"""Pydantic models for the eval system.

Used both as the contract for what each expert LLM must return (JSON validated
against ExpertResult) and as the storage shape for evaluation rows.
"""

from __future__ import annotations

from datetime import datetime
from typing import Literal
from uuid import UUID

from pydantic import BaseModel, Field, field_validator


# Axes the judge must score. Order does not matter for storage but matches
# rubric_v1.yaml for readability when humans inspect output.
AXIS_NAMES: tuple[str, ...] = (
    "task_complexity",
    "goal_completion",
    "tool_usage_quality",
    "efficiency",
    "communication",
    "subagent_orchestration",
    "self_extension",
)

# Axes that may be null when the session never used the relevant mechanic.
NULLABLE_AXES: frozenset[str] = frozenset({"subagent_orchestration", "self_extension"})


class EvalScores(BaseModel):
    """Per-axis numeric scores. Fully open-ended scale: any non-negative integer."""

    task_complexity: int = Field(ge=0)
    goal_completion: int = Field(ge=0)
    tool_usage_quality: int = Field(ge=0)
    efficiency: int = Field(ge=0)
    communication: int = Field(ge=0)
    subagent_orchestration: int | None = Field(default=None, ge=0)
    self_extension: int | None = Field(default=None, ge=0)


class ExpertResult(BaseModel):
    """One expert's verdict for one session. Matches the JSON the LLM emits."""

    expert_id: Literal["strict_critic", "pragmatist", "tech_lead"]
    scores: EvalScores
    comment: str

    @field_validator("comment")
    @classmethod
    def _strip_comment(cls, v: str) -> str:
        v = (v or "").strip()
        if not v:
            raise ValueError("comment must not be empty")
        return v


class EvalRunMetadata(BaseModel):
    """Metadata that pins the rubric / judge version for a single run."""

    eval_run_id: UUID
    eval_date: datetime
    judge_model: str
    judge_version: str
    rubric_version: str


class StoredEvaluation(BaseModel):
    """Row shape returned by db.list_evaluations()."""

    id: UUID
    session_id: str
    eval_run_id: UUID
    eval_date: datetime
    judge_model: str
    judge_version: str
    rubric_version: str
    expert_id: str
    scores: EvalScores
    comment: str


# ── REST response shapes ─────────────────────────────────────────────────


class SessionOverview(BaseModel):
    """Row in GET /eval/sessions."""

    session_id: str
    profile_id: str
    name: str | None = None
    created_at: datetime
    last_active: datetime
    pinned: bool
    msg_count: int
    likes: int
    dislikes: int
    eval_status: Literal["evaluated", "pending", "stale"]
    latest_avg: dict | None = None
    latest_eval_date: datetime | None = None
    latest_judge_version: str | None = None
    latest_rubric_version: str | None = None


class SessionDetail(BaseModel):
    """GET /eval/sessions/{id}."""

    session_id: str
    profile_id: str
    name: str | None = None
    created_at: datetime
    last_active: datetime
    msg_count: int
    feedback: list[dict]                # [{message_index, rating, ...}]
    evaluations: list[StoredEvaluation] # newest first


class WeeklyAxisMeans(BaseModel):
    week_start: str
    bucket: str
    sample_count: int
    axis_means: dict


class StatsResponse(BaseModel):
    buckets: list[str]
    weekly: list[WeeklyAxisMeans]


# ── Run trigger / status ─────────────────────────────────────────────────


class RunRequest(BaseModel):
    """POST /eval/run body."""

    scope: Literal["unevaluated", "session", "all"] = "unevaluated"
    session_id: str | None = None
    since: datetime | None = None
    limit: int | None = None
    model: str = "gemma4:31b-cloud"
    backend: str = "ollama"


class RunSessionStatus(BaseModel):
    session_id: str
    state: Literal["pending", "running", "ok", "failed"]
    avg: dict | None = None
    error: str | None = None


class RunStatus(BaseModel):
    """GET /eval/run/{run_id}."""

    run_id: str
    state: Literal["running", "done", "failed"]
    started_at: datetime
    finished_at: datetime | None = None
    judge_model: str
    judge_version: str
    rubric_version: str
    sessions: list[RunSessionStatus]