from __future__ import annotations

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

from pydantic import BaseModel, Field


DetectionEvidenceRole = Literal["false_positive", "false_negative"]
DetectionReviewDecision = Literal[
    "confirmed_model_false_positive",
    "confirmed_model_false_negative",
    "reference_gap_or_change",
    "qa_alignment_mismatch",
    "imagery_obscured_or_uncertain",
    "uncertain",
    "unreviewed",
]


class DetectionReviewUpsert(BaseModel):
    evidence_role: DetectionEvidenceRole
    evidence_feature_id: str = Field(min_length=1, max_length=255)
    decision: DetectionReviewDecision
    notes: str | None = Field(default=None, max_length=2000)
    reviewed_by: str = Field(default="operator", min_length=1, max_length=120)


class DetectionReviewRead(BaseModel):
    id: UUID | None = None
    project_id: UUID
    quality_check_id: UUID
    analysis_run_id: UUID | None = None
    evidence_role: DetectionEvidenceRole
    evidence_feature_id: str
    detection_id: UUID | None = None
    reference_feature_id: UUID | None = None
    decision: DetectionReviewDecision = "unreviewed"
    notes: str | None = None
    reviewed_by: str | None = None
    confidence: float | None = None
    class_name: str | None = None
    source_tile_path: str | None = None
    created_at: datetime | None = None
    updated_at: datetime | None = None


class DetectionReviewSummary(BaseModel):
    total: int
    reviewed: int
    remaining: int
    false_positive_total: int
    false_negative_total: int
    decision_counts: dict[str, int]
    # The score with the operator's verdicts applied, next to the raw one. A
    # finding adjudicated as a reference gap is not the model's error, and an
    # interval covers what the unreviewed remainder could still turn out to be.
    reviewed_metrics: dict | None = None


class DetectionReviewList(BaseModel):
    items: list[DetectionReviewRead]
    total: int
    limit: int
    offset: int
    summary: DetectionReviewSummary