o2o-triple-pick/backend/app/routers/leaderboard.py

"""리더보드 API — 누적 포인트 1위 (docs/SCORING.md §3 랭킹 규칙).

user_points 테이블(채점 시 누적 갱신) 기준 랭킹.
1차 정렬: 누적 포인트. 동점: 정확스코어 횟수 → 적중률 → 참여수 → 최초도달.
주최측/임직원 배제(P1). 이메일은 마스킹하여 노출.
"""
from __future__ import annotations

from fastapi import APIRouter, Depends, Query
from sqlalchemy import func, select
from sqlalchemy.ext.asyncio import AsyncSession
from sqlalchemy.orm import selectinload

from ..database import get_db
from ..models import Match, UserPoints, UserPrediction
from ..scoring import is_excluded, score_prediction
from ..schemas import (
    AILeaderboardOut,
    AIStandingOut,
    LeaderboardOut,
    StandingOut,
)

router = APIRouter(prefix="/api/leaderboard", tags=["leaderboard"])


def _mask(email: str) -> str:
    name, _, domain = email.partition("@")
    head = name[:2] if len(name) >= 2 else name
    return f"{head}{'*' * max(1, len(name) - 2)}@{domain}"


@router.get("", response_model=LeaderboardOut)
async def leaderboard(
    limit: int = Query(50, ge=1, le=200),
    league: str = Query("", description="wc | kbo | mlb — 빈값이면 전체 합산"),
    db: AsyncSession = Depends(get_db),
) -> LeaderboardOut:
    if league:
        # 리그별 랭킹 — 채점된 픽(points)을 리그 경기로 한정해 이메일별 재집계.
        picks = (
            await db.execute(
                select(UserPrediction, Match)
                .join(Match, Match.match_id == UserPrediction.match_id)
                .where(
                    Match.league == league,
                    Match.result_outcome.is_not(None),
                    UserPrediction.points.is_not(None),
                    UserPrediction.email.is_not(None),
                )
            )
        ).all()
        agg: dict[str, list] = {}  # email → [pts, exact, played, first_ts]
        for pk, m in picks:
            row = agg.setdefault(pk.email, [0, 0, 0, None])
            row[0] += pk.points or 0
            row[1] += 1 if (
                pk.score_a == m.result_score_a and pk.score_b == m.result_score_b
            ) else 0
            row[2] += 1
        ranked_rows = sorted(
            ((e, v) for e, v in agg.items() if not is_excluded(e)),
            key=lambda kv: (-kv[1][0], -kv[1][1], -kv[1][2]),
        )
        scored_matches = (
            await db.execute(
                select(func.count()).select_from(Match).where(
                    Match.league == league, Match.result_outcome.is_not(None)
                )
            )
        ).scalar() or 0
        standings = [
            StandingOut(
                rank=i + 1, emailMasked=_mask(e),
                totalPoints=v[0], exactCount=v[1], matchesPlayed=v[2],
            )
            for i, (e, v) in enumerate(ranked_rows[:limit])
        ]
        return LeaderboardOut(standings=standings, scoredMatches=scored_matches)

    rows = (
        await db.execute(select(UserPoints))
    ).scalars().all()

    ranked = sorted(
        (r for r in rows if not is_excluded(r.email)),
        key=lambda r: (
            -r.total_points,
            -r.exact_count,
            -(r.total_points / r.matches_played if r.matches_played else 0),
            -r.matches_played,
            r.first_scored_at.timestamp() if r.first_scored_at else 0,
        ),
    )

    scored_matches = (
        await db.execute(
            select(func.count())
            .select_from(Match)
            .where(Match.result_outcome.is_not(None))
        )
    ).scalar() or 0

    standings = [
        StandingOut(
            rank=i + 1,
            emailMasked=_mask(r.email),
            totalPoints=r.total_points,
            exactCount=r.exact_count,
            matchesPlayed=r.matches_played,
        )
        for i, r in enumerate(ranked[:limit])
    ]
    return LeaderboardOut(standings=standings, scoredMatches=scored_matches)


@router.get("/ai", response_model=AILeaderboardOut)
async def ai_leaderboard(
    league: str = Query("", description="wc | kbo | mlb — 빈값이면 전체"),
    db: AsyncSession = Depends(get_db),
) -> AILeaderboardOut:
    """AI 모델 누적 랭킹 — 종료된 경기의 AI 예측을 유저와 동일한 배점으로 채점·합산.

    별도 누적 테이블 없이 매 조회 시 종료 경기 전수 재계산 → 결과/배점 변동에
    항상 일관. (모델 3개 × 경기 수라 비용 무시 가능.)
    """
    q = (
        select(Match)
        .where(Match.result_outcome.is_not(None))
        .options(selectinload(Match.predictions))
    )
    if league:
        q = q.where(Match.league == league)
    matches = (await db.execute(q)).scalars().all()

    # model → [points, exact_count, matches_played]
    agg: dict[str, list[int]] = {}
    for m in matches:
        bb = m.league in ("kbo", "mlb")
        for p in m.predictions:
            pts = score_prediction(
                p.score_a, p.score_b, m.result_score_a, m.result_score_b, baseball=bb
            )
            row = agg.setdefault(p.model, [0, 0, 0])
            row[0] += pts
            row[1] += 1 if (p.score_a == m.result_score_a and p.score_b == m.result_score_b) else 0
            row[2] += 1

    order = {"GPT": 0, "Claude": 1, "Gemini": 2}
    ranked = sorted(
        agg.items(),
        key=lambda kv: (-kv[1][0], -kv[1][1], -kv[1][2], order.get(kv[0], 9)),
    )

    standings = [
        AIStandingOut(
            rank=i + 1,
            model=model,  # type: ignore[arg-type]
            totalPoints=pts,
            exactCount=exact,
            matchesPlayed=played,
        )
        for i, (model, (pts, exact, played)) in enumerate(ranked)
    ]
    return AILeaderboardOut(standings=standings, scoredMatches=len(matches))