Capstone - Complete Analytics System

library(tidyverse)

# Simulate comprehensive match event data
set.seed(42)

# Generate team and league data
teams <- c("Manchester City", "Liverpool", "Arsenal", "Chelsea",
           "Man United", "Tottenham", "Newcastle", "Brighton")

# Function to simulate match events
simulate_match_events <- function(team_name, n_matches = 10) {
  map_dfr(1:n_matches, function(m) {
    n_events <- sample(400:600, 1)

    tibble(
      match_id = m,
      team = team_name,
      minute = sort(sample(1:95, n_events, replace = TRUE)),
      event_type = sample(c("Pass", "Carry", "Shot", "Dribble", "Pressure", "Duel",
                            "Clearance", "Interception", "Cross", "Foul"),
                          n_events, replace = TRUE,
                          prob = c(0.55, 0.12, 0.03, 0.05, 0.08, 0.07, 0.04, 0.03, 0.02, 0.01)),
      location_x = runif(n_events, 0, 120),
      location_y = runif(n_events, 0, 80),
      end_location_x = runif(n_events, 0, 120),
      end_location_y = runif(n_events, 0, 80),
      pass_length = runif(n_events, 2, 50),
      pass_outcome = sample(c("Complete", "Incomplete"), n_events, replace = TRUE, prob = c(0.8, 0.2)),
      shot_xg = ifelse(event_type == "Shot", runif(n_events, 0.02, 0.6), NA),
      shot_outcome = ifelse(event_type == "Shot",
                           sample(c("Goal", "Saved", "Off Target", "Blocked"),
                                  n_events, replace = TRUE, prob = c(0.1, 0.4, 0.3, 0.2)), NA),
      dribble_outcome = ifelse(event_type == "Dribble",
                               sample(c("Complete", "Incomplete"), n_events, replace = TRUE, prob = c(0.55, 0.45)), NA),
      player_name = sample(paste0("Player_", 1:14), n_events, replace = TRUE)
    )
  })
}

# Generate data for all teams
all_team_data <- map_dfr(teams, ~simulate_match_events(.x))

# Comprehensive style metrics calculation
calculate_style_metrics <- function(team_events) {
  passes <- team_events %>% filter(event_type == "Pass")
  shots <- team_events %>% filter(event_type == "Shot")
  pressures <- team_events %>% filter(event_type == "Pressure")
  crosses <- team_events %>% filter(event_type == "Cross")
  dribbles <- team_events %>% filter(event_type == "Dribble")

  n_matches <- n_distinct(team_events$match_id)

  tibble(
    # Possession & Build-up
    passes_per_90 = nrow(passes) / n_matches,
    pass_completion = mean(passes$pass_outcome == "Complete") * 100,
    avg_pass_length = mean(passes$pass_length),
    long_ball_pct = mean(passes$pass_length > 30) * 100,
    progressive_pass_pct = mean(passes$end_location_x - passes$location_x > 10) * 100,
    backward_pass_pct = mean(passes$end_location_x - passes$location_x < -5) * 100,

    # Possession zones
    possession_own_third = mean(team_events$location_x < 40) * 100,
    possession_mid_third = mean(team_events$location_x >= 40 & team_events$location_x < 80) * 100,
    possession_final_third = mean(team_events$location_x >= 80) * 100,

    # Width
    avg_width = sd(team_events$location_y),
    left_side_pct = mean(team_events$location_y < 30) * 100,
    right_side_pct = mean(team_events$location_y > 50) * 100,

    # Pressing
    pressures_per_90 = nrow(pressures) / n_matches,
    high_press_pct = mean(pressures$location_x > 80) * 100,
    mid_press_pct = mean(pressures$location_x >= 40 & pressures$location_x <= 80) * 100,

    # Attacking
    shots_per_90 = nrow(shots) / n_matches,
    xg_per_shot = mean(shots$shot_xg, na.rm = TRUE),
    xg_per_90 = sum(shots$shot_xg, na.rm = TRUE) / n_matches,
    shot_accuracy = mean(shots$shot_outcome %in% c("Goal", "Saved"), na.rm = TRUE) * 100,

    # Crosses & Dribbles
    crosses_per_90 = nrow(crosses) / n_matches,
    dribbles_per_90 = nrow(dribbles) / n_matches,
    dribble_success = mean(dribbles$dribble_outcome == "Complete", na.rm = TRUE) * 100,

