The Defense Showed Up. The Offense Never Finished.

Week 2 2026 — Syracuse 18, California 21 — exhibits for publication

library(tidyverse)
library(arrow)
library(gt)
library(ggrepel)
library(scales)

source("../R/04_storyline_engine.R")
source("../R/05_viz_style.R")

TEAM   <- "Syracuse"
OPP    <- "California"
SEASON <- params$season
WEEK   <- params$week

FIG_DIR <- sprintf("figures/%d_wk%02d_story", SEASON, WEEK)
SCOPE   <- sprintf("%d Week %d | %s vs %s", SEASON, WEEK, TEAM, OPP)

hist <- read_parquet("~/cfb-data/processed/team_game_metrics.parquet")
cur  <- read_parquet("~/cfb-data/processed/team_game_metrics_2026.parquet")
raw  <- read_parquet(sprintf("~/cfb-data/raw/pbp_inseason/pbp_%d_wk%02d.parquet",
                             SEASON, WEEK))

fbs_teams <- unique(hist$team[hist$is_fbs_matchup])

# The peer group: every FBS-vs-FBS team-game this week. Cal is FBS, so unlike
# Week 1 there is no FCS discount to apply -- these are true peers.
peers <- cur |>
  filter(team %in% fbs_teams, !low_volume, is_fbs_matchup, week == WEEK)

su  <- peers |> filter(team == TEAM)
su1 <- cur   |> filter(team == TEAM, week == 1)

peer_rank <- function(col, higher_better = TRUE) {
  v <- peers[[col]]
  r <- rank(if (higher_better) -v else v, na.last = "keep")
  as.integer(round(r[peers$team == TEAM]))
}
n_peers <- nrow(peers)

Exhibit 1 — The defense showed up

def_rank <- peer_rank("def_epa_play", higher_better = FALSE)

d1 <- peers |>
  mutate(is_su = team == TEAM) |>
  arrange(desc(def_epa_play)) |>
  mutate(team = factor(team, levels = team))

p_def <- ggplot(d1, aes(x = def_epa_play, y = team, color = is_su)) +
  geom_vline(xintercept = median(d1$def_epa_play), linetype = "dashed",
             color = SU$muted, linewidth = 0.4) +
  geom_point(aes(size = is_su)) +
  ggrepel::geom_text_repel(
    data = filter(d1, is_su),
    aes(label = sprintf("%s — %+.3f EPA allowed\n%s of %d",
                        team, def_epa_play, scales::ordinal(def_rank), n_peers)),
    nudge_x = 0.22, size = 3.4, fontface = "bold", segment.color = NA,
    lineheight = 0.95, force = 0
  ) +
  scale_color_manual(values = c(`FALSE` = SU$muted, `TRUE` = SU$orange), guide = "none") +
  scale_size_manual(values = c(`FALSE` = 1.6, `TRUE` = 3.6), guide = "none") +
  labs(
    title = sprintf("Syracuse's defense ranked %s of %d against FBS opposition in Week 2",
                    scales::ordinal(def_rank), n_peers),
    subtitle = "EPA allowed per play — further left is better | Dashed line = group median",
    x = "Defensive EPA allowed per play", y = NULL,
    caption = su_caption(SCOPE)
  ) +
  theme_su(legend_pos = "none") +
  theme(axis.text.y = element_text(size = 5.5))

save_fig(p_def, "01_defense_vs_peers", height = 8)

The defense was also the week’s biggest improvement: EPA allowed moved from +0.098 in Week 1 to -0.165 against a Power Four opponent.

tibble::tibble(
  Metric = c("EPA allowed per play", "Success rate allowed",
             "Havoc rate", "Sack rate"),
  `Week 1` = c(su1$def_epa_play, su1$def_success_rate,
               su1$def_havoc_core, su1$def_sack_rate),
  `Week 2` = c(su$def_epa_play, su$def_success_rate,
               su$def_havoc_core, su$def_sack_rate),
  `Rank of 96` = c(peer_rank("def_epa_play", FALSE), peer_rank("def_success_rate", FALSE),
                   peer_rank("def_havoc_core"), peer_rank("def_sack_rate"))
) |>
  gt() |>
  tab_header(
    title = md("**Efficient without being disruptive**"),
    subtitle = md("Syracuse held California down while generating almost no negative plays — an unusual profile worth watching")
  ) |>
  fmt_number(columns = c(`Week 1`, `Week 2`), decimals = 3) |>
  cols_align("center", columns = -Metric) |>
  tab_style(style = cell_text(weight = "bold"),
            locations = cells_body(columns = Metric)) |>
  tab_source_note(md(paste0("**", su_caption(SCOPE),
    "** — ranks are among the ", n_peers,
    " FBS teams that played FBS opponents in Week 2."))) |>
  gt_theme_su()

Efficient without being disruptive
Syracuse held California down while generating almost no negative plays — an unusual profile worth watching
Metric	Week 1	Week 2	Rank of 96
EPA allowed per play	0.098	−0.165	16
Success rate allowed	0.276	0.350	25
Havoc rate	0.103	0.017	72
Sack rate	0.161	0.029	73
Data: CollegeFootballData via cfbfastR | 2026 Week 2 | Syracuse vs California — ranks are among the 96 FBS teams that played FBS opponents in Week 2.

