Dataset Viewer
Auto-converted to Parquet Duplicate
The dataset viewer is not available for this split.
Not found.
Error code:   ResponseNotFound

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

Constellation Demo Fleet

The ready-made fleet that ConstellationOS loads into a new account, so you can see what it does before you add your own satellites. It is one Earth observation operator: five sun-synchronous satellites that dump their data on X and Ka band at polar, mid-latitude and tropical stations, and one geostationary satellite (GOES 19) on its public L-band broadcast to its own antenna in Ohio, a continuous contact counted per day, not as passes. Troll goes down for simulated maintenance every day from 06:00 to 09:00 UTC, and a simulated rain cell crosses the Singapore teleport every evening from 17:00 UTC, so a 48 hour run always has a rain-fade swap to show. 5 ground stations, 7 antennas, 6 satellites, the same network config POST /topology/demo-fleet loads, plus 2 hours of link telemetry (2,134 records, one every 5 s through each pass on 17 links) and a week's pass log to audit.

Simulated data. The mission names, NORAD ids, orbits and station sites are real and public; the antennas, RF terms, link telemetry and pass outcomes are nominal or simulated. Nothing here is telemetry from, or access to, these missions or stations. Where each value comes from says which is which.

The fleet

The satellites are real, with their published downlink bands; the ids are norad-<NORAD catalog number>, which you can look up on CelesTrak or Space-Track.

Satellite NORAD Operator Downlink Frequency Bandwidth
GOES 19 (norad-60133) 60133 NOAA L 1.6866 GHz 7.826 MHz
Landsat 9 (norad-49260) 49260 USGS X 8.2005 GHz 120 MHz
NOAA 20 (norad-43013) 43013 NOAA Ka 26.7034 GHz 95 MHz
NOAA 21 (norad-54234) 54234 NOAA Ka 26.7034 GHz 95 MHz
Sentinel-2C (norad-60989) 60989 ESA X 8.1 GHz 160 MHz
Suomi NPP (norad-37849) 37849 NOAA X 8.2125 GHz 95 MHz

The stations sit at the sites the public asset catalog lists, with the bands it lists there: KSAT Svalbard and Troll see every polar orbit, NASA Fairbanks most of them, AWS Ohio a few a day. singapore-teleport is the operator's own tropical X and Ka antenna, where rain costs the most.

Station Latitude, longitude Antennas (dish, bands, mask)
aws-ohio-1 39.96, -83.0 oh-grb1 4.5 m, 1.67 to 1.71 GHz, 5 deg; oh-x1 5.4 m, 7.75 to 8.5 GHz, 10 deg
ksat-svalbard 78.229, 15.408 sg-x1 7.3 m, 7.75 to 8.5 GHz, 5 deg; sg-xka1 7.3 m, 7.75 to 8.5 GHz and 25.5 to 27 GHz, 15 deg
ksat-troll -72.012, 2.535 tr-xka1 7.3 m, 7.75 to 8.5 GHz and 25.5 to 27 GHz, 15 deg
nasa-fairbanks 64.859, -147.849 fb-x11m 11.3 m, 7.75 to 8.5 GHz, 5 deg
singapore-teleport 1.35, 103.82 sin-xka1 4.5 m, 7.75 to 8.5 GHz and 25.5 to 27 GHz, 10 deg

An antenna works only the satellites whose downlink sits in one of its bands: the X-only antennas never see the Ka satellites. At the polar sites the passes come every orbit, more than the antennas can take, so they compete.

The story, in the Platform

Sign in at platform.constellation.space (Explore is free), then:

  1. Fleet: Load the demo fleet. The stations and satellites above land in your network config. The demo never counts toward your plan's satellites, your telemetry usage or your bill, and you can add your own from the catalog beside it. The Platform reloads it once a session, so its telemetry ends near now; Reload the demo on the demo's label does it on demand, and Reset the demo puts back what you edited.
  2. Passes: Start from a template, pick Pass yield, then Price the next 48 hours. A few hundred passes: polar passes every orbit at Svalbard and Troll, up to 14 minutes each and a few to tens of GB each at the mission's rate, and a handful a day at Ohio and Singapore. On Explore a run covers 2 satellites; the Platform picks the busiest.
  3. Automations: copy Rain-fade swap and run it. Ka passes that rise low, and every Ka pass at the tropical site, lose more than half their data in the wet tail of the rain climatology. Each one comes with a later pass of the same satellite on another antenna that keeps more of it, and the deadline to cancel by. On a plan whose runs deliver, put an approval policy on it and each swap waits on you in Passes, under Waiting on you; on Explore it runs as a test run and lists them.
  4. Automations: MODCOD plan and Antenna capacity. A pass steps through several MODCODs as it rises and sets; the polar antennas carry most of the fleet's GB per day.
  5. Forecasts. SNR on the links in a pass right now (the load posts every pass of the last two hours), and weather attenuation on any station and satellite.
  6. Passes: switch to Past, then Audit, and Use a sample pass log. A week of NOAA 21 (norad-54234) and Sentinel-2C (norad-60989): most passes delivered, the ones rain put at risk failed, a few missed because their antenna was still on another pass, some delivered short and some never logged. The audit lands at a believable share delivered, and the failed passes are the ones Rain-fade swap would have moved.

