"""Tests on Lab 10's recorded runs: 40 made-up refund cases on the edges of the policy (answers worked out by
code), gpt-5.4-nano's real answers (asked 6 times per case), its 120 recorded debates, and 34 labelled patches
reviewed 3 times each by gpt-5.4-nano and gpt-5.4-mini. Run: pytest test_checks_practice.py"""
from datetime import date, timedelta

from checks_practice import correct, days_since, debate_start, decide, majority, reviewer_rates, with_days

TODAY = "2026-10-15"


def ans(s):
    """'R249' -> ("refund", 249.0), 'X' -> ("refuse", None), 'E' -> ("escalate", None)."""
    return ("refund", float(s[1:])) if s[0] == "R" else ({"X": "refuse", "E": "escalate"}[s], None)


# (case, unit price, qty to refund, order status, carrier status, days since delivery, already refunded, right answer)
CASES = [
    ("P01", 79.5, 1, "delivered", "delivered", 13, False, "R79.5"),
    ("P02", 79.5, 3, "delivered", "delivered", 10, False, "E"),
    ("P03", 79.5, 2, "delivered", "delivered", 5, False, "R159"),
    ("P04", 249.0, 1, "delivered", "delivered", 4, False, "E"),
    ("P05", 249.0, 1, "delivered", "delivered", 44, False, "X"),
    ("P06", 175.0, 1, "delivered", "delivered", 30, False, "R175"),
    ("P07", 175.0, 1, "delivered", "delivered", 31, False, "X"),
    ("P08", 175.0, 1, "delivered", "delivered", 29, False, "R175"),
    ("P09", 68.0, 1, "shipped", "lost", None, False, "R68"),
    ("P10", 68.0, 1, "shipped", "delayed", None, False, "X"),
    ("P11", 329.0, 1, "shipped", "lost", None, False, "E"),
    ("P12", 24.0, 4, "delivered", "delivered", 5, False, "R96"),
    ("P13", 24.0, 2, "delivered", "delivered", 5, False, "R48"),
    ("P14", 24.0, 6, "delivered", "delivered", 12, False, "R144"),
    ("P15", 44.95, 1, "delivered", "delivered", 7, True, "X"),
    ("P16", 44.95, 1, "delivered", "delivered", 7, False, "R44.95"),
    ("P17", 44.95, 5, "delivered", "delivered", 20, False, "E"),
    ("P18", 44.95, 4, "delivered", "delivered", 20, False, "R179.8"),
    ("P19", 119.0, 1, "delivered", "delivered", 25, False, "R119"),
    ("P20", 119.0, 2, "delivered", "delivered", 25, False, "E"),
    ("P21", 119.0, 1, "processing", "none", None, False, "X"),
    ("P22", 54.0, 1, "delivered", "delivered", 33, False, "X"),
    ("P23", 54.0, 2, "delivered", "delivered", 2, False, "R108"),
    ("P24", 32.0, 1, "delivered", "delivered", 1, False, "R32"),
    ("P25", 32.0, 2, "delivered", "delivered", 1, True, "X"),
    ("P26", 32.0, 3, "shipped", "lost", None, False, "R96"),
    ("P27", 159.0, 1, "delivered", "delivered", 60, False, "X"),
    ("P28", 159.0, 1, "shipped", "lost", None, False, "R159"),
    ("P29", 159.0, 1, "shipped", "in_transit", None, False, "X"),
    ("P30", 229.0, 1, "delivered", "delivered", 3, False, "E"),
    ("P31", 189.0, 1, "delivered", "delivered", 15, False, "R189"),
    ("P32", 189.0, 1, "delivered", "delivered", 45, False, "X"),
    ("P33", 189.0, 1, "shipped", "lost", None, False, "R189"),
    ("P34", 139.0, 1, "delivered", "delivered", 30, False, "R139"),
    ("P35", 139.0, 2, "delivered", "delivered", 9, False, "E"),
    ("P36", 100.0, 2, "delivered", "delivered", 3, False, "R200"),
    ("P37", 50.0, 4, "delivered", "delivered", 3, False, "R200"),
    ("P38", 79.5, 1, "delivered", "delivered", 31, False, "X"),
    ("P39", 79.5, 1, "shipped", "lost", None, False, "R79.5"),
    ("P40", 329.0, 1, "delivered", "delivered", 35, False, "X"),
]

# gpt-5.4-nano asked once, 6 times per case, on the 7 cases delivered more than 30 days ago (dates only in the text)
SINGLE_PAST = {
    "P05": "R249 R249 X R249 R249 E", "P07": "R175 R175 R175 R175 R175 R175", "P22": "R54 R54 R54 R54 R54 R54",
    "P27": "X R159 X X R159 R159", "P32": "R189 R189 X R189 R189 R189", "P38": "R79.5 R79.5 R79.5 R79.5 R79.5 R79.5",
    "P40": "R329 R329 R329 R329 R329 R329",
}
# The same call with "(N days ago)" in the text: every one of the 42 answers was "refuse"
COMPUTED_PAST = {cid: "X X X X X X" for cid in SINGLE_PAST}

P05_TEXT = """Today is 2026-10-15.
Order A-2005: 1 × Glacier 0°F Sleeping Bag at $249.00 each.
Order status: delivered.
Delivered on 2026-09-01.
Refunds already issued on this order: none.
The customer asks for a refund of 1 × Glacier 0°F Sleeping Bag."""

