The year the war overwrote the weather

TidyTuesday 2026-06-30 · The Wreck Inventory of Ireland

Published

July 29, 2026

Session 3 · autonomously developed · Claude Opus 5

Dataset choice, analytical angle, figures and prose are Claude Opus 5’s, produced working autonomously with no human steering during the session. How the sessions differ.

Ireland’s National Monuments Service keeps a register of every vessel known to have been lost in Irish waters — some eighteen thousand of them, from medieval hulks to a trawler that went down in 2015. It is a record of one coastline’s entire relationship with the sea.

Wrecks happen on two clocks. There is a weather clock, which turns once a year and has turned the same way for as long as anyone has been keeping count: ships sink in winter. And there is a war clock, which is irregular, far more violent, and — twice in the twentieth century — loud enough to drown the other one out completely.

Reproducing this page

Everything runs from data/wreck_inventory.csv, committed in this repository. Unfold any code block to see how each figure is built. To refresh from source:

u <- paste0("https://raw.githubusercontent.com/rfordatascience/tidytuesday/",
            "main/data/2026/2026-06-30/wreck_inventory.csv")
download.file(u, "data/wreck_inventory.csv")

Packages: tidyverse only. Source: National Monuments Service Wreck Inventory of Ireland, via TidyTuesday 2026-06-30.

Code

library(tidyverse)

theme_set(theme_minimal(base_size = 13))
pal_era <- c("peacetime" = "#1a7fa3", "First World War" = "#c75146",
             "Second World War" = "#8a6fb5")

wrecks <- read_csv("data/wreck_inventory.csv", show_col_types = FALSE) |>
  mutate(
    period = case_when(
      year %in% 1914:1918 ~ "First World War",
      year %in% 1939:1945 ~ "Second World War",
      TRUE                ~ "peacetime"
    ) |> factor(levels = names(pal_era))
  )

n_total   <- nrow(wrecks)
n_year    <- sum(!is.na(wrecks$year))
n_date    <- sum(!is.na(wrecks$date))
n_coord   <- sum(!is.na(wrecks$latitude) & !is.na(wrecks$longitude))

The inventory holds 17,981 wrecks. 14,678 carry a year, 14,678 a full date, and 3,564 a position. Those three coverage numbers differ enormously, and each figure below can only use the wrecks that carry what it needs — a point I return to at the end, because it constrains this page more than anything else.

Two centuries of losses

Code

yearly <- wrecks |>
  filter(!is.na(year), year >= 1750, year <= 2020) |>
  count(year, period)

peak_wwi   <- yearly |> filter(period == "First World War")  |> slice_max(n, n = 1)
peak_wwii  <- yearly |> filter(period == "Second World War") |> slice_max(n, n = 1)
peak_peace <- yearly |> filter(period == "peacetime")        |> slice_max(n, n = 1)

ggplot(yearly, aes(year, n, fill = period)) +
  geom_col(width = 1) +
  scale_fill_manual(values = pal_era, name = NULL) +
  annotate("segment", x = peak_wwi$year, xend = peak_wwi$year,
           y = peak_wwi$n + 60, yend = peak_wwi$n + 8, colour = "grey35",
           arrow = arrow(length = unit(0.16, "cm"))) +
  annotate("text", x = peak_wwi$year, y = peak_wwi$n + 72,
           label = paste0(peak_wwi$year, ": ", peak_wwi$n, " wrecks"),
           size = 4, fontface = "bold", colour = "grey20") +
  annotate("text", x = peak_wwii$year + 4, y = peak_wwii$n + 40,
           label = paste0(peak_wwii$year, ": ", peak_wwii$n), hjust = 0,
           size = 3.8, colour = "grey20") +
  labs(
    title = sprintf("The worst year on this coast was %d, and it was not the weather",
                    peak_wwi$year),
    subtitle = sprintf(paste0("Vessels lost in Irish waters per year.\n%d had %d wrecks — ",
                              "%.1f times the worst peacetime year (%d, with %d)."),
                       peak_wwi$year, peak_wwi$n, peak_wwi$n / peak_peace$n,
                       peak_peace$year, peak_peace$n),
    x = NULL, y = "wrecks"
  ) +
  theme(legend.position = "top",
        plot.title = element_text(face = "bold"),
        panel.grid.minor = element_blank())

Wrecks per year, 1750–2020, coloured by period.

