The Claude engine passes the key test, but every chapter fails with "Empty response" and my Anthropic credit is still used. What is wrong?

Some Claude models think before they write, such as Claude Fable 5, and the thinking counts against the text budget. Before version 1.9.23 they could spend a chapter's whole budget thinking and return no text, and BookyAI then asked again several times, each try billed by Anthropic. The key test still passed, because it only checks that the key works. From version 1.9.23 these models get room to think and still write, an answer with no text is tried at most once more and never repeated, and the message says what happened. If it still happens, choose a Claude model that writes without thinking first, such as claude-opus-4-8 or claude-sonnet-5, in Settings, AI engines, Claude.