Proposal: Triangle data-quality diagnostics

[son1403]

I would like to discuss a small triangle quality-check helper.

Before selecting development factors, I usually check for incomplete cells, negative values, non-monotonic cumulative development, sparse rows or columns, and unusually volatile age-to-age factors. These checks are repetitive and easy to apply inconsistently across triangles.

A first version could return a structured, read-only result for the selected checks. It should flag the data and explain why it was flagged, without changing the triangle or prescribing actuarial treatment.

The earlier prototype is in #1246, which I have closed so the API and scope can be discussed here first.

[henrydingliu]

great idea!

to move forward, i would recommend opening sub-issues for each diagnostic as a property. some of the individual diagnostic proposed here would appear duplicative, but we can vet them out in their respective sub-issues. then we can add an aggregator property that returns all of them.

[kennethshsu]

I like all these ideas except the "unusually volatile age-to-age factors"? Can't this easily be accomplished with .heatmap()? Also, it's hard to give a standard definition on above/below what value is volatile or not?

• "incomplete cells" as in the NaN issue?
• "negative values": good!
• "non-monotonic cumulative development": good!
• sparse rows or columns: and diagonals?