Exhibit 2 — But Syracuse could not stay on schedule

sd_rank <- peer_rank("off_sr_standard", higher_better = TRUE)

d2 <- peers |>
  filter(!is.na(off_sr_standard)) |>
  mutate(is_su = team == TEAM) |>
  arrange(off_sr_standard) |>
  mutate(team = factor(team, levels = team))

p_std <- ggplot(d2, aes(x = off_sr_standard, y = team, color = is_su)) +
  geom_vline(xintercept = median(d2$off_sr_standard), linetype = "dashed",
             color = SU$muted, linewidth = 0.4) +
  geom_point(aes(size = is_su)) +
  ggrepel::geom_text_repel(
    data = filter(d2, is_su),
    aes(label = sprintf("%s — %.1f%% on standard downs\n%s of %d",
                        team, 100 * off_sr_standard,
                        scales::ordinal(sd_rank), nrow(d2))),
    nudge_x = 0.1, size = 3.4, fontface = "bold", segment.color = NA,
    lineheight = 0.95, force = 0
  ) +
  scale_color_manual(values = c(`FALSE` = SU$muted, `TRUE` = SU$orange), guide = "none") +
  scale_size_manual(values = c(`FALSE` = 1.6, `TRUE` = 3.6), guide = "none") +
  scale_x_continuous(labels = scales::percent) +
  labs(
    title = sprintf("On first and second down, Syracuse ranked %s of %d",
                    scales::ordinal(sd_rank), nrow(d2)),
    subtitle = sprintf(
      "Standard-downs success rate — further right is better | Syracuse fell from %.1f%% in Week 1 to %.1f%% | Dashed line = group median",
      100 * su1$off_sr_standard, 100 * su$off_sr_standard),
    x = "Standard-downs success rate", y = NULL,
    caption = su_caption(SCOPE)
  ) +
  theme_su(legend_pos = "none") +
  theme(axis.text.y = element_text(size = 5.5))

save_fig(p_std, "02_standard_downs_vs_peers", height = 8)

Failing on early downs is the mechanism behind everything in Exhibit 3. An offense that does not win first and second down arrives at third-and-long, stalls short of the end zone, and settles for field goals.

Exhibit 3 — Nine trips inside the 40, eighteen points

drives <- raw |>
  filter(pos_team == TEAM) |>
  group_by(drive_id) |>
  summarise(
    start_ytg = first(drive_start_yards_to_goal),
    closest   = min(yards_to_goal, na.rm = TRUE),
    pts       = first(drive_pts),
    result    = first(drive_result),
    plays     = n(),
    .groups = "drop"
  ) |>
  filter(!is.na(start_ytg), is.finite(closest))

opps <- drives |> filter(closest <= 40) |> arrange(closest)

opps |>
  transmute(
    `Drive started` = sprintf("own %.0f", 100 - start_ytg),
    `Got as close as` = ifelse(closest == 0, "the goal line",
                               sprintf("the %s %.0f", OPP, closest)),
    Plays = plays,
    Result = result,
    Points = pts
  ) |>
  gt() |>
  tab_header(
    title = md("**Syracuse reached California's 40-yard line nine times and scored 18 points**"),
    subtitle = md("Two missed field goals and an interception at the goal line, in a three-point loss")
  ) |>
  cols_align("center", columns = c(Plays, Points)) |>
  cols_align("left", columns = c(`Drive started`, `Got as close as`, Result)) |>
  data_color(columns = Points, method = "numeric", palette = SU_SEQ) |>
  tab_style(
    style = list(cell_fill(color = "#FFF3E6"), cell_text(weight = "bold")),
    locations = cells_body(rows = Points == 0)
  ) |>
  tab_source_note(md(paste0(
    "**", su_caption(SCOPE), "** — a scoring opportunity is any drive reaching the opponent's 40. ",
    sprintf("Syracuse averaged %.2f points per trip, %s of %d nationally.",
            su$off_pts_per_opp, scales::ordinal(peer_rank("off_pts_per_opp")), n_peers)))) |>
  gt_theme_su()

Syracuse reached California’s 40-yard line nine times and scored 18 points
Two missed field goals and an interception at the goal line, in a three-point loss
Drive started	Got as close as	Plays	Result	Points
own 28	the goal line	6	TD	6
own 52	the goal line	8	INT	0
own 100	the goal line	22	MISSED FG	0
own 25	the California 2	15	TD	6
own 36	the California 7	7	FG	3
own 40	the California 15	15	FG	3
own 25	the California 22	14	MISSED FG	0
own 28	the California 28	2	PUNT	0
own 20	the California 40	7	PUNT	0
Data: CollegeFootballData via cfbfastR | 2026 Week 2 | Syracuse vs California — a scoring opportunity is any drive reaching the opponent’s 40. Syracuse averaged 2.25 points per trip, 83rd of 96 nationally.

