package growth

import (
	"slices"
	"time"
)

const (
	defaultHorizonDays = 90
	minimumHistoryDays = 56
	maxTrendBPS        = 2_000
)

func BuildForecast(history []DailyPoint, catalog []Event, start time.Time, horizonDays int) Forecast {
	if horizonDays <= 0 {
		horizonDays = defaultHorizonDays
	}
	result := Forecast{StartDate: normalizeDate(start), HorizonDays: horizonDays, Confidence: ConfidenceInsufficientData, SuppressionReason: SuppressionInsufficientHistory, Points: []ForecastPoint{}}
	ordered := normalizedHistory(history, start)
	if len(ordered) == 0 {
		return result
	}
	result.Backtest = backtest(ordered, catalog)
	result.Confidence = classifyConfidence(result.Backtest)
	result.SuppressionReason = ""
	result.Points = forecastPoints(ordered, catalog, result.StartDate, horizonDays, result.Confidence)
	return result
}

func forecastPoints(history []DailyPoint, catalog []Event, start time.Time, horizon int, confidence Confidence) []ForecastPoint {
	sessionResiduals, amountResiduals := modelResiduals(history, catalog)
	points := make([]ForecastPoint, 0, horizon)
	for index := 0; index < horizon; index++ {
		day := start.AddDate(0, 0, index)
		baselineSessions, baselineAmount, predictedSessions, predictedAmount, eventIDs := predictCenter(history, catalog, day)
		points = append(points, ForecastPoint{
			Date: day, BaselineSessions: baselineSessions, BaselineAmountRial: baselineAmount,
			PaidSessions:   forecastRange(predictedSessions, sessionResiduals, confidence),
			PaidAmountRial: forecastRange(predictedAmount, amountResiduals, confidence), EventIDs: eventIDs,
		})
	}
	return points
}

func normalizedHistory(history []DailyPoint, before time.Time) []DailyPoint {
	ordered := make([]DailyPoint, 0, len(history))
	for _, point := range history {
		if dateBefore(point.Date, before) {
			point.Date = normalizeDate(point.Date)
			ordered = append(ordered, point)
		}
	}
	slices.SortFunc(ordered, func(a, b DailyPoint) int { return a.Date.Compare(b.Date) })
	return ordered
}

func weekdayMedian(history []DailyPoint, weekday time.Weekday, value func(DailyPoint) int64) int64 {
	values := make([]int64, 0, 8)
	for index := len(history) - 1; index >= 0 && len(values) < 8; index-- {
		if history[index].Date.Weekday() == weekday {
			values = append(values, value(history[index]))
		}
	}
	if len(values) == 0 {
		for index := len(history) - 1; index >= 0 && len(values) < 8; index-- {
			values = append(values, value(history[index]))
		}
	}
	return median(values)
}

func median(values []int64) int64 {
	if len(values) == 0 {
		return 0
	}
	copyOfValues := slices.Clone(values)
	slices.Sort(copyOfValues)
	middle := len(copyOfValues) / 2
	if len(copyOfValues)%2 == 1 {
		return copyOfValues[middle]
	}
	return (copyOfValues[middle-1] + copyOfValues[middle]) / 2
}

func modelResiduals(history []DailyPoint, catalog []Event) ([]int64, []int64) {
	sessions := make([]int64, 0, maxInt(0, len(history)-minimumHistoryDays))
	amounts := make([]int64, 0, maxInt(0, len(history)-minimumHistoryDays))
	for index := minimumHistoryDays; index < len(history); index++ {
		prior := history[:index]
		_, _, predictedSessions, predictedAmount, _ := predictCenter(prior, catalog, history[index].Date)
		sessions = append(sessions, history[index].PaidSessionCount-predictedSessions)
		amounts = append(amounts, history[index].PaidAmountRial-predictedAmount)
	}
	return sessions, amounts
}

func rangeFromResiduals(center int64, residualValues []int64) Range {
	return Range{
		P10: maxInt64(0, center+quantile(residualValues, 10)),
		P50: maxInt64(0, center+quantile(residualValues, 50)),
		P90: maxInt64(0, center+quantile(residualValues, 90)),
	}
}

