# OrbitStack

> LEO-aware congestion control (LeoAware) + optional ASCENT satellite state for Starlink-class paths.
> Aimed at engineers shipping traffic over LEO (including high-volume consumer apps).

## Measured headline (leo_fast_ho, 90s, 7 handovers)

| CCA | Goodput | p95 RTT |
|-----|---------|---------|
| CUBIC | 6.44 Mbps | 134.3 ms |
| BBRv3approx | 65.38 Mbps | 188.6 ms |
| LeoAware | 68.31 Mbps | 164.0 ms |

Terrestrial control: LeoAware 77.98 Mbps vs BBR approx 78.81 Mbps (no LEO-only regression).

Sim: discrete-time path model with handover RTT/capacity redraw + non-congestive loss bursts.
Educational BBR approx - not bit-exact production BBRv3.

## Components

1. LeoAware - endpoint reconfiguration detection, sample invalidation, mobility vs congestion
2. ASCENT - optional structured telemetry/agent control plane (https://ascent.jonbailey.xyz/)
3. Integration - quiche-shaped on_ack / on_loss / on_path_hint

## Links

- Site: https://orbitstack.jonbailey.xyz/
- Code (MIT): https://github.com/Pitchfork-and-Torch/leo-aware-transport
- ASCENT: https://ascent.jonbailey.xyz/
- SkyCache: https://skycache.jonbailey.xyz/

Reproduce: pip install -r requirements.txt && python -m experiments.run_suite

Not affiliated with X, Cloudflare, or SpaceX. Research brief.