I almost hate to spoil the fun here, but ... > Thanks for your input Larry - I totally get that relying on undocumented behaviour is not a good strategy! > The thrust of my argument now is that the current behaviour should be fixed and documented. Yes, I understood that. What is documented is that an ORDER clause enforces some kind of ordering. If feasible, it will be optimized away where the implementation naturally (in its current evolution) achieves the specified ordering. You can rely on that. The extra guarantee you seek/suggest is highly unlikely to be made by the present SQLite project. In part this is because VDBE code generation is complex enough without imposing requirements which are contrary to long-standing SQL convention. And, I venture, further cause to decline your invitation is that it would weaken the message: If you want ordering, use ORDER BY. The docs have been quite clear that, absent an ORDER BY clause, no ordering is guaranteed. I do not foresee an ever-growing list of exceptions appearing with those warnings. This stance on ordering is especially important for input to the windowing functions, where ordering (or misordering) is baked into the result rather than merely affecting result ordering.