func forecastRange(center int64, residualValues []int64, confidence Confidence) Range {
	if confidence == ConfidenceLow && len(residualValues) == 0 {
		return Range{P10: center / 2, P50: center, P90: center + center/2}
	}
	return rangeFromResiduals(center, residualValues)
}

func quantile(values []int64, percentile int) int64 {
	if len(values) == 0 {
		return 0
	}
	ordered := slices.Clone(values)
	slices.Sort(ordered)
	index := (len(ordered) - 1) * percentile / 100
	return ordered[index]
}

func trendBPS(history []DailyPoint, value func(DailyPoint) int64) int64 {
	if len(history) < 56 {
		return 0
	}
	var recent, previous int64
	for _, point := range history[len(history)-28:] {
		recent += value(point)
	}
	for _, point := range history[len(history)-56 : len(history)-28] {
		previous += value(point)
	}
	trend := changeBPS(recent, previous)
	if trend > maxTrendBPS {
		return maxTrendBPS
	}
	if trend < -maxTrendBPS {
		return -maxTrendBPS
	}
	return trend
}

func backtest(history []DailyPoint, catalog []Event) BacktestResult {
	result := BacktestResult{}
	var sessionError, amountError, covered int64
	var evaluated int64
	first := len(history) - 56
	for origin := first; origin < len(history); origin += 14 {
		if origin < minimumHistoryDays {
			continue
		}
		result.Origins++
		training := history[:origin]
		sessionResiduals, amountResiduals := modelResiduals(training, catalog)
		for index := origin; index < minInt(origin+14, len(history)); index++ {
			actual := history[index]
			_, _, sessionCenter, amountCenter, _ := predictCenter(training, catalog, actual.Date)
			sessionRange := rangeFromResiduals(sessionCenter, sessionResiduals)
			amountRange := rangeFromResiduals(amountCenter, amountResiduals)
			sessionError += absolutePercentageErrorBPS(actual.PaidSessionCount, sessionRange.P50)
			amountError += absolutePercentageErrorBPS(actual.PaidAmountRial, amountRange.P50)
			if actual.PaidSessionCount >= sessionRange.P10 && actual.PaidSessionCount <= sessionRange.P90 && actual.PaidAmountRial >= amountRange.P10 && actual.PaidAmountRial <= amountRange.P90 {
				covered++
			}
			evaluated++
		}
	}
	if evaluated > 0 {
		result.SessionMAPEBPS = sessionError / evaluated
		result.AmountMAPEBPS = amountError / evaluated
		result.IntervalCoverageBPS = covered * 10_000 / evaluated
	}
	return result
}

func predictCenter(history []DailyPoint, catalog []Event, day time.Time) (int64, int64, int64, int64, []string) {
	baselineSessions := weekdayMedian(history, day.Weekday(), func(p DailyPoint) int64 { return p.PaidSessionCount })
	baselineAmount := weekdayMedian(history, day.Weekday(), func(p DailyPoint) int64 { return p.PaidAmountRial })
	predictedSessions := applyBPS(baselineSessions, trendBPS(history, func(p DailyPoint) int64 { return p.PaidSessionCount }))
	predictedAmount := applyBPS(baselineAmount, trendBPS(history, func(p DailyPoint) int64 { return p.PaidAmountRial }))
	eventIDs := make([]string, 0, 1)
	familyEffects := eventFamilyEffects(history, catalog)
	for _, event := range catalog {
		if dateBefore(day, event.StartDate) || dateAfter(day, event.EndDate) {
			continue
		}
		eventIDs = append(eventIDs, event.ID)
		if uplift, ok := familyEffects[event.Family]; ok {
			predictedSessions = applyBPS(predictedSessions, uplift.sessions)
			predictedAmount = applyBPS(predictedAmount, uplift.amount)
		}
	}
	return baselineSessions, baselineAmount, predictedSessions, predictedAmount, eventIDs
}

func classifyConfidence(backtest BacktestResult) Confidence {
	if backtest.Origins < 4 {
		return ConfidenceLow
	}
	worstMAPE := maxInt64(backtest.SessionMAPEBPS, backtest.AmountMAPEBPS)
	if worstMAPE <= 2_000 && backtest.IntervalCoverageBPS >= 7_000 {
		return ConfidenceHigh
	}
	if worstMAPE <= 4_000 && backtest.IntervalCoverageBPS >= 5_000 {
		return ConfidenceMedium
	}
	return ConfidenceLow
}

