backend/tests/test_expectation_loop.py

"""Expectation quality loop pure control plane (no LLM)."""
from __future__ import annotations

from agents.world_class_advisor import evaluate_application
from core.generation.expectation_loop import (
    is_regulation_ready,
    run_advisor_expectation_loop,
    run_expectation_loop,
)

BRIEF = {
    "key_rules": ["Wnioskodawca musi posiadać status MŚP", "Projekt musi spełniać zasadę DNSH"],
    "required_sections": ["Opis projektu", "Budżet"],
    "required_attachments": [],
    "attention_points": ["Sprawdź DNSH / wpływ środowiskowy w opisie projektu."],
    "usable": True,
}


def test_is_regulation_ready_false_for_structure_only():
    assert (
        is_regulation_ready(
            {
                "passed": True,
                "score": 95,
                "regulation_grounded_pass": False,
                "blockers": [],
                "grounding_mode": "structure_only",
            }
        )
        is False
    )


def test_loop_stops_when_ready():
    sections = {
        "Opis projektu": (
            "Projekt MŚP z DNSH i pełnym opisem innowacji. " * 10
        ),
        "Budżet": (
            "Budżet z wkładem własnym i kosztami kwalifikowalnymi. " * 10
        ),
    }

    def evaluate(state):
        return evaluate_application(
            sections=state.get("generated_sections") or {},
            brief=BRIEF,
            grounding_mode="regulation",
        )

    result = run_expectation_loop(
        evaluate=evaluate,
        apply_fixes=None,
        initial_state={"generated_sections": sections},
        max_iterations=3,
        min_score=70,
    )
    assert result.ready is True
    assert result.stop_reason == "ready"
    assert result.regulation_grounded_pass is True
    assert result.iterations >= 1
    assert not result.remaining_blockers


def test_loop_max_iter_with_blockers():
    weak = {"Opis projektu": "za mało"}

    def evaluate(state):
        return evaluate_application(
            sections=state.get("generated_sections") or {},
            brief=BRIEF,
            grounding_mode="regulation",
        )

    def fix(state, report):
        # Improves slightly but not enough to cover Budżet
        gen = dict(state.get("generated_sections") or {})
        gen["Opis projektu"] = (gen.get("Opis projektu") or "") + " dodatek MŚP DNSH " * 5
        state = dict(state)
        state["generated_sections"] = gen
        state["fixed_sections"] = ["Opis projektu"]
        return state

    result = run_expectation_loop(
        evaluate=evaluate,
        apply_fixes=fix,
        initial_state={"generated_sections": weak},
        max_iterations=2,
        min_score=70,
    )
    assert result.ready is False
    assert result.stop_reason == "max_iter"
    assert result.iterations == 2
    assert result.remaining_blockers
    assert result.regulation_grounded_pass is False


def test_loop_structure_only_never_ready():
    sections = {
        "Opis projektu": "x" * 250,
        "Budżet": "y" * 250,
    }

    def evaluate(state):
        return evaluate_application(
            sections=state.get("generated_sections") or {},
            brief=BRIEF,
            grounding_mode="structure_only",
        )

    result = run_expectation_loop(
        evaluate=evaluate,
        apply_fixes=lambda s, r: s,
        initial_state={
            "generated_sections": sections,
            "external_context": {"grounding_mode": "structure_only"},
            "grounding_mode": "structure_only",
        },
        max_iterations=2,
    )
    assert result.ready is False
    assert result.regulation_grounded_pass is False
    assert result.stop_reason in ("max_iter", "blocked_grounding")


def test_quality_expectation_step_wired_for_production():
    """Production helper run_quality_expectation_step drives real advisor + fix path."""
    from core.generation.quality_loop import run_quality_expectation_step

    state = {
        "sections_plan": [
            {"title": "Opis projektu", "type": "desc"},
            {"title": "Budżet", "type": "budget"},
        ],
        "generated_sections": {
            "Opis projektu": "Pełny opis projektu MŚP z DNSH i celami. " * 8,
            "Budżet": "Budżet z wkładem własnym 30% i kosztami kwalifikowalnymi. " * 8,
        },
        "external_context": {
            "grounding_mode": "regulation",
            "advisor_brief": BRIEF,
            "required_sections": BRIEF["required_sections"],
            "key_rules": BRIEF["key_rules"],
        },
    }
    out = run_quality_expectation_step(state, source="holistic", min_score=70)
    assert "advisor_before" in out
    assert "regulation_ready" in out
    assert out["stop_reason"] in ("ready", "needs_retry")
    # Strong content with brief signals should be ready without needing rewrite
    assert out["regulation_ready"] is True
    assert out["stop_reason"] == "ready"


def test_soft_pass_gate_requires_regulation_grounded_pass_semantics():
    """Mirror generator gate: usable brief → only regulation_grounded_pass is ready."""
    from core.generation.expectation_loop import is_regulation_ready

    fluff_rep = evaluate_application(
        sections={
            "Opis projektu": "Projekt innowacyjny z bogatym doświadczeniem zespołu. " * 20,
            "Budżet": "Budżet obejmuje koszty osobowe i sprzęt w pełnym zakresie. " * 20,
        },
        brief=BRIEF,
        grounding_mode="regulation",
    )
    assert fluff_rep.regulation_grounded_pass is False
    assert is_regulation_ready(fluff_rep) is False


def test_run_advisor_expectation_loop_improves_to_ready():
    state = {
        "generated_sections": {
            "Opis projektu": "krótki start",
        },
        "external_context": {
            "grounding_mode": "regulation",
            "advisor_brief": BRIEF,
            "required_sections": BRIEF["required_sections"],
        },
    }

    def pure_fix(st, report):
        # Simulate targeted rewrite filling required sections
        st = dict(st)
        st["generated_sections"] = {
            "Opis projektu": (
                "Pełny opis projektu MŚP z DNSH, innowacją i celami programu. " * 8
            ),
            "Budżet": (
                "Szczegółowy budżet, wkład własny 30%, koszty kwalifikowalne. " * 8
            ),
        }
        st["fixed_sections"] = ["Opis projektu", "Budżet"]
        return st

    result = run_advisor_expectation_loop(
        state,
        max_iterations=3,
        min_score=70,
        apply_fixes=pure_fix,
    )
    assert result.ready is True
    assert result.stop_reason == "ready"
    assert result.regulation_grounded_pass is True