    # Directness
    directness_index = (sum(shots$shot_xg, na.rm = TRUE) / nrow(passes)) * 1000
  )
}

# Calculate metrics for all teams
league_metrics <- map_dfr(teams, function(t) {
  team_events <- all_team_data %>% filter(team == t)
  metrics <- calculate_style_metrics(team_events)
  metrics$team <- t
  metrics
})

# League average
league_avg <- league_metrics %>%
  summarise(across(where(is.numeric), mean, na.rm = TRUE)) %>%
  mutate(team = "League Average")

# Opposition report generator
generate_opposition_report <- function(opponent_name, team_data, league_metrics, league_avg) {

  opponent_events <- team_data %>% filter(team == opponent_name)
  opponent_style <- league_metrics %>% filter(team == opponent_name)

  cat("\n")
  cat("=" , rep("=", 70), "\n", sep = "")
  cat("         OPPOSITION ANALYSIS REPORT: ", opponent_name, "\n", sep = "")
  cat("=" , rep("=", 70), "\n\n", sep = "")

  # 1. Style Profile
  cat("1. PLAYING STYLE PROFILE\n")
  cat(rep("-", 50), "\n", sep = "")

  style_label <- case_when(
    opponent_style$possession_final_third > league_avg$possession_final_third &
      opponent_style$avg_pass_length < league_avg$avg_pass_length ~ "Possession-Based",
    opponent_style$long_ball_pct > league_avg$long_ball_pct + 5 ~ "Direct",
    opponent_style$high_press_pct > league_avg$high_press_pct + 10 ~ "High-Pressing",
    opponent_style$crosses_per_90 > league_avg$crosses_per_90 + 2 ~ "Wing-Focused",
    TRUE ~ "Balanced"
  )

  cat("Style Classification:", style_label, "\n\n")

  # Key metrics vs league
  key_metrics <- c("pass_completion", "progressive_pass_pct", "pressures_per_90",
                   "high_press_pct", "shots_per_90", "xg_per_shot", "crosses_per_90")

  cat("Key Metrics vs League Average:\n")
  for (metric in key_metrics) {
    opp_val <- opponent_style[[metric]]
    league_val <- league_avg[[metric]]
    diff <- opp_val - league_val
    direction <- ifelse(diff > 0, "ABOVE", "BELOW")

    cat(sprintf("  %-25s: %.1f (League: %.1f) - %s\n",
                gsub("_", " ", metric), opp_val, league_val, direction))
  }

  # 2. Weakness Analysis
  cat("\n2. EXPLOITABLE WEAKNESSES\n")
  cat(rep("-", 50), "\n", sep = "")

  weaknesses <- list()

  # Check for weaknesses
  if (opponent_style$high_press_pct < league_avg$high_press_pct - 5) {
    weaknesses$press <- "LOW PRESS - Space available in midfield. Build from back confidently."
  }

  if (opponent_style$pass_completion < league_avg$pass_completion - 3) {
    weaknesses$turnovers <- "TURNOVER PRONE - Press high to force errors and create transitions."
  }

  if (opponent_style$left_side_pct < opponent_style$right_side_pct - 10) {
    weaknesses$left_side <- "WEAK LEFT SIDE - Attack down their left; overload with RW + RB."
  }

  if (opponent_style$possession_own_third > league_avg$possession_own_third + 5) {
    weaknesses$deep_play <- "DEEP BUILD-UP - High press can force long balls and turnovers."
  }

  if (opponent_style$dribble_success < league_avg$dribble_success - 5) {
    weaknesses$dribble_def <- "POOR DRIBBLE DEFENSE - 1v1 situations favourable; encourage attackers to take on players."
  }

  if (length(weaknesses) == 0) {
    cat("  No major weaknesses identified.\n")
  } else {
    for (i in seq_along(weaknesses)) {
      cat(sprintf("  %d. %s\n", i, weaknesses[[i]]))
    }
  }