What is in it

File Rows What it holds
network_config.json The stations, antennas and satellites as one network config, the body PATCH /topology/config takes
ground_stations.csv 5 Station id, WGS84 latitude, longitude and altitude, antenna count
antennas.csv One row per antenna band: antenna id, its station, dish size, efficiency, elevation mask, band edges and G/T
satellites.csv One row per radio: satellite id, NORAD catalog number, name, operator, frequency, bandwidth and EIRP
link_telemetry.parquet 2,134 The telemetry of every pass in the 2 hours before 2026-11-09T00:00:00Z: Es/N0, elevation, range, carrier frequency and system noise temperature, every 5 s
pass-log-sample.csv 500 A pass log for an audit: NOAA 21 (norad-54234) and Sentinel-2C (norad-60989) from 2026-11-02T00:00:00Z to 2026-11-09T00:00:00Z, in the CSV POST /telemetry/pass-logs takes (484 ok, 7 failed, 9 missed)
workflows/*.json The workflow templates (pass-yield, antenna-capacity, modcod-plan, rain-fade-swap, audit-ledger, rfi-watch), each the body put_workflow takes, plus demo-pass-yield.json and scenario-example.json for the quickstart
mcp/ MCP client configs for Claude Code, Claude Desktop and Cursor, and prompts that walk the story
manifest.json Each file's SHA-256, the row counts, and the ConstellationOS commit it was exported from

link_telemetry.parquet

Column Unit Meaning
time UTC When the modem reported it
ground_station_id The station (ground_stations.csv)
ground_antenna_id The antenna (antennas.csv); a link is named by its antenna
satellite_id The satellite the antenna tracks in that pass (satellites.csv)
contact_id The pass: its antenna, satellite and first step in the file
es_n0_db dB Symbol energy over noise density on the downlink
elevation_deg deg Antenna elevation above the local horizon
range_m m Slant range from antenna to satellite
frequency_ghz GHz Downlink carrier frequency
system_noise_temp_k K The receive chain's system noise temperature, the rfi noise channel
noise_diode_on Whether the calibration diode was on; always false here

The file is the telemetry a load at 2026-11-09T00:00:00Z posts, on the element sets in the export. The API makes the same thing for the two hours before you load, from the live orbit catalog, so a reload rewrites the same points and the passes are the ones that really happened.

Things to know

  • The RF terms are nominal, not datasheets. Each radio's EIRP is set so its Es/N0 is about 12 dB on X band and 17 dB on Ka band at 30 degrees elevation into a 7.3 m antenna; each bandwidth so the DVB-S2 rate there is near the mission's published downlink rate. A real downlink carries more margin on a fixed MODCOD; at these values rain moves the MODCOD, so the workflows have something to show.
  • The telemetry is synthetic. Each pass step's Es/N0 is the link's clear-sky budget at its geometry plus a planted offset of -1 to -4 dB (a known answer for calibration, not a measured loss), a slow wander and noise. When the API loads it, the last 30 minutes of each link also drop by the rain attenuation the platform holds for that station from real weather. That fade is applied at load time and is not in this file.
  • The noise channel holds a known answer for rfi. Each step's system noise temperature is the one its antenna's G/T implies, with 0.1 dB of seeded jitter. Mid-pass on every Ka pass that the file holds whole and that is long enough, unknown noise raises it by 3 dB for 60 s and Es/N0 falls by the same: the rfi model prices those steps as interference events, and every other step names no cause. X band passes are the quiet control. The API prices the noise channel only within 5 s of a sample, so a forecast for a later instant does not see it.
  • Rain risk is climatology unless a weather model is served. The workflows' weather lane is one climate everywhere, scaled by frequency and elevation, so Ka passes that rise low are the fragile ones at every site; where the deployment serves the weather model, it prices each step from the station's own weather.
  • The pass log is a story, not a record. Its passes are the ones the sim computes for the week from the element sets in the export; how each went (delivered, short, failed in rain, missed, not logged) is assigned by rule, not observed. Nothing here is any operator's network, configuration or traffic.

