Answer in the user's language

[adamjohnwright]

Spec 004, stages 1 and 2. detected_language reaches the answer prompt in React-to-Me and Plant Reactome, as its own variable.

What was going wrong

Reproduced before fixing:

step	result
question	Quel role joue TP53 dans l'apoptose ?
detected	French ✓
rephrased for retrieval	What role does TP53 play in apoptosis? ✓
retrieval	good ✓
answer	English ✗

Four steps of five already worked. chat_history still held the original French question and the model answered in English anyway — so it was never that the model couldn't tell; nothing asked it to reply in French.

Why a prompt variable and not #140's approach

create_retrieval_chain passes only input to the retriever:

retrieval_docs = (lambda x: x["input"]) | retriever

So a separate variable cannot reach BM25, the vector store, or the query expansion in front of them. #140 appended its instruction to input; measured through the whole retriever, that changes about half the fused documents.

A test asserts the retrieval query is byte-identical across four languages — exact, where a retrieval baseline would only show noise and cost an hour of API time.

The instruction wording is @bleedblack1's, kept because it gets the hard part right: SET, MAX and CAT are gene symbols and ordinary English words, and R-HSA-9612973 means nothing translated.

The adversarial review caught my plan contradicting my spec

FR-007 promised English questions "no additional prompt content", while the plan always passes the language — which does add a sentence to every English prompt. FR-007 was the wrong half: it now promises no extra model call and a byte-identical retrieval query, and states the added sentence plainly. Branching instead would leave the common path exercised only by users who never ask in English.

Verified against the Release95 bundle

	answer	R-HSA ids	docs
French question	French	preserved	40
English question	unchanged English	preserved	40

Reverting the call site fails the test that covers it. 231 tests pass.

Worth watching: the French answer carried 2 R-HSA citations against English's 9. One sample, not a finding — but the kind of thing an answer-quality run should look at.

#140 and #125 close once this lands.

🤖 Generated with Claude Code