app/services/cross_validation_numeric_runner.py

from __future__ import annotations

from app.schemas.cross_validation import AnswerKind, TaskType
from app.schemas.numeric import NumericCheckItem
from app.schemas.pedagogy import PedagogicalContext
from app.services.cross_validation_numeric_adapter import cross_validation_report_to_numeric_check
from app.services.math_cross_validation_orchestrator import run_cross_validation_with_wolfram_reference
from app.services.math_llm_cross_validation import solve_exercise_independently_with_llm, translate_exercise_to_wolfram_with_llm


async def run_solution_set_cross_validation_numeric_check(
    *,
    answer_text: str,
    pedagogical_context: PedagogicalContext,
) -> NumericCheckItem:
    expected_kind = AnswerKind.SOLUTION_SET

    wolfram_candidate = await translate_exercise_to_wolfram_with_llm(
        pedagogical_context=pedagogical_context,
        expected_answer_kind=expected_kind,
    )

    candidate_kind = getattr(wolfram_candidate, "expected_answer_kind", expected_kind)
    candidate_task = getattr(wolfram_candidate, "task_type", TaskType.EQUATION)

    if candidate_kind != expected_kind or candidate_task != TaskType.EQUATION:
        actual_kind = getattr(candidate_kind, "value", str(candidate_kind))
        actual_task = getattr(candidate_task, "value", str(candidate_task))
        return NumericCheckItem(
            check_type="equivalence",
            expression="cross_validation_skipped",
            expected=expected_kind.value,
            is_valid=True,
            details=f"cross_validation_skipped_incompatible_task={actual_task}; answer_kind={actual_kind}",
            sympy_status="skipped",
            wolfram_status="skipped",
            cas_status="NO_CAS_VALIDATION",
            preferred_engine="none",
            confidence_impact="neutral",
            human_review_reason=None,
        )

    llm_answer = await solve_exercise_independently_with_llm(
        pedagogical_context=pedagogical_context,
        expected_answer_kind=expected_kind,
    )

    report = await run_cross_validation_with_wolfram_reference(
        answer_text=answer_text,
        expected_answer_kind=expected_kind,
        llm_answer=llm_answer,
        wolfram_query=wolfram_candidate.wolfram_query,
    )

    return cross_validation_report_to_numeric_check(report)