fern-go: error bodies are `any` for OpenAPI-sourced specs, and group names ignore domain acronyms

[adidavid014]

Two smaller findings from the same customer-SDK review. Filing together since both are "the generated output is right but not as good as it could be", and both are unreachable from config today.

1. Error bodies are any for OpenAPI inputs

Fern types error bodies for Fern-definition inputs — seed/go-sdk/errors produces:

type BadRequestError struct {
    *core.APIError
    Body *ErrorBody
}

But an OpenAPI spec whose error response is a named $ref produces:

type BadRequestError struct {
    *core.APIError
    Body any
}

Our overlay attaches an explicit 400 carrying $ref: '#/components/schemas/PlaidError', and PlaidError is a named schema present in the SDK as plaid.PlaidError. Callers still get map[string]any and must string-index error_code / error_type.

Things I ruled out:

• Not caused by default coexisting with explicit codes — removing the default response via overlay changes nothing.
• Not a missing-schema problem — PlaidError is generated as a type.
• I could not find any x-fern-* extension for naming an error schema.

Expected: an OpenAPI error response pointing at a named schema should produce Body *ThatType, matching the Fern-definition behavior.

2. Group names ignore domain acronyms, and there's no override

smart-casing: true handles a fixed initialism list but not domain acronyms, and x-fern-sdk-group-name is taken verbatim. Measured on 1.57.2:

supplied	generated
api_key	APIKey ✅
url_thing	URLThing ✅
oauth	Oauth ❌
o_auth	OAuth ✅ (workaround)
fdx	Fdx ❌
f_d_x	FDx ❌
cra_report	CraReport ❌

So OAuth is reachable only by misspelling the group as o_auth, and FDX / CRAReport / EWAReport are not reachable at all. We shipped client.OAuth via the workaround and left client.Fdx / client.CraReport as-is.

Expected: either a configurable acronym list (e.g. additionalAcronyms: [FDX, CRA, EWA]) or a per-group casing override. The workaround is obscure enough that a future reader would likely "fix" the spelling and silently regress it.

[adidavid014]

Closing in favour of an internal ticket (FSE-86) — I'm on the Fern team, so this belongs in our tracker rather than the public issue list. No content lost; the ticket carries the full findings including the expanded casing measurements (uppercase inputs like FDX also do not survive, so this is not fixable via x-fern-sdk-group-name).