j0%SrSSKJr SSKJr SSKJrJrJrJ r J r J r J r SSK JrJrJr SSKJrJr SSKJr SSKJr SS KJr SS KJrJr S r\ S r\\ \\\\ S .r!S \"S'Sr#"SS\5r$"SS\5r%\"SS9SSj5r&SSjr'SSjr(g)uAnswering into a schema built for the question that was asked. Two calls. A planner reads the question alone — never the context, never the answer — and declares the fields the answer should have. Those declarations are compiled into a Pydantic model, and the answering agent is constrained to it through the same call `declared` uses. The only difference between the two approaches is who authored the model. The planner emits field declarations rather than raw JSON Schema. A model writing free-form JSON Schema produces dialects the validator then rejects, and that surfaces as a retry loop rather than an answer; a closed set of field types cannot fail that way. ) annotations) lru_cache)AnyDictListLiteralOptionalTupleType) BaseModelField create_model)Agent NativeOutput) ModelSettings)usage_metadata_from)anthropic_model) run_metadatatracerzclaude-haiku-4-5)stringnumber string_list number_listzDict[str, Any] _PY_TYPESuYou design the return shape for a financial-document question. You are given the QUESTION only — never the document, never the answer. Do not attempt to answer it. Decide first whether the question asks for data or for explanation. - If it asks you to explain, compare, summarize, or reason, set `wants_value` false and return no fields. It will be answered in prose. - Otherwise set `wants_value` true and declare the fields that hold the answer, and only those. Rules for fields: - Declare the narrowest set that answers the question. A question naming one figure gets ONE field holding that figure, plus at most the qualifiers the question itself names. - Name fields in snake_case after what they hold. - Each field's description must tell the answering agent to report the figure exactly as printed in the source, with no surrounding words. - Never declare a field for citations or sources; those are added for you. - Never declare a field whose value would be a sentence. cf\rSrSr%\"SS9rS\S'\"SS9rS\S'\"S S9rS\S 'S r S \S 'Sr g) FieldSpec?zsnake_case field name. descriptionstrnamezThe field's type. FieldTypetypez'What the answering agent must put here.rTboolrequiredN) __name__ __module__ __qualname____firstlineno__r r!__annotations__r#rr%__static_attributes__r&>/home/mande/repo/quber/src/quber/playground/answers/planned.pyrr?s<":;D#;(;rr@runoutputmodel_dump_jsonrusageoutputs)questionpromptinputsrMresultrNs r.plan_forrVYs( $F. 9 / F xv]<>  z~~f--}}"-&:P:P:RST1&,,?   M. MsZACB5CB;#B7$?B;# C.B9/C7B;9C;CC  CCc0nURHnURS;aM[URnUR(a#U[ UR S94XR'M\[U[ SUR S94XR'M [[[ [SS94US'[[[ SSS94US '[S 0UD6$) aLTurn a plan's field declarations into a model the answer agent fills. `cited_ids` and `not_found` are appended to every compiled model, so grounding and the not-present case survive whatever the planner declared. A plan that declared a field by either of those names loses it to the appended one, which is the intended precedence. ) cited_ids not_foundrN)defaultrz,Chunk and/or cell ids supporting the answer.r3rXz5Set only when the context lacks the answer; say what.rY) PlannedAnswer) r5r!rr#r%r rr rr r6r)planr5specpys r. compile_modelr_ms*,F  992 2  tyy ! ==!#Ut7G7G%H IF99 !)"uTtO_O_/` aF99  S  d0^_F;   d(_`F;  26 22r-N)returnzAgent[None, Plan])rRr r`r0)r\r0r`zType[BaseModel]))__doc__ __future__r functoolsrtypingrrrrr r r pydanticr r r pydantic_airrpydantic_ai.settingsrquber.agents.langsmith_tracerrquber.playground.agentrquber.playground.tracingrrr>r"r floatrr+r?rr0r@rVr_r&r-r.rls #BBB33+.=29" D E 9;  >0 9 1(3r-