  # 3. Key Player Analysis
  cat("\n3. KEY PLAYER THREATS\n")
  cat(rep("-", 50), "\n", sep = "")

  player_threats <- opponent_events %>%
    group_by(player_name) %>%
    summarise(
      events = n(),
      shots = sum(event_type == "Shot"),
      xg = sum(shot_xg, na.rm = TRUE),
      goals = sum(shot_outcome == "Goal", na.rm = TRUE),
      key_passes = sum(event_type == "Pass" & end_location_x > 100, na.rm = TRUE),
      progressive_carries = sum(event_type == "Carry" & (end_location_x - location_x) > 15),
      pressures = sum(event_type == "Pressure"),
      .groups = "drop"
    ) %>%
    mutate(
      threat_score = xg * 5 + key_passes * 0.5 + progressive_carries * 0.3
    ) %>%
    arrange(desc(threat_score)) %>%
    head(5)

  for (i in 1:nrow(player_threats)) {
    p <- player_threats[i, ]
    cat(sprintf("  %d. %s - Threat Score: %.1f\n", i, p$player_name, p$threat_score))
    cat(sprintf("     xG: %.2f | Shots: %d | Key Passes: %d\n\n", p$xg, p$shots, p$key_passes))
  }

  # 4. Tactical Recommendations
  cat("4. TACTICAL RECOMMENDATIONS\n")
  cat(rep("-", 50), "\n", sep = "")

  cat("\n  IN POSSESSION:\n")
  if (opponent_style$high_press_pct > league_avg$high_press_pct) {
    cat("    - Play through their press with quick combinations\n")
    cat("    - Use wide areas to escape pressure\n")
    cat("    - Goalkeeper comfortable playing with feet essential\n")
  } else {
    cat("    - Build patiently; they don't press high\n")
    cat("    - Look for progressive passes into midfield pockets\n")
    cat("    - Switch play to exploit width\n")
  }

  cat("\n  OUT OF POSSESSION:\n")
  if (opponent_style$long_ball_pct > league_avg$long_ball_pct) {
    cat("    - Win second balls; strong aerial presence needed\n")
    cat("    - Compress midfield to limit space for knockdowns\n")
    cat("    - Fullbacks stay tight to wingers\n")
  } else {
    cat("    - Press in middle third to disrupt build-up\n")
    cat("    - Man-mark their key playmaker\n")
    cat("    - Force play into wide areas\n")
  }

  cat("\n  SET PIECES:\n")
  cat("    - Assign dedicated markers for aerial threats\n")
  cat("    - Watch for short corner routines\n")

  cat("\n")
  cat("=" , rep("=", 70), "\n", sep = "")

  return(invisible(list(
    style = opponent_style,
    weaknesses = weaknesses,
    threats = player_threats
  )))
}

# Generate report for Manchester City
report <- generate_opposition_report("Manchester City", all_team_data, league_metrics, league_avg)

# Radar chart comparison
create_style_radar <- function(team_name, league_metrics, league_avg) {
  # Select key metrics for radar
  radar_metrics <- c("pass_completion", "progressive_pass_pct", "pressures_per_90",
                     "high_press_pct", "xg_per_90", "directness_index")

  team_vals <- league_metrics %>%
    filter(team == team_name) %>%
    select(all_of(radar_metrics)) %>%
    pivot_longer(everything(), names_to = "metric", values_to = "team_value")

  avg_vals <- league_avg %>%
    select(all_of(radar_metrics)) %>%
    pivot_longer(everything(), names_to = "metric", values_to = "avg_value")

  radar_data <- team_vals %>%
    left_join(avg_vals, by = "metric") %>%
    mutate(
      # Normalize to 0-100 scale
      team_norm = (team_value / max(team_value, avg_value)) * 100,
      avg_norm = (avg_value / max(team_value, avg_value)) * 100
    )

  print(radar_data)
}

library(tidyverse)

# Simulate set piece data
set.seed(42)
n_setpieces <- 150

set_piece_data <- tibble(
  match_id = sample(1:15, n_setpieces, replace = TRUE),
  set_piece_type = sample(c("Corner", "Free Kick"), n_setpieces,
                          replace = TRUE, prob = c(0.6, 0.4)),
  side = sample(c("Left", "Right"), n_setpieces, replace = TRUE),
  taker = sample(paste0("Taker_", 1:4), n_setpieces, replace = TRUE),

  # Delivery characteristics
  delivery_type = sample(c("Inswinger", "Outswinger", "Short", "Driven", "Floated"),
                         n_setpieces, replace = TRUE,
                         prob = c(0.35, 0.25, 0.15, 0.15, 0.1)),

