networkx kosaraju_strongly_connected_components(G, source=s) returns sets that are not strongly connected (is_strongly_connected False)
Symptom: passing source= to nx.kosaraju_strongly_connected_components can yield a set that is not a strongly connected component; without source the output is correct. networkx 3.7:
G = nx.DiGraph([(0,1),(1,0),(0,2),(2,3),(3,2)])
list(nx.kosaraju_strongly_connected_components(G, source=0)) # [{0, 1, 2}]True SCCs are {0,1} and {2,3} (everything is reachable from 0, so the reachable subgraph has the same SCCs). nx.is_strongly_connected(G.subgraph({0,1,2})) is False: 2 cannot reach 0, and the real component {2,3} is split. A random sweep (2,000 small gnp digraphs, every node as source) found a wrong set in 2,001 of 8,061 cases, so it is not an edge case. This silently corrupts cycle/ring detection in transaction or social graphs.
Fix: do not pass source. Use nx.strongly_connected_components(G) and pick the component containing your node, or run it on G.subgraph(nx.descendants(G, s) | {s}).
Related: https://github.com/networkx/networkx/issues/9003