Exhibit 4 — Outplayed California, and lost anyway

# Final scores reconstructed from the play-by-play, so this never depends on a
# live API call. A handful of games carry corrupted score fields, so margins
# are filtered to a plausible range before use.
scores <- bind_rows(
  raw |> transmute(game_id, team = pos_team,     pts = pos_team_score),
  raw |> transmute(game_id, team = def_pos_team, pts = def_pos_team_score)
) |>
  filter(!is.na(team), !is.na(pts), pts >= 0, pts <= 100) |>
  group_by(game_id, team) |>
  summarise(final = max(pts, na.rm = TRUE), .groups = "drop")

margins <- scores |>
  group_by(game_id) |>
  filter(n() == 2) |>
  mutate(margin = final - rev(final)) |>
  ungroup() |>
  filter(abs(margin) <= 80)

j <- peers |>
  inner_join(margins, by = c("game_id", "team")) |>
  mutate(is_su = team == TEAM,
         won_epa = net_epa_play > 0,
         won_game = margin > 0)

share_won <- mean(j$won_game[j$won_epa])
n_epa_win <- sum(j$won_epa)

p_scatter <- ggplot(j, aes(x = net_epa_play, y = margin)) +
  annotate("rect", xmin = 0, xmax = Inf, ymin = -Inf, ymax = 0,
           fill = SU$orange, alpha = 0.06) +
  geom_hline(yintercept = 0, color = SU$muted, linewidth = 0.4) +
  geom_vline(xintercept = 0, color = SU$muted, linewidth = 0.4) +
  geom_point(data = filter(j, !is_su), color = SU$muted, alpha = 0.5, size = 2.2) +
  geom_point(data = filter(j, is_su), color = SU$orange, size = 4.2) +
  ggrepel::geom_text_repel(
    data = filter(j, is_su),
    aes(label = sprintf("%s\n%+.3f EPA margin, lost by %d", team, net_epa_play, abs(margin))),
    nudge_y = -14, nudge_x = 0.05, size = 3.4, fontface = "bold",
    segment.color = SU$orange, lineheight = 0.95
  ) +
  annotate("text", x = Inf, y = -Inf, hjust = 1.05, vjust = -1.2,
           label = "Won the efficiency battle, lost the game",
           size = 3.2, fontface = "italic", color = SU$ink2) +
  labs(
    title = sprintf("%d of %d teams that won the efficiency battle in Week 2 won the game. Syracuse did not.",
                    sum(j$won_epa & j$won_game), n_epa_win),
    subtitle = sprintf(
      "Net EPA per play against final margin | %d FBS-vs-FBS team-games | Shaded quadrant: outplayed the opponent and still lost",
      nrow(j)),
    x = "Net EPA per play (offense minus defense allowed)",
    y = "Final point margin",
    caption = su_caption(SCOPE)
  ) +
  theme_su(legend_pos = "none")

save_fig(p_scatter, "04_outplayed_and_lost", height = 6)

Among the 16 teams that won the efficiency battle and lost, Syracuse’s 2.25 points per scoring opportunity was among the lowest. The pattern is not bad luck — it is what happens to a team that reaches scoring range and cannot convert.

Figures written to disk

figs <- list.files(FIG_DIR, pattern = "\\.png$", full.names = TRUE)
if (length(figs) == 0) {
  cat("No figures written.\n")
} else {
  cat("300-dpi PNGs ready for the piece:\n\n")
  for (f in figs) cat("  ", f, "  (", round(file.info(f)$size / 1024), " KB)\n", sep = "")
  cat("\nThe gt tables render above — copy them from the browser into Google Docs,\n")
  cat("where they paste as native editable tables.\n")
}

## 300-dpi PNGs ready for the piece:
## 
##   figures/2026_wk02_story/01_defense_vs_peers.png  (350 KB)
##   figures/2026_wk02_story/02_standard_downs_vs_peers.png  (359 KB)
##   figures/2026_wk02_story/04_outplayed_and_lost.png  (227 KB)
## 
## The gt tables render above — copy them from the browser into Google Docs,
## where they paste as native editable tables.