  # Target zone (6-yard box: near post, central, far post; edge of box)
  target_zone = sample(c("Near Post", "Central 6-yard", "Far Post",
                         "Penalty Spot", "Edge of Box"),
                       n_setpieces, replace = TRUE,
                       prob = c(0.2, 0.3, 0.25, 0.15, 0.1)),

  # Outcome
  outcome = sample(c("Cleared", "Shot Created", "Goal", "Keeper Claim",
                     "Out of Play", "Second Ball"),
                   n_setpieces, replace = TRUE,
                   prob = c(0.35, 0.2, 0.05, 0.15, 0.15, 0.1)),

  # First contact player
  first_contact = sample(c("Defender_1", "Defender_2", "Midfielder_1",
                           "Forward_1", "Forward_2", "Own Player"),
                        n_setpieces, replace = TRUE,
                        prob = c(0.25, 0.2, 0.15, 0.2, 0.15, 0.05)),

  # xG if shot created
  shot_xg = ifelse(outcome %in% c("Shot Created", "Goal"),
                   runif(n_setpieces, 0.05, 0.35), NA),

  # Free kick distance (for FK only)
  fk_distance = ifelse(set_piece_type == "Free Kick",
                       runif(n_setpieces, 20, 35), NA),
  fk_angle = ifelse(set_piece_type == "Free Kick",
                    sample(c("Central", "Wide Left", "Wide Right"), n_setpieces, replace = TRUE), NA)
)

# Analyze set piece patterns
analyze_set_pieces <- function(data) {

  cat("=" , rep("=", 60), "\n", sep = "")
  cat("SET PIECE ANALYSIS REPORT\n")
  cat("=" , rep("=", 60), "\n\n", sep = "")

  # Overall summary
  cat("1. SET PIECE OVERVIEW\n")
  cat(rep("-", 40), "\n", sep = "")

  summary_stats <- data %>%
    summarise(
      total_set_pieces = n(),
      corners = sum(set_piece_type == "Corner"),
      free_kicks = sum(set_piece_type == "Free Kick"),
      shots_created = sum(outcome %in% c("Shot Created", "Goal")),
      goals = sum(outcome == "Goal"),
      total_xg = sum(shot_xg, na.rm = TRUE)
    )

  cat(sprintf("  Total Set Pieces: %d\n", summary_stats$total_set_pieces))
  cat(sprintf("  Corners: %d | Free Kicks: %d\n", summary_stats$corners, summary_stats$free_kicks))
  cat(sprintf("  Shots Created: %d (%.1f%%)\n", summary_stats$shots_created,
              summary_stats$shots_created / summary_stats$total_set_pieces * 100))
  cat(sprintf("  Goals: %d | Total xG: %.2f\n\n", summary_stats$goals, summary_stats$total_xg))

  # Corner delivery analysis
  cat("2. CORNER DELIVERY PATTERNS\n")
  cat(rep("-", 40), "\n", sep = "")

  corner_patterns <- data %>%
    filter(set_piece_type == "Corner") %>%
    group_by(side, delivery_type) %>%
    summarise(
      count = n(),
      shots = sum(outcome %in% c("Shot Created", "Goal")),
      xg = sum(shot_xg, na.rm = TRUE),
      .groups = "drop"
    ) %>%
    arrange(side, desc(count))

  print(corner_patterns)

  # Most common delivery
  most_common <- corner_patterns %>%
    group_by(delivery_type) %>%
    summarise(total = sum(count), .groups = "drop") %>%
    arrange(desc(total)) %>%
    slice(1)

  cat(sprintf("\n  Most Common Delivery: %s (%d corners)\n\n", most_common$delivery_type, most_common$total))

  # Target zone analysis
  cat("3. TARGET ZONE ANALYSIS\n")
  cat(rep("-", 40), "\n", sep = "")

  target_analysis <- data %>%
    filter(set_piece_type == "Corner") %>%
    group_by(target_zone) %>%
    summarise(
      count = n(),
      pct = n() / nrow(data %>% filter(set_piece_type == "Corner")) * 100,
      xg = sum(shot_xg, na.rm = TRUE),
      goals = sum(outcome == "Goal"),
      .groups = "drop"
    ) %>%
    arrange(desc(count))

  print(target_analysis)

  # Identify preferred zone
  preferred_zone <- target_analysis %>% slice(1)
  cat(sprintf("\n  Primary Target: %s (%.1f%% of corners)\n\n",
              preferred_zone$target_zone, preferred_zone$pct))