Where each value comes from

Value Source
Satellite names, NORAD ids, operators Published (CelesTrak)
Orbits (the element sets behind every pass) Published (CelesTrak), pinned on 2026-11-09T00:00:00Z
Downlink band Published by each mission
Carrier frequency Published where the mission gives one; nominal for Landsat 9 and Sentinel-2C
EIRP and bandwidth Nominal, set for the Es/N0 above
Station sites (latitude, longitude, altitude) Published (the public asset catalog)
Antennas: dish size, efficiency, mask, bands, G/T Nominal; singapore-teleport is the demo operator's own, and fictional
Pass times and geometry (AOS, LOS, elevation, range) Computed from the published orbits
Link telemetry: Es/N0, system noise temperature, the rfi burst Simulated
Pass log outcomes: ok, short, failed, missed, not logged Simulated, assigned by rule

Quickstart: from this page to a forecast in ten minutes

You need a free Explore account. The steps use curl, jq and the MCP endpoint at https://api.constellation.space/mcp. The API takes your key in the x-api-key header, and the MCP endpoint as Authorization: Bearer. Set your key once:

export CONSTELLATION_API_KEY="<YOUR_KEY>"

1. Get a free key

Sign up at platform.constellation.space (the Explore plan is free) and create a live key on the Keys page. It is shown once. Check it works:

curl -sS https://api.constellation.space/account/state -H "x-api-key: $CONSTELLATION_API_KEY"

The answer names your plan and this month's usage. Usage is counted, never capped: no plan limits prediction calls.

2. Load the fleet

Pick one:

  • MCP (recommended). Ask your agent to "load the demo fleet" (step 5 connects it), or call the tool directly:

    curl -sS https://api.constellation.space/mcp \
      -H "Authorization: Bearer $CONSTELLATION_API_KEY" \
      -H "Content-Type: application/json" -H "Accept: application/json, text/event-stream" \
      -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"load_demo_fleet","arguments":{}}}'
    

    The result counts the accepted records (every pass step of the last two hours, none rejected) and says whether the network config was merged. Loading is safe to repeat.

  • Platform. Use Load the demo fleet in the Platform.

  • From these files. Merge network_config.json into your network config with PATCH /topology/config (a key with config:write). The telemetry in link_telemetry.parquet is a past window, for study; load the fleet with MCP or the Platform for telemetry that ends now.

3. Wait until it is ready

The data lands within seconds. The links in a pass right now have a fresh Es/N0; list them:

curl -sS "https://api.constellation.space/topology?freshness_seconds=120" \
  -H "x-api-key: $CONSTELLATION_API_KEY"

Each record in state is a link with fresh telemetry, with fields.es_n0_db (most read 5 to 19 dB): GOES 19 to Ohio all the time, the polar orbiters only during a pass. An empty state means the load has not landed yet, so wait a few seconds and list again.

4. Forecast

SNR at the next pass of norad-54234. List its passes over the next day, then ask for SNR on that pass's antenna at its midpoint, where the satellite is well above the antenna's minimum elevation. The window starts 5 minutes from now, so the first pass starts in the future even when the satellite is overhead now: a forecast's at must be later than the time the API receives it.

FROM=$(date -u -v+5M +%Y-%m-%dT%H:%M:%SZ 2>/dev/null || date -u -d '+5 minutes' +%Y-%m-%dT%H:%M:%SZ)
TO=$(date -u -v+1d +%Y-%m-%dT%H:%M:%SZ 2>/dev/null || date -u -d '+1 day' +%Y-%m-%dT%H:%M:%SZ)

