j:SrSSKJr SSKJr SSKJrJrJrJ r SSK J r J r SSK Jr SSKJr SSKJrJrJr SS KJrJr SS KJr SS KJr SS KJrJr S r\"SS9SSj5r SSSjjr!SSjr"SSSjjr#g)uAnswering into a fixed envelope whose payload is a closed set of shapes. The agent always returns `Answer`. It chooses the payload — scalar, series, grid, prose, unanswerable — from the union, so deciding whether the question wants a figure or an explanation happens inside the same call that answers it. No separate classification step, and no schema is invented at runtime: every shape a caller can receive is written down in `shapes`. ) annotations) lru_cache)AnyDictListOptional)Agent NativeOutput) ModelSettings)usage_metadata_from) DEFAULT_MODEL answer_model build_context)DocumentIdentity with_document)Answer)RetrievedChunk) run_metadatatraceruYou are a financial-document analyst answering questions about ONE parsed document, from a set of context chunks retrieved from it. Answer ONLY from the provided context. Never use outside knowledge, never invent figures. If the context does not hold the answer, return the `unanswerable` payload. CHOOSING THE PAYLOAD Read what the question asks for and return that shape: - `scalar` — the question names ONE figure. This is the common case for questions phrased like "Return the Q1 2026 value" or "What was X as of ". Put the figure in `value` exactly as the source prints it, with its currency sign and separators, and NOTHING else. "$14.47" — not "$14.47 per share", not "Book value was $14.47", not "$14.47 as of March 31, 2026". The qualifiers belong in `unit`, `period`, and `label`. - `series` — the question asks for one measure across several periods, segments, or categories. - `grid` — the question asks for a whole table. - `prose` — the question asks you to explain, compare, summarize, or reason. Only here may you write sentences. - `unanswerable` — the context does not contain what was asked. When a question could be read either way, prefer the narrower shape: if a single figure answers it, return `scalar`. GROUNDING Populate `cited_ids` with the ids supporting your answer. Each context chunk begins with `[chunk id= | page | type ]`. Cite the chunk ``, or — for tighter grounding — the finer-grained element ids exposed inside a chunk's content, such as table cells tagged ``. On a `scalar`, also set `source_id` to the single cell or chunk the figure was read from, and make `value` character-for-character what that cell prints. Only cite ids that appear verbatim in the context. Never invent or construct an id. )maxsizec Z[[5[[5[[ SS9S9$)Ng) temperature) output_type system_promptmodel_settings)r rr r SYSTEM_PROMPTr - 8 / +11&Qjj&1   2s,c# [XU5n[5RS[X15[[ 5S9IShvN n[ 5RU5IShvN nURnSUR5S./[UR5S.Ul SSS5IShvN U$NrNSN !,IShvN (df  W$=f7f)N answer_fixed)modelextra_metadata assistantr*)r.usage_metadata) r'rllm_runr2r rr!runoutputmodel_dump_jsonr usageoutputs)r#r$r%r&r:resultr;s r answerr@bs( 3Fx V4MZfZh  "}((00"-&:P:P:RST1&,,?   M1 MsZACB5CB;#B7$?B;# C.B9/C7B;9C;CC  CC)returnzAgent[None, Answer])N)r#strr$List[RetrievedChunk]r%Optional[DocumentIdentity]rArB)r&rBr$rCrAzDict[str, Any])r#rBr$rCr%rDrAr)$__doc__ __future__r functoolsrtypingrrrr pydantic_air r pydantic_ai.settingsr quber.agents.langsmith_tracerr quber.playground.agentr rr!quber.playground.answers.documentrrquber.playground.answers.shapesrquber.playground.retrievalrquber.playground.tracingrrrr!r'r2r@rrr rQs#,,+.=MMM259' T 1Y]++/+;U++Y]  / ;U   r