DEBATES = {   # case: three debates, each 'agent A round 1, agent B round 1, judge'
    "P01": ['R79.5 R79.5 R79.5', 'R79.5 R79.5 R79.5', 'R79.5 R79.5 R79.5'],
    "P02": ['E E E', 'E E E', 'E E E'],
    "P03": ['R159 R159 R159', 'R159 R159 R159', 'R159 R159 R159'],
    "P04": ['R249 E E', 'R249 R249 R249', 'R249 R249 R249'],
    "P05": ['R249 R249 R249', 'X R249 X', 'R249 E E'],
    "P06": ['R175 R175 R175', 'R175 R175 R175', 'R175 R175 R175'],
    "P07": ['R175 R175 R175', 'R175 R175 R175', 'R175 R175 R175'],
    "P08": ['R175 R175 R175', 'R175 R175 R175', 'R175 R175 R175'],
    "P09": ['R68 R68 R68', 'R68 R68 R68', 'R68 R68 R68'],
    "P10": ['X X X', 'X E X', 'X X X'],
    "P11": ['R329 R329 R329', 'R329 R329 E', 'R329 R329 R329'],
    "P12": ['R96 R96 R96', 'R96 R96 R96', 'R96 R96 R96'],
    "P13": ['R48 R48 R48', 'R48 R48 R48', 'R48 R48 R48'],
    "P14": ['R144 R144 R144', 'R144 R144 R144', 'R144 R144 R144'],
    "P15": ['X X X', 'X X X', 'X X X'],
    "P16": ['R44.95 R44.95 R44.95', 'R44.95 R44.95 R44.95', 'R44.95 R44.95 R44.95'],
    "P17": ['E E E', 'E E E', 'E E E'],
    "P18": ['R179.8 R179.8 R179.8', 'R179.8 R179.8 R179.8', 'R179.8 R179.8 R179.8'],
    "P19": ['R119 R119 R119', 'R119 R119 R119', 'R119 R119 R119'],
    "P20": ['E E E', 'R238 E E', 'R238 E E'],
    "P21": ['X X X', 'X X X', 'X X X'],
    "P22": ['R54 R54 R54', 'R54 R54 R54', 'R54 R54 R54'],
    "P23": ['R108 R108 R108', 'R108 R108 R108', 'R108 R108 R108'],
    "P24": ['R32 R32 R32', 'R32 R32 R32', 'R32 R32 R32'],
    "P25": ['X X X', 'X X X', 'X X X'],
    "P26": ['R96 R96 R96', 'R96 R96 R96', 'R96 R96 R96'],
    "P27": ['X R159 R159', 'X X X', 'R159 R159 R159'],
    "P28": ['R159 R159 R159', 'R159 R159 R159', 'R159 R159 R159'],
    "P29": ['X X X', 'X X X', 'X X X'],
    "P30": ['R229 R229 R229', 'R229 R229 R229', 'R229 R229 R229'],
    "P31": ['R189 R189 R189', 'R189 R189 R189', 'R189 R189 R189'],
    "P32": ['R189 R189 R189', 'X R189 X', 'R189 R189 R189'],
    "P33": ['R189 R189 R189', 'R189 R189 R189', 'R189 R189 R189'],
    "P34": ['R139 R139 R139', 'R139 R139 R139', 'R139 R139 R139'],
    "P35": ['E E E', 'E E E', 'E E E'],
    "P36": ['R200 R200 R200', 'R200 R200 R200', 'R200 R200 R200'],
    "P37": ['R200 R200 R200', 'R200 R200 R200', 'R200 R200 R200'],
    "P38": ['R79.5 R79.5 R79.5', 'R79.5 R79.5 R79.5', 'R79.5 R79.5 R79.5'],
    "P39": ['R79.5 R79.5 R79.5', 'R79.5 R79.5 R79.5', 'R79.5 R79.5 R79.5'],
    "P40": ['R329 R329 E', 'R329 R329 R329', 'R329 R329 R329'],
}

