Skip to main content

Run simulations

Difficulty: Intermediate. Time: about 25 minutes. Needs: a browser with a fleet loaded (demo fleet works); guest queries apply.

You will run the console's flagship what-if (a gateway outage), degrade it further with deterministic rain, and save the whole thing as a scenario that reproduces its numbers exactly on reload. Every engine involved is pure and deterministic: what-ifs mutate a copy of the scene, and the live fleet is never edited by a hypothesis.

1. Simulate a gateway outage​

Ask the assistant:

Simulate losing the Alaska gateway.

(Any gateway works; you can also select by country or region.) The outage engine re-derives feeder access geometrically from the surviving gateways, using the same feasibility rules as the live scene, and reports:

  • Re-homed satellites: each one that lost its serving gateway but found another, with from and to gateways named.
  • Orphaned satellites: left with no feasible gateway at all.
  • Before and after fleet statistics computed on both snapshots at the same sim time, so the delta is an honest subtraction.

Expected outcome: the disabled gateway dims, re-homed feeder links redraw, orphaned satellites are highlighted, and the statistics card shows before/after. The answer states that access links are re-derived from geometry and can differ from the replayed store's links; that disclosure is deliberate.

2. Add deterministic rain​

There is exactly one rain model in the console, an ITU-R P.618 approximation inside the Ka fronthaul link budget, and no random weather anywhere. Ask:

Now apply 25 mm/h rain over Alaska and the Pacific Northwest and re-evaluate.

Rain is an explicit field: uniform (one rate everywhere) or regional (lat/lon boxes with a rate each; a station takes the maximum of the cells covering it, zero outside). It resolves per station because attenuation happens on the slant path above the site, and coverage gates each sample on whether the link budget still closes at that elevation and range under the scenario rate.

Expected outcome: clear-sky and rain-degraded numbers presented together. That pairing is a fixed reporting rule, so rain sensitivity always reads as a delta, never as an unanchored degraded figure.

3. Save the scenario​

Save the run as a scenario. What persists is everything needed to rebuild the exact run: literal injected payloads (if you added synthetic assets), the sim epoch, the outage mutations and their selectors, and the evaluation config (window hours, step seconds, elevation mask, weather).

4. Prove reproducibility​

Reset the scenario (mutations were applied to copies, so reset restores the unmodified scene), then reload the saved scenario. Loading seeks the timeline to the saved epoch and re-runs the deterministic engines.

Expected outcome: every number matches the original run to the digit. This is the reproducibility contract that makes a scenario quotable in a design review weeks later. See Determinism for why the engines can promise this.

5. Variations worth trying​

  • A multi-gateway outage by region: "take every gateway in Scandinavia offline".
  • Combine with a synthetic design: build a Walker constellation (tutorial), then ask what the outage does to it.
  • A trade study: two designs scored under identical outage and rain conditions; resets between runs keep each configuration scored independently on identical windows and masks.
Troubleshooting
  • "Outage matched zero gateways": the selector did not resolve; the engine reports the empty match plainly rather than inventing an effect. Check the gateway name or region and retry.
  • Numbers differ from the stored replay's links: expected; outage access is re-derived from geometry and the answer says so. Compare like with like (before/after within the scenario).
  • Reloaded scenario gives different numbers: confirm the timeline actually seeked to the saved epoch and no extra filter or asset change is active; determinism holds inputs-to-outputs, so a differing input is the cause.
  • Rain seems to have no effect: check your rain cells actually cover the stations involved; a station outside every cell honestly takes zero.

Where next​