"""A real consult over the normalized route (L9).

The point is the prompt-level contract: a live model must produce something
that parses against our local schema, since no API-level enforcement exists.
"""

import io
from pathlib import Path

import pytest

from nisse_core.loop.expert import BudgetExhausted, ExpertDesk
from nisse_core.models.llm import LiteLLMClient, RoleMap
from nisse_core.oplog import OpLog
from nisse_dev.evals import collect
from nisse_dev.state import StateStore

pytestmark = pytest.mark.integration


@pytest.fixture
def desk(tmp_path: Path, litellm: LiteLLMClient, live_model: str):
    def make(budget: int = 2) -> tuple[ExpertDesk, StateStore]:
        model = live_model
        roles = RoleMap(driver=model, helper=model, expert=model)
        store = StateStore(
            tmp_path / "state.db", oplog=OpLog("nisse-dev", io.StringIO())
        )
        return ExpertDesk(litellm, roles, budget=budget), store

    return make


def test_live_consult_parses_against_the_local_schema(desk) -> None:
    expert, store = desk()

    answer = expert.ask(
        store.transcript("t", "expert"),
        "For a single-process agent storing a few thousand rows of task state, "
        "is SQLite sufficient, or is PostgreSQL warranted? Answer briefly.",
    )

    assert answer.confidence in {"high", "medium", "low"}
    assert len(answer.answer) > 20
    # The consult is captured for L9 without any extra bookkeeping.
    [record] = collect(store)
    assert record.role == "expert"
    assert record.exit_status == "answered"


def test_live_budget_stops_a_circling_caller(desk) -> None:
    expert, store = desk(budget=1)
    expert.ask(
        store.transcript("t", "expert"),
        "Name one advantage of WAL mode in SQLite. Be brief.",
    )
    with pytest.raises(BudgetExhausted):
        expert.ask(store.transcript("t", "expert"), "And another?")