PASS=$(curl -sS -G "https://api.constellation.space/topology/passes" \
  -H "x-api-key: $CONSTELLATION_API_KEY" \
  --data-urlencode "from=$FROM" --data-urlencode "to=$TO" \
  -d step_s=60 -d satellite_id=norad-54234 | jq -ec '.items[0]') || echo "No pass listed: set AT, ANTENNA and STATION yourself"
AT=$(echo "$PASS" | jq -r '[.aos_utc, .los_utc] | map(sub("\\+00:00$"; "Z") | fromdateiso8601) | add / 2 | floor | todateiso8601')
ANTENNA=$(echo "$PASS" | jq -r .antenna_id)
STATION=$(echo "$PASS" | jq -r .ground_station_id)
echo "AT=$AT ANTENNA=$ANTENNA STATION=$STATION"

curl -sS -G "https://api.constellation.space/predictions" \
  -H "x-api-key: $CONSTELLATION_API_KEY" \
  -d model_family=snr -d ground_antenna_id=$ANTENNA -d satellite_id=norad-54234 \
  --data-urlencode "at=$AT"

The response has one item, that antenna's forecast at AT (values are illustrative):

{
  "model_family": "snr",
  "captured_at": "2026-11-12T12:10:41Z",
  "items": [
    {
      "ground_antenna_id": "tr-xka1",
      "ground_station_id": "ksat-troll",
      "satellite_id": "norad-54234",
      "target_at": "2026-11-12T12:45:22Z",
      "value": {"snr_db_p50": 13.7, "orbit": {"source": "gp", "age_h": 16.5, "stale": false}},
      "predicted_at": "2026-11-12T12:10:41Z",
      "model_version": "...",
      "feature_view_id": "...",
      "model_release_id": "...",
      "status": "uncertain",
      "reason": "link_budget_only"
    }
  ],
  "entity_errors": []
}
  • snr_db_p50 is the predicted SNR in dB.
  • status: "uncertain" with reason: "link_budget_only" marks a forecast that is the link budget alone. On Explore every SNR forecast is; Operator and up add the model's learned corrections.
  • Read entity_errors as well as items: an antenna with no forecast is listed there with a code, not returned as a zero forecast.

Weather attenuation on the links from that pass's station to the satellite, at the same instant:

curl -sS -G "https://api.constellation.space/predictions" \
  -H "x-api-key: $CONSTELLATION_API_KEY" \
  -d model_family=weather-attenuation -d ground_station_id=$STATION \
  -d satellite_id=norad-54234 --data-urlencode "at=$AT"
{
  "model_family": "weather-attenuation",
  "captured_at": "2026-11-12T12:10:42Z",
  "items": [
    {
      "ground_station_id": "ksat-troll",
      "ground_antenna_id": "tr-xka1",
      "satellite_id": "norad-54234",
      "target_at": "2026-11-12T12:45:22Z",
      "predicted_at": "2026-11-12T12:10:42Z",
      "network_config_version": 1,
      "value": {
        "attenuation_db": 0.42,
        "attenuation_rain_db": 0.0,
        "attenuation_gas_db": 0.31,
        "attenuation_cloud_db": 0.08,
        "attenuation_scintillation_db": 0.03,
        "elevation_deg": 30.75,
        "rain_source": "...",
        "nwp_source": "..."
      },
      "model_release_id": "...",
      "assumptions": []
    }
  ],
  "entity_errors": []
}

With both a station and a satellite, every antenna at the station is priced against the satellite, so this answers one item per antenna. value carries the total attenuation_db and its rain, gas, cloud and scintillation terms in dB, the path geometry, and where the weather came from.

5. Connect an AI agent (MCP)

The server is https://api.constellation.space/mcp (Streamable HTTP), authenticated with Authorization: Bearer <YOUR_KEY>. Ready-made configs are in mcp/:

  • Claude Code: run mcp/claude-code.sh with your key in place of <YOUR_KEY>.
  • Claude Desktop: merge mcp/claude_desktop_config.json into claude_desktop_config.json and restart. It needs Node.js.
  • Cursor: merge mcp/cursor_mcp.json into ~/.cursor/mcp.json.

Then try the prompts in mcp/prompts.txt, for example "Which demo links are in a pass right now, and what is their Es/N0?". The agent calls load_demo_fleet, get_topology, get_predictions, get_usage and the workflow tools below with your key, inside your plan's limits. Keep the config files private: anyone with the key can act as you.

6. Run a workflow

A workflow is a saved graph that prices every pass of your fleet. The quickstart copies the pass-yield template (expected GB per pass from the physics SNR and weather lanes) as demo-pass-yield, limited to norad-54234 and norad-60989, because an Explore run covers at most 2 satellites. workflows/demo-pass-yield.json is that copy; workflows/pass-yield.json is the template as is.