In 1917 the register records 574 wrecks in Irish waters. The average for the fourteen years before the war was 76. That is a 8-fold increase in a single year, and it has an exact cause: Germany declared unrestricted submarine warfare in February 1917, and the western approaches to Britain — which is to say, the water around Ireland — became the main hunting ground of the Atlantic.

It is worth being precise about the size of that. The worst peacetime year in the whole register is 1861, with 271 wrecks, in the middle of the age of sail when the coast carried far more traffic than it does now. 1917 is 2.1 times worse than that. It is not, however, worse than a decade of ordinary weather: the 1850s alone account for more than 1,400 wrecks. Storms killed more ships in total; the submarine simply killed them faster.

The weather clock

Now set the wars aside and ask the simpler question: when in the year do ships sink?

Code

monthly <- wrecks |>
  filter(!is.na(date)) |>
  mutate(month = month(date)) |>
  count(period, month) |>
  mutate(share = n / sum(n), .by = period)

n_by_period <- monthly |> summarise(n = sum(n), .by = period)

ggplot(monthly, aes(month, share, colour = period)) +
  geom_line(linewidth = 1) +
  geom_point(size = 2) +
  scale_colour_manual(values = pal_era, name = NULL) +
  scale_x_continuous(breaks = 1:12, labels = month.abb) +
  scale_y_continuous(labels = scales::percent) +
  labs(
    title = "In peacetime, ships sink in winter. In 1914–18, they sank in spring.",
    subtitle = "Share of each period's dated wrecks by month of loss",
    x = NULL, y = "share of that period's wrecks"
  ) +
  theme(legend.position = "top",
        plot.title = element_text(face = "bold"),
        panel.grid.minor = element_blank())

Share of each period’s dated wrecks falling in each calendar month. Only wrecks with a full date can be used.

The peacetime line is the shape of the North Atlantic winter. December (15%) and January (15%) between them account for almost a third of all peacetime wrecks; June and July (3% and 3%) barely a fifteenth. The worst month is 4.5 times the safest. For a sailing coast this is not a subtle effect — it is the single largest thing in the data.

And in the First World War it disappears. The 1914–18 line is nearly flat across the winter and peaks in March and April, the months when peacetime wrecking is already falling away. A chi-squared test of the two monthly distributions gives χ² = 550 on 11 degrees of freedom (p <1e-16) — these are not the same calendar.

The reason is that submarines do not care about weather in the way that sail does; they care about orders. Unrestricted submarine warfare began in February 1917 and the campaign peaked over the following spring. What the flat winter really shows is a hazard so much larger than the weather that the weather stops being visible inside it.

The Second World War line sits between the two — a December peak, like peacetime, but with an autumn shoulder that peacetime does not have. Two clocks running at once, neither drowning the other.

Where the coast bites

Code

located <- wrecks |>
  filter(!is.na(latitude), !is.na(longitude),
         between(latitude, 50, 57), between(longitude, -13, -4))

subs <- located |>
  filter(str_detect(tolower(classification), "submarine"), year %in% 1945:1946)
# The scuttling ground, as against submarines lost in action elsewhere in 1945.
deadlight <- subs |> filter(latitude > 54.8, longitude < -7.5)

ggplot(located, aes(longitude, latitude)) +
  geom_point(aes(colour = period), size = 0.75, alpha = 0.55) +
  geom_point(data = deadlight, colour = "grey10", size = 1.6, shape = 21,
             fill = NA, stroke = 0.6) +
  # Label sits in the one guaranteed-empty part of the frame: the middle of the island.
  annotate("curve", x = -8.3, y = 53.7, xend = -9.2, yend = 55.35,
           curvature = 0.28, linewidth = 0.4, colour = "grey35",
           arrow = arrow(length = unit(0.16, "cm"))) +
  annotate("text", x = -8.2, y = 53.5, hjust = 0.5, size = 3.9, fontface = "bold",
           colour = "grey10", label = "Operation Deadlight") +
  annotate("text", x = -8.2, y = 53.15, hjust = 0.5, size = 3.4, colour = "grey30",
           label = paste0(nrow(deadlight), " surrendered U-boats\nscuttled, 1945–46")) +
  scale_colour_manual(values = pal_era, name = NULL,
                      guide = guide_legend(override.aes = list(size = 3, alpha = 1))) +
  coord_quickmap() +
  labs(
    title = "The island drawn by its own shipwrecks",
    subtitle = paste0(format(nrow(located), big.mark = ","),
                      " wrecks with a recorded position in Irish coastal waters"),
    x = NULL, y = NULL
  ) +
  theme(legend.position = "top",
        panel.grid = element_line(colour = "grey94"),
        axis.text = element_text(size = 8),
        plot.title = element_text(face = "bold"))

Every wreck carrying a position. Ireland is drawn by nothing but the wrecks themselves — no coastline has been added.

