Yes, this is accurate. You often end up with many solutions and human judgement is sometimes required in the end to pick the right strategy. However, what I hav

Yes, this is accurate. You often end up with many solutions and human judgement is sometimes required in the end to pick the right strategy. However, what I have found in practice is that heuristics and hinting can rapidly cut through that complexity. E.g. "always pick the shortest path between tables, and never use this set of tables as intermediate nodes on any path" rules out a ton of options, and usually will leave you with one or two, and usually those are the natural choices. In this way you can use polynomially-many constraints or rules to avoid exponentially-many weird or exceptional routes through the schema. I am optimistic that you can build a system where by default, the autojoin solution is the natural one maybe 90% of the time. There will certainly be exceptions where you have to express the join conditions explicitly. But I think you can dramatically reduce the amount of code required. I would also hazard the suggestion that this might produce productive backpressure on the system. If the autojoiner is struggling to find a good route through the schema, it's possible that the schema is not properly normalized or otherwise messed up.