  # Aerial threat analysis
  cat("4. AERIAL THREATS\n")
  cat(rep("-", 40), "\n", sep = "")

  aerial_threats <- data %>%
    filter(outcome %in% c("Shot Created", "Goal")) %>%
    group_by(first_contact) %>%
    summarise(
      aerial_wins = n(),
      xg = sum(shot_xg, na.rm = TRUE),
      goals = sum(outcome == "Goal"),
      .groups = "drop"
    ) %>%
    arrange(desc(xg))

  print(aerial_threats)

  # Top threat recommendation
  top_threat <- aerial_threats %>% slice(1)
  cat(sprintf("\n  Primary Aerial Threat: %s (%.2f xG from set pieces)\n", top_threat$first_contact, top_threat$xg))
  cat("  RECOMMENDATION: Assign dedicated marker to this player at all corners\n\n")

  # Free kick analysis
  cat("5. FREE KICK ANALYSIS\n")
  cat(rep("-", 40), "\n", sep = "")

  fk_analysis <- data %>%
    filter(set_piece_type == "Free Kick") %>%
    group_by(fk_angle, delivery_type) %>%
    summarise(
      count = n(),
      xg = sum(shot_xg, na.rm = TRUE),
      .groups = "drop"
    ) %>%
    arrange(desc(xg))

  print(fk_analysis)

  # Defensive recommendations
  cat("\n6. DEFENSIVE RECOMMENDATIONS\n")
  cat(rep("-", 40), "\n", sep = "")

  # Based on analysis
  cat("  CORNERS:\n")

  if (preferred_zone$target_zone == "Near Post") {
    cat("    - Strong man at near post; attack ball aggressively\n")
    cat("    - Second defender 2 yards behind near post marker\n")
  } else if (preferred_zone$target_zone == "Far Post") {
    cat("    - Stack defenders at far post; zonal coverage\n")
    cat("    - Keeper to command 6-yard box\n")
  } else {
    cat("    - Mixed zonal-man marking; key players man-marked\n")
    cat("    - Two at posts for short corner threat\n")
  }

  cat(sprintf("    - Primary marker on: %s\n\n", top_threat$first_contact))

  cat("  FREE KICKS:\n")
  cat("    - Wall: 4-5 players depending on distance\n")
  cat("    - Watch for dummy runs behind wall\n")
  cat("    - One player on post for near post deliveries\n")

  return(invisible(list(
    corner_patterns = corner_patterns,
    target_zones = target_analysis,
    aerial_threats = aerial_threats
  )))
}

# Run analysis
analysis <- analyze_set_pieces(set_piece_data)

# Visualization: Corner target zones
corner_viz <- set_piece_data %>%
  filter(set_piece_type == "Corner") %>%
  count(target_zone, delivery_type)

ggplot(corner_viz, aes(x = target_zone, y = n, fill = delivery_type)) +
  geom_bar(stat = "identity", position = "dodge") +
  labs(
    title = "Corner Delivery Patterns by Target Zone",
    x = "Target Zone", y = "Count", fill = "Delivery Type"
  ) +
  theme_minimal() +
  theme(axis.text.x = element_text(angle = 45, hjust = 1))

library(tidyverse)

# Simulate match state data with tactical changes
set.seed(42)
n_matches <- 15

# Generate match data with state changes
generate_match_state_data <- function(n_matches) {
  map_dfr(1:n_matches, function(match_id) {
    # Each match has multiple time periods
    periods <- tibble(
      match_id = match_id,
      period = c("0-15", "15-30", "30-45", "45-60", "60-75", "75-90"),
      minute_start = c(0, 15, 30, 45, 60, 75),
      minute_end = c(15, 30, 45, 60, 75, 90)
    )

    # Simulate match states and goals
    states <- c("Drawing", "Winning", "Losing")
    current_state <- "Drawing"
    goals_for <- 0
    goals_against <- 0

    periods <- periods %>%
      rowwise() %>%
      mutate(
        goal_scored = rbinom(1, 1, 0.15),
        goal_conceded = rbinom(1, 1, 0.12),
        match_state = {
          if (goal_scored) goals_for <<- goals_for + 1
          if (goal_conceded) goals_against <<- goals_against + 1
          if (goals_for > goals_against) "Winning"
          else if (goals_for < goals_against) "Losing"
          else "Drawing"
        }
      ) %>%
      ungroup()

