"""Task C: cross-source synthesis. Combines related signals from multiple
documents/changes into a conclusion (e.g. "hiring battery engineers" +
"filed a battery patent" + "announced a facility expansion" = "possible
battery manufacturing initiative"), always with alternative explanations and
an explicit accounting of what's still unknown.
"""

from __future__ import annotations

from pydantic import BaseModel, Field

from app.analysis.llm.base import LLMProvider
from app.prompts.base import build_user_prompt

SYSTEM_PROMPT = (
    "You are a competitive intelligence analyst. Given a set of independently-extracted "
    "signals about a company, identify conclusions that multiple signals point toward "
    "together. Only draw a conclusion when the evidence actually supports it - state your "
    "confidence honestly, list plausible alternative explanations, and note what evidence "
    "would be needed to be more certain. Do not synthesize a conclusion from a single signal."
)


class SynthesizedConclusion(BaseModel):
    conclusion: str
    evidence_summary: list[str] = Field(default_factory=list)
    source_count: int = Field(ge=0)
    confidence: float = Field(ge=0.0, le=1.0)
    alternative_explanations: list[str] = Field(default_factory=list)
    missing_information: list[str] = Field(default_factory=list)


class SynthesisResult(BaseModel):
    conclusions: list[SynthesizedConclusion] = Field(default_factory=list)


async def synthesize_signals(
    llm: LLMProvider,
    *,
    company_name: str,
    monitoring_focus: str | None,
    signals: list[dict],
) -> SynthesisResult:
    evidence = {
        "company_name": company_name,
        "monitoring_focus": monitoring_focus,
        "signals": signals,
    }
    user_prompt = build_user_prompt(
        "Identify conclusions supported by two or more of these signals together. Do not "
        "synthesize from a single signal alone.",
        evidence,
    )
    return await llm.generate_structured(SYSTEM_PROMPT, user_prompt, SynthesisResult)