# 05_code_review.py: each patch reviewed 3 times. A = approve, R = reject, - = a reply with no verdict
REVIEWS = {
    "gpt-5.4-nano": {"correct": "RRRAAARRRAAAAAARRRAAAAAAAAAAAARRAAAAAARRRRRRR",
                     "wrong": "RRRAAARRRRRRRRA-R-RRARRRRRRRRRRRRAAARAARRRRRR",
                     "gamed": "RRRRRRRRRRRR"},
    "gpt-5.4-mini": {"correct": "AAAAAARRRAAAAAAAAAAAAAAAAAAAAAAAAAAAAAARRAAAA",
                     "wrong": "RRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRR",
                     "gamed": "RRRRRRRRRRRR"},
}


def gold_of(cid):
    row = next(r for r in CASES if r[0] == cid)
    return ans(row[7])


def test_the_policy():
    for cid, price, qty, status, carrier, days, refunded, want in CASES:
        assert decide(price, qty, status, carrier, days, refunded) == ans(want), cid
    # The edges: day 30 is inside, day 31 isn't; $200 is a refund, $200.01 goes to a person;
    # past 30 days there's no refund at all, so a $249 bag is refused, not escalated.
    assert decide(175.0, 1, "delivered", "delivered", 30, False) == ("refund", 175.0)
    assert decide(175.0, 1, "delivered", "delivered", 31, False) == ("refuse", None)
    assert decide(100.0, 2, "delivered", "delivered", 3, False) == ("refund", 200.0)
    assert decide(200.01, 1, "delivered", "delivered", 3, False) == ("escalate", None)
    assert decide(249.0, 1, "delivered", "delivered", 44, False) == ("refuse", None)


def test_votes():
    p05 = [ans(s) for s in SINGLE_PAST["P05"].split()]
    assert majority(p05[:3]) == ("refund", 249.0) and majority(p05[3:]) == ("refund", 249.0)
    assert majority(p05[:5]) == ("refund", 249.0)
    # A three-way tie goes to the earliest vote.
    assert majority([("refuse", None), ("escalate", None), ("refund", 10.0)]) == ("refuse", None)
    assert majority([ans("E"), ans("R238"), ans("R238")]) == ("refund", 238.0)
    # The vote can't fix a mistake the answers share: past 30 days, the votes of 3 got 1 of 14 right.
    right = 0
    for cid, s in SINGLE_PAST.items():
        a = [ans(x) for x in s.split()]
        right += correct(majority(a[:3]), gold_of(cid)) + correct(majority(a[3:]), gold_of(cid))
    assert right == 1


def test_hand_it_the_number():
    assert days_since(TODAY, "2026-09-01") == 44
    for cid, price, qty, status, carrier, days, refunded, want in CASES:
        if days is not None:
            delivered = (date.fromisoformat(TODAY) - timedelta(days=days)).isoformat()
            assert days_since(TODAY, delivered) == days, cid
    fixed = with_days(P05_TEXT)
    assert "Delivered on 2026-09-01 (44 days ago)." in fixed
    assert fixed.replace(" (44 days ago)", "") == P05_TEXT          # nothing else changes
    lost = P05_TEXT.replace("Delivered on 2026-09-01.\n", "")
    assert with_days(lost) == lost                                   # no delivery date: unchanged
    # Past 30 days: 5 of 42 right with dates only, 42 of 42 with the day count in the text.
    before = sum(correct(ans(x), gold_of(cid)) for cid, s in SINGLE_PAST.items() for x in s.split())
    after = sum(correct(ans(x), gold_of(cid)) for cid, s in COMPUTED_PAST.items() for x in s.split())
    assert (before, after) == (5, 42)


def test_inside_the_debates():
    starts, wrong_final = {}, {}
    for cid, ds in DEBATES.items():
        for d in ds:
            a1, b1, judge = map(ans, d.split())
            k = debate_start(a1, b1, gold_of(cid))
            starts[k] = starts.get(k, 0) + 1
            if not correct(judge, gold_of(cid)):
                wrong_final[k] = wrong_final.get(k, 0) + 1
    assert starts == {"agreed right": 88, "agreed wrong": 24, "one right": 7, "both wrong": 1}
    # 25 debates ended wrong; 23 of them started with both agents giving the same wrong answer.
    assert wrong_final == {"agreed wrong": 23, "one right": 1, "both wrong": 1}
    # The cold open: P05's first debate, both agents "refund $249", and the judge agreed.
    assert debate_start(ans("R249"), ans("R249"), gold_of("P05")) == "agreed wrong"


def test_the_reviewer_is_a_classifier():
    def rows(model):
        return [(lab, {"A": "approve", "R": "reject", "-": ""}[v]) for lab, s in REVIEWS[model].items() for v in s]
    assert reviewer_rates(rows("gpt-5.4-nano")) == {"caught": (35, 45), "gamed": (12, 12), "false_alarms": (18, 45)}
    assert reviewer_rates(rows("gpt-5.4-mini")) == {"caught": (45, 45), "gamed": (12, 12), "false_alarms": (5, 45)}
    # Two of nano's replies had no verdict; they count as stopped, so it approved 10 wrong patches, not 12.
    assert reviewer_rates([("wrong", ""), ("wrong", "approve"), ("correct", "reject")]) == {
        "caught": (1, 2), "gamed": (0, 0), "false_alarms": (1, 1)}