// IYmtg_App_iOS/Features/Help/Models/TrainingStatus.swift
import SwiftUI

struct TrainingCategory: Identifiable {
    let id = UUID()
    let name: String
    let group: String
    let functionalCount: Int
    let solidCount: Int
    let highAccuracyCount: Int
}

class TrainingGuideViewModel: ObservableObject {
    @Published var categories: [TrainingCategory] = [
        // MARK: Foil Model (Image Classification)
        // 13 classes — false negatives cause wrong foil label; all classes need balanced data.
        .init(name: "NonFoil",          group: "Foil Model", functionalCount: 15, solidCount: 50, highAccuracyCount: 100),
        .init(name: "Traditional",      group: "Foil Model", functionalCount: 15, solidCount: 50, highAccuracyCount: 100),
        .init(name: "Etched",           group: "Foil Model", functionalCount: 15, solidCount: 50, highAccuracyCount: 100),
        .init(name: "PreModern",        group: "Foil Model", functionalCount: 15, solidCount: 50, highAccuracyCount: 100),
        .init(name: "Textured",         group: "Foil Model", functionalCount: 15, solidCount: 50, highAccuracyCount: 100),
        .init(name: "Galaxy",           group: "Foil Model", functionalCount: 15, solidCount: 50, highAccuracyCount: 100),
        .init(name: "Surge",            group: "Foil Model", functionalCount: 15, solidCount: 50, highAccuracyCount: 100),
        .init(name: "Oil Slick",        group: "Foil Model", functionalCount: 15, solidCount: 50, highAccuracyCount: 100),
        .init(name: "Step & Compleat",  group: "Foil Model", functionalCount: 15, solidCount: 50, highAccuracyCount: 100),
        .init(name: "Halo",             group: "Foil Model", functionalCount: 15, solidCount: 50, highAccuracyCount: 100),
        .init(name: "Confetti",         group: "Foil Model", functionalCount: 15, solidCount: 50, highAccuracyCount: 100),
        .init(name: "Neon Ink",         group: "Foil Model", functionalCount: 15, solidCount: 50, highAccuracyCount: 100),
        .init(name: "Fracture",         group: "Foil Model", functionalCount: 15, solidCount: 50, highAccuracyCount: 100),

        // MARK: Stamp Model (Image Classification)
        // Detects promo stamp presence. Binary classification — easy to train quickly.
        .init(name: "Stamped (Promo)",  group: "Stamp Model", functionalCount: 15, solidCount: 50, highAccuracyCount: 100),
        .init(name: "Clean (No Stamp)", group: "Stamp Model", functionalCount: 15, solidCount: 50, highAccuracyCount: 100),

        // MARK: Condition Model (Object Detection — requires bounding-box annotations in Create ML)
        // Higher minimums than image classification because each image needs manual annotation.
        // Surface defects
        .init(name: "Light Scratches",  group: "Condition Model", functionalCount: 20, solidCount: 50, highAccuracyCount: 100),
        .init(name: "Clouding",         group: "Condition Model", functionalCount: 20, solidCount: 50, highAccuracyCount: 100),
        .init(name: "Dirt",             group: "Condition Model", functionalCount: 20, solidCount: 50, highAccuracyCount: 100),
        .init(name: "Dents",            group: "Condition Model", functionalCount: 20, solidCount: 50, highAccuracyCount: 100),
        // Edge defects
        .init(name: "Whitening (Edges)", group: "Condition Model", functionalCount: 20, solidCount: 50, highAccuracyCount: 100),
        .init(name: "Chipping (Edges)", group: "Condition Model", functionalCount: 20, solidCount: 50, highAccuracyCount: 100),
        .init(name: "Corner Wear",      group: "Condition Model", functionalCount: 20, solidCount: 50, highAccuracyCount: 100),
        // Structure defects
        .init(name: "Creases",          group: "Condition Model", functionalCount: 20, solidCount: 50, highAccuracyCount: 100),
        .init(name: "Shuffle Bend",     group: "Condition Model", functionalCount: 20, solidCount: 50, highAccuracyCount: 100),
        .init(name: "Binder Dents",     group: "Condition Model", functionalCount: 20, solidCount: 50, highAccuracyCount: 100),
        // Critical damage — a single false positive will downgrade a NM card to Damaged.
        // Train these more aggressively to minimise false positives.
        .init(name: "Water Damage",     group: "Condition Model", functionalCount: 20, solidCount: 75, highAccuracyCount: 150),
        .init(name: "Inking",           group: "Condition Model", functionalCount: 20, solidCount: 75, highAccuracyCount: 150),
        .init(name: "Rips",             group: "Condition Model", functionalCount: 20, solidCount: 75, highAccuracyCount: 150),
    ]

    var groups: [String] {
        var seen = Set<String>()
        return categories.compactMap { seen.insert($0.group).inserted ? $0.group : nil }
    }

    func categories(for group: String) -> [TrainingCategory] {
        categories.filter { $0.group == group }
    }
}