    # Add tactical metrics that change by state
    periods %>%
      mutate(
        # Formation shifts when losing late
        formation = case_when(
          match_state == "Losing" & minute_start >= 60 ~ sample(c("4-3-3", "3-4-3", "4-2-4"), 1, prob = c(0.3, 0.4, 0.3)),
          match_state == "Winning" & minute_start >= 75 ~ sample(c("5-4-1", "4-5-1", "5-3-2"), 1, prob = c(0.4, 0.4, 0.2)),
          TRUE ~ "4-3-3"
        ),

        # Pressing intensity by state
        pressing_intensity = case_when(
          match_state == "Losing" ~ runif(1, 70, 95),
          match_state == "Winning" & minute_start >= 60 ~ runif(1, 40, 60),
          TRUE ~ runif(1, 55, 75)
        ),

        # Directness by state
        directness = case_when(
          match_state == "Losing" & minute_start >= 60 ~ runif(1, 60, 85),
          match_state == "Winning" ~ runif(1, 35, 55),
          TRUE ~ runif(1, 45, 65)
        ),

        # Defensive line height
        defensive_line = case_when(
          match_state == "Winning" & minute_start >= 60 ~ runif(1, 35, 50),
          match_state == "Losing" ~ runif(1, 55, 70),
          TRUE ~ runif(1, 45, 60)
        ),

        # Width in attack
        attacking_width = case_when(
          match_state == "Losing" ~ runif(1, 65, 85),
          TRUE ~ runif(1, 50, 70)
        ),

        # Crosses per period
        crosses = case_when(
          match_state == "Losing" & minute_start >= 60 ~ rpois(1, 6),
          TRUE ~ rpois(1, 3)
        ),

        # Long balls per period
        long_balls = case_when(
          match_state == "Losing" & minute_start >= 75 ~ rpois(1, 8),
          TRUE ~ rpois(1, 4)
        )
      )
  })
}

match_state_data <- generate_match_state_data(n_matches)

# Substitution patterns
substitution_data <- tibble(
  match_id = rep(1:n_matches, each = 3),
  sub_number = rep(1:3, n_matches),
  minute = c(replicate(n_matches, sort(sample(c(55:90), 3, prob = c(rep(1, 10), rep(2, 15), rep(3, 11)))))),
  match_state_at_sub = sample(c("Drawing", "Winning", "Losing"), n_matches * 3, replace = TRUE, prob = c(0.4, 0.3, 0.3)),
  player_off_type = sample(c("Forward", "Midfielder", "Defender", "Goalkeeper"), n_matches * 3, replace = TRUE, prob = c(0.3, 0.35, 0.3, 0.05)),
  player_on_type = sample(c("Forward", "Midfielder", "Defender"), n_matches * 3, replace = TRUE, prob = c(0.4, 0.35, 0.25))
)

# Analyze tactical changes by match state
analyze_state_tactics <- function(state_data, sub_data) {

  cat("=" , rep("=", 65), "\n", sep = "")
  cat("DYNAMIC GAME MODEL - TACTICAL STATE ANALYSIS\n")
  cat("=" , rep("=", 65), "\n\n", sep = "")

  # 1. Style by Match State
  cat("1. TACTICAL PROFILE BY MATCH STATE\n")
  cat(rep("-", 50), "\n", sep = "")

  state_metrics <- state_data %>%
    group_by(match_state) %>%
    summarise(
      avg_pressing = mean(pressing_intensity),
      avg_directness = mean(directness),
      avg_def_line = mean(defensive_line),
      avg_width = mean(attacking_width),
      avg_crosses = mean(crosses),
      avg_long_balls = mean(long_balls),
      .groups = "drop"
    )

  print(state_metrics)

  cat("\nKey Observations:\n")
  losing_data <- state_metrics %>% filter(match_state == "Losing")
  winning_data <- state_metrics %>% filter(match_state == "Winning")

  cat(sprintf("  When LOSING: Press intensity %.0f%% (+%.0f%% vs winning)\n",
              losing_data$avg_pressing, losing_data$avg_pressing - winning_data$avg_pressing))
  cat(sprintf("  When WINNING: Def line at %.0fm (%.0fm deeper than losing)\n",
              winning_data$avg_def_line, losing_data$avg_def_line - winning_data$avg_def_line))

