I second this bug report. I have been able to consistently reproduce it just by watching the AI play.

Hello folks,

Thank you for your report and the save files, this is really helpful.

When a construction is cancelled, the invested industry is not lost. It is converted into a surplus that can be used for another construction. So the AI doesn't actually loose anything (as long as it queues enough construction to consume all that surplus at the end of the turn). We could argue that, most of the time, finishing a construction providing Industry is better than cancelling it and use the surplus for something else. This could be a good improvement, I'll think about it.

I've also noticed something odd in the decision making. The Makers Quarter was queued because the Industry net was evaluated as too low. Then, due to a narrative event, the Industry net gets a boost. That boost is high enough to make the Industry need drop significantly, hence the decision to queue units instead. I don't think that's a good decision. This temporary boost should probably be ignored by the AI when it comes to prioritizing constructions. That could be a great improvement too.

Cheers,

Alex.

Given a situation, the AI will assess how well a city is performing for each FIMS production. The lower the resource production, the higher it prioritizes improving this resource production. Constructions are then given a score, depending on the gains they can provide (among other things...) and the AI pick the one with the highest score. Affinity can provide a boost on resources (eg. Industry receives a slight boost when the AI has a Builder affinity). An on-going construction can then be canceled when the AI finds a construction with a higher score. So, indirectly, yes, affinities are taken into account when cancelling a construction for another. But the AI shouldn't be more likely to cancel a construction depending on its current affinity.

I'd be really interested in having a look at the Italian situation you're mentionning. We noticed the AI had a hard time managing its Public order due to poor City cap management. That could lead to the kind of situations you're describing. I've worked on the City cap management, which should now be under control, but I may have missed another source of disastrous Public order management. From what I see in the data, Common Quarter could be considered as a light source of Influence for the AI but I don't think it would be enough to make it want to spam that district like that.