These are our internal representations?

Why don't we just preserve the ones returned from Bedrock?

They sort of forced our hand, although I could've kept to it a little more faithfully.

Concatenation of their delta blocks won't result in a full content block. They're structurally dissimilar, so I need to translate those to make that work.

Otherwise they discriminate their content block types via nesting them under mutually exclusive keys on the content block object, while we use the type field as a discriminator. I'm just translating from their discriminator scheme to ours here. I suppose I could've made them be { type: "reasoningContent", reasoningContent: { ... } } instead of spreading the reasoningContent object there directly into the block, but that felt like poor design.

Ok - only other concern then is to try to keep it inline with changes in Python as well

Not sure who's picking up the python one, but we're consciously not forcing structural similarity between the two libs anymore unless there's a good reason to (e.g. they need to interop). Winds up causing too many pythonic conventions that feel weird in JS.

Edit, to elaborate for external parties: The goal behind this decision is to make the JS lib more predictable (and therefore easier to adopt and use) for JS/TS devs. We'll still aim to keep things functionally and conceptually similar, and if a design works well for both sides there's no reason to waste cycles redoing it for the other. We only want to be forcing close similarity in cases when doing so adds more direct value to users than the alternative would.