  # 2. Formation Changes
  cat("\n2. FORMATION CHANGES BY STATE & TIME\n")
  cat(rep("-", 50), "\n", sep = "")

  formation_changes <- state_data %>%
    filter(minute_start >= 60) %>%
    group_by(match_state, formation) %>%
    summarise(count = n(), .groups = "drop") %>%
    arrange(match_state, desc(count))

  print(formation_changes)

  # 3. Substitution Patterns
  cat("\n3. SUBSTITUTION PATTERNS\n")
  cat(rep("-", 50), "\n", sep = "")

  sub_patterns <- sub_data %>%
    group_by(match_state_at_sub, player_on_type) %>%
    summarise(count = n(), avg_minute = mean(minute), .groups = "drop") %>%
    arrange(match_state_at_sub, desc(count))

  print(sub_patterns)

  # Key sub pattern
  losing_subs <- sub_data %>%
    filter(match_state_at_sub == "Losing") %>%
    count(player_on_type) %>%
    arrange(desc(n)) %>%
    slice(1)

  cat(sprintf("\n  When losing, most common sub: %s (%.0f%% of subs)\n",
              losing_subs$player_on_type,
              losing_subs$n / sum(sub_data$match_state_at_sub == "Losing") * 100))

  # 4. Time-Based Tactical Phases
  cat("\n4. TACTICAL PHASES BY TIME PERIOD\n")
  cat(rep("-", 50), "\n", sep = "")

  time_phases <- state_data %>%
    group_by(period) %>%
    summarise(
      pressing = mean(pressing_intensity),
      directness = mean(directness),
      crosses = mean(crosses),
      .groups = "drop"
    )

  print(time_phases)

  # 5. Generate Scenario-Based Game Plans
  cat("\n5. SCENARIO-BASED GAME PLANS\n")
  cat(rep("-", 50), "\n", sep = "")

  cat("\n  SCENARIO A: We are WINNING\n")
  cat("    - Expect them to push higher (def line +", round(losing_data$avg_def_line - winning_data$avg_def_line, 0), "m)\n")
  cat("    - Counter-attacks become more viable\n")
  cat("    - Watch for increased crosses (", round(losing_data$avg_crosses - mean(state_metrics$avg_crosses), 1), " more per 15 min)\n")
  cat("    - Likely formation change to 3-4-3 or 4-2-4 after 60'\n")
  cat("    - Expect striker brought on around 65-70'\n")

  cat("\n  SCENARIO B: We are LOSING\n")
  cat("    - They will drop deeper and reduce pressing\n")
  cat("    - Expect 5-4-1 or 5-3-2 formation after 75'\n")
  cat("    - Counter-attacks become their primary threat\n")
  cat("    - Watch for time-wasting substitutions late\n")

  cat("\n  SCENARIO C: DRAWING late (75'+)\n")
  cat("    - Both teams likely to take more risks\n")
  cat("    - Space opens up in transitions\n")
  cat("    - Set pieces become crucial\n")

  # 6. Key Tactical Triggers
  cat("\n6. KEY TACTICAL TRIGGERS TO WATCH\n")
  cat(rep("-", 50), "\n", sep = "")

  cat("  Minute 60: Formation likely changes if losing\n")
  cat("  Minute 65-70: Attacking substitution if losing\n")
  cat("  Minute 75: Defensive switch if winning (5-back)\n")
  cat("  Minute 80+: Long balls increase significantly (+", round(max(state_data$long_balls[state_data$minute_start >= 75]) - mean(state_data$long_balls), 1), " per period)\n")

  return(invisible(list(
    state_metrics = state_metrics,
    formation_changes = formation_changes,
    sub_patterns = sub_patterns
  )))
}

# Run analysis
game_model <- analyze_state_tactics(match_state_data, substitution_data)

# Visualization
state_summary <- match_state_data %>%
  pivot_longer(cols = c(pressing_intensity, directness, defensive_line),
               names_to = "metric", values_to = "value")

ggplot(state_summary, aes(x = match_state, y = value, fill = metric)) +
  geom_boxplot() +
  facet_wrap(~metric, scales = "free_y") +
  labs(
    title = "Tactical Metrics by Match State",
    x = "Match State", y = "Value"
  ) +
  theme_minimal() +
  theme(legend.position = "none")