type familyUplift struct{ sessions, amount int64 }

func eventFamilyEffects(history []DailyPoint, catalog []Event) map[string]familyUplift {
	result := make(map[string]familyUplift)
	counts := make(map[string]int64)
	for _, event := range catalog {
		if event.Family == "" || dateAfter(event.StartDate, history[len(history)-1].Date) {
			continue
		}
		effect := AnalyzeEventEffect(history, event, catalog)
		if !effect.Eligible {
			continue
		}
		value := result[event.Family]
		value.sessions += effect.PaidSessionUpliftBPS
		value.amount += effect.AmountUpliftBPS
		result[event.Family] = value
		counts[event.Family]++
	}
	for family, value := range result {
		value.sessions /= counts[family]
		value.amount /= counts[family]
		result[family] = value
	}
	return result
}

func applyBPS(value, adjustment int64) int64 { return maxInt64(0, value*(10_000+adjustment)/10_000) }

func absolutePercentageErrorBPS(actual, predicted int64) int64 {
	if actual <= 0 {
		if predicted == 0 {
			return 0
		}
		return 10_000
	}
	difference := actual - predicted
	if difference < 0 {
		difference = -difference
	}
	return difference * 10_000 / actual
}

func maxInt64(a, b int64) int64 {
	if a > b {
		return a
	}
	return b
}
func minInt(a, b int) int {
	if a < b {
		return a
	}
	return b
}

func maxInt(a, b int) int {
	if a > b {
		return a
	}
	return b
}

func DetectRisks(forecast Forecast, limit int) []Risk {
	if forecast.Confidence == ConfidenceInsufficientData || limit <= 0 {
		return nil
	}
	candidates := make([]Risk, 0, len(forecast.Points))
	for _, point := range forecast.Points {
		if point.BaselineSessions < MinBaselineSessions {
			continue
		}
		sessionDecline := -changeBPS(point.PaidSessions.P50, point.BaselineSessions)
		amountDecline := -changeBPS(point.PaidAmountRial.P50, point.BaselineAmountRial)
		if sessionDecline < MaterialChangeBPS && amountDecline < MaterialChangeBPS {
			continue
		}
		candidates = append(candidates, Risk{Date: point.Date, EndDate: point.Date, SessionDeclineBPS: maxInt64(0, sessionDecline), AmountDeclineBPS: maxInt64(0, amountDecline), ExpectedSessions: point.PaidSessions, ExpectedAmountRial: point.PaidAmountRial})
	}
	coalesced := coalesceRisks(candidates)
	slices.SortStableFunc(coalesced, func(a, b Risk) int {
		aSeverity := maxInt64(a.SessionDeclineBPS, a.AmountDeclineBPS)
		bSeverity := maxInt64(b.SessionDeclineBPS, b.AmountDeclineBPS)
		if aSeverity > bSeverity {
			return -1
		}
		if aSeverity < bSeverity {
			return 1
		}
		return a.Date.Compare(b.Date)
	})
	if len(coalesced) > limit {
		coalesced = coalesced[:limit]
	}
	return coalesced
}

func coalesceRisks(candidates []Risk) []Risk {
	if len(candidates) == 0 {
		return nil
	}
	result := make([]Risk, 0, len(candidates))
	current := candidates[0]
	for _, candidate := range candidates[1:] {
		if dayKey(candidate.Date) != dayKey(current.EndDate.AddDate(0, 0, 1)) {
			current.EvidenceID = EvidenceID("risk", dayKey(current.Date)+":"+dayKey(current.EndDate))
			result = append(result, current)
			current = candidate
			continue
		}
		current.EndDate = candidate.EndDate
		if candidate.SessionDeclineBPS > current.SessionDeclineBPS {
			current.SessionDeclineBPS = candidate.SessionDeclineBPS
			current.ExpectedSessions = candidate.ExpectedSessions
		}
		if candidate.AmountDeclineBPS > current.AmountDeclineBPS {
			current.AmountDeclineBPS = candidate.AmountDeclineBPS
			current.ExpectedAmountRial = candidate.ExpectedAmountRial
		}
	}
	current.EvidenceID = EvidenceID("risk", dayKey(current.Date)+":"+dayKey(current.EndDate))
	return append(result, current)
}