption key, tenant identity derives from the API key (never the hostname), writes are tenant-stamped server-side, and disabled tenants fail closed. Dedicated stacks in their own AWS accounts exist above that. See Tenant isolation.

What happens when I hit the rate limit?

You get 429 with {"error": "rate_limited", "retry_after_seconds": N} and a Retry-After header. Limits are 30 requests per minute per IP and 60 per minute per tenant with a 10 per second burst. Sleep the stated seconds and resume; batch harder if it recurs. See Errors and limits.

Can synthetic satellites end up in my real analyses?

They participate wherever their capabilities qualify them, and they are permanently labeled synthetic in every answer that touches them, kept separate from live telemetry. A synthetic asset can inform a design decision; it cannot silently pose as hardware. See Simulations and scenarios.

Why does the demo fleet look so complete compared to my live fleet?

The demo is a bundled, curated constellation with a stored telemetry window, built for learning the console. Your live fleet shows exactly what your telemetry reports, no more: entities appear as samples arrive, and gaps are shown as gaps. If your live topology looks sparse, check what your pipeline actually writes. See Connect telemetry.