Ask your agent: "Save workflows/demo-pass-yield.json as demo-pass-yield, create the scenario in workflows/scenario-example.json, run demo-pass-yield against it over the next 48 hours, and show me the best passes." Or with curl:

H=(-H "x-api-key: $CONSTELLATION_API_KEY" -H "Content-Type: application/json")

# Save the workflow (put_workflow)
curl -sS -X PUT https://api.constellation.space/predictions/workflows/demo-pass-yield "${H[@]}" -d @workflows/demo-pass-yield.json

# Create the scenario (create_scenario): ksat-troll down, 3 dB extra fade at singapore-teleport
curl -sS -X POST https://api.constellation.space/topology/scenarios "${H[@]}" -d @workflows/scenario-example.json
# -> {"scenario_id": "...", ...}

# Run it over the next 48 hours (run_workflow)
FROM=$(python3 -c "from datetime import UTC, datetime; print(datetime.now(UTC).isoformat(timespec='seconds'))")
TO=$(python3 -c "from datetime import UTC, datetime, timedelta; print((datetime.now(UTC) + timedelta(hours=48)).isoformat(timespec='seconds'))")
curl -sS -X POST https://api.constellation.space/predictions "${H[@]}" \
  -d "{\"workflow\": \"demo-pass-yield\", \"from\": \"$FROM\", \"to\": \"$TO\", \"scenario_id\": \"<scenario_id>\"}"
# -> {"run_id": "...", "status": "queued"}

# Read the result (get_run) until status is done
curl -sS "https://api.constellation.space/predictions?run_id=<run_id>" "${H[@]}"

A finished run has "status": "done" and one item per pass: entity (the pass, its antenna and satellite), target_at and until (AOS and LOS), value (expected GB at p10, p50 and p90) and explain (each lane's value and where it came from). A run against a scenario is a what-if: it prices and stores its items and sends nothing. Without a scenario, an Explore run must be a test run: add "deliver": false. Workflow runs and reads do not spend predictions.

If your agent does not list list_workflows, workflows are not enabled for your account yet; steps 1 to 3 and 5 still work, and in step 4 you set AT, ANTENNA and STATION yourself, because listing passes needs workflows.

7. Audit a pass log

An audit sets the passes your fleet flew, as your pass log records them, against every pass it could have flown. pass-log-sample.csv is a pass log for the demo fleet, one row per contact: satellite_id, antenna_id, aos_utc, los_utc, bytes_delivered and status (ok, failed or missed). It logs NOAA 21 (norad-54234) and Sentinel-2C (norad-60989) from 2026-11-02T00:00:00Z to 2026-11-09T00:00:00Z: every pass an audit of that week finds, the ones the rain climatology put at risk failed, the ones whose antenna was still on another pass missed, and the rest delivered, a few of them short and a few not logged at all. Every row names an antenna and satellite of the demo fleet, so load the fleet first: a pass log is matched against your fleet, and a row naming anything else is refused.

curl -sS -X POST https://api.constellation.space/telemetry/pass-logs -H "x-api-key: $CONSTELLATION_API_KEY" \
  -H "Content-Type: text/csv" --data-binary @pass-log-sample.csv

Then run the audit-ledger template with "dataset": "audit" over 2026-11-02T00:00:00Z to 2026-11-09T00:00:00Z, limited to NOAA 21 (norad-54234) and Sentinel-2C (norad-60989), the week and satellites the log covers: an audit of any other week finds none of its contacts. The Platform's Audit page does all of this in a few clicks, and its "Use a sample pass log" builds the same kind of log for the past week.

What Explore covers

Explore (free) Operator
Load the demo fleet, topology, MCP Yes Yes
SNR and weather attenuation forecasts Yes Yes
Workflow runs Test runs and scenarios, physics-only on all five lanes, at most 2 satellites Learned corrections, all lanes, the whole fleet
Model lanes (demand, RFI, SSA) in workflows, so the rfi-watch template Yes Yes
Runs that deliver to a feed or destination No Yes

Where the deployment serves the demand and SSA models, they price the five low orbit satellites' passes: they fly at 700 to 830 km, inside SSA's 400 to 2,000 km. get_network_config shows each satellite's readiness per lane.

Use it in Python

from datasets import load_dataset

satellites = load_dataset("constellationspace/demo-fleet-v2", "satellites", split="train").to_pandas()
print(satellites[["name", "norad_id", "frequency_ghz", "bandwidth_mhz", "eirp_dbw"]])

License

CC BY 4.0. Credit "Constellation Space" when you use it.

Downloads last month
119