We present a sound and complete algorithm, called iterative causal discovery\n(ICD), for recovering causal graphs in the presence of latent confounders and\nselection bias. ICD relies on the causal Markov and faithfulness assumptions\nand recovers the equivalence class of the underlying causal graph. It starts\nwith a complete graph, and consists of a single iterative stage that gradually\nrefines this graph by identifying conditional independence (CI) between\nconnected nodes. Independence and causal relations entailed after any iteration\nare correct, rendering ICD anytime. Essentially, we tie the size of the CI\nconditioning set to its distance on the graph from the tested nodes, and\nincrease this value in the successive iteration. Thus, each iteration refines a\ngraph that was recovered by previous iterations having smaller conditioning\nsets -- a higher statistical power -- which contributes to stability. We\ndemonstrate empirically that ICD requires significantly fewer CI tests and\nlearns more accurate causal graphs compared to FCI, FCI+, and RFCI algorithms\n(code is available at https://github.com/IntelLabs/causality-lab).\n