Models

Context is the window available under normal capacity. Under heavy load a request may be served with a smaller window; when that happens the request returns an explicit error rather than silently dropping your input. If you rely on the full window, handle that error and retry.