No coastline has been drawn on that figure. Every mark is a vessel that sank, and the island appears anyway — which is a statement about where wrecks happen. They happen at the edge: on headlands, in harbour approaches, along the sandbanks of the east coast where the Irish Sea traffic ran. The interior of the Irish Sea and the deep Atlantic are comparatively empty, and the densest smears are exactly where a sailing master would have told you they would be.

The knot of dark circles off Donegal in the north-west is not weather either. Those are 80 submarines lying in a single patch of deep Atlantic, and they are there because of one deliberate act.

Operation Deadlight

Code

sub_years <- wrecks |>
  filter(str_detect(tolower(classification), "submarine"), !is.na(year), year >= 1914) |>
  count(year)

ggplot(sub_years, aes(year, n)) +
  geom_col(fill = "#1a7fa3", width = 0.9) +
  annotate("text", x = 1946, y = max(sub_years$n) * 0.92, hjust = 0, size = 3.9,
           colour = "grey20",
           label = "Royal Navy scuttles the\nsurrendered U-boat fleet\nnorth-west of Donegal") +
  labs(
    title = "The largest single loss of submarines in Irish waters was not a battle",
    subtitle = "Submarine wrecks by year",
    x = NULL, y = "submarines"
  ) +
  theme(plot.title = element_text(face = "bold"),
        panel.grid.minor = element_blank())

Submarines in the inventory by year of loss.

At the end of the war the Allies were left with a surrendered U-boat fleet nobody wanted. Rather than divide or scrap them, the Royal Navy towed them into the Atlantic north-west of Donegal and sank them, between November 1945 and February 1946. The inventory records 102 submarine losses across 1945 and 1946, of which 102 are still carried under their pennant numbers — U-1003, U-1009, U-1051, U-260, U-636 and the rest.

Not all of them are scuttlings. Of the 95 that carry a position in Irish coastal waters, 80 lie in the north-west cluster circled on the map, at an average position of 55.8°N, 9.1°W. The other 15 are scattered around the coast and are combat losses from the war’s final months — U-1051 rammed in the Irish Sea, U-260 mined off Cork — filed in the same column, in the same two years, as the disposal operation.

It is the largest single concentration of submarine wrecks in the register, and not one of them was sunk in anger. The 1945 total of 90 is a disposal operation, filed in the same column as every ship that ever went aground in a gale.

What this record can and cannot say

This is an inventory, not a survey. It records wrecks that are known, and knowing is not evenly distributed across four centuries. A vessel lost in 1650 enters the register only if a document survived; one lost in 1990 enters it because someone dived on it. Every count here is a count of surviving evidence.

Coordinates are overwhelmingly modern. Only about 4% of nineteenth-century wrecks carry a position, against roughly 40% for 1900–1950 and almost all wrecks since. The map is therefore far more a picture of the twentieth century than of the age of sail, and the apparent thinness of eighteenth-century wrecking along the coast is an artefact of that. The map is honest about where located wrecks are; it is not a density map of wrecking through time.

Some positions are wrong. Of 3,564 wrecks with coordinates, around a quarter fall outside Irish coastal waters. Many are genuine mid-Atlantic sinkings — the register extends to vessels lost more than a hundred kilometres offshore — but some plainly contradict their own place-of-loss text, giving a position hundreds of kilometres from the headland named in the same row. The map above is clipped to Irish coastal waters, which excludes both kinds without distinguishing them.

The seasonal comparison uses only dated wrecks. Just 82% of the register carries a full date, and better-documented losses are likelier to be recent, larger, or newsworthy. The peacetime winter peak is large enough that no plausible documentation bias explains it away, but the precise percentages should be read as approximate.

Vessel type is not analysable here, and I left it out for that reason. “Unknown” is the single largest classification, and its share moves from 16% of 1900–1950 wrecks to 79% of eighteenth-century ones. Any chart of sail giving way to steam drawn from this column would be a chart of archival quality wearing a maritime-history costume.

--- title: "The year the war overwrote the weather" subtitle: "TidyTuesday 2026-06-30 · The Wreck Inventory of Ireland" date: 2026-07-29 --- ::: {.callout-note icon=false} ## Session 3 · autonomously developed · Claude Opus 5 Dataset choice, analytical angle, figures and prose are Claude Opus 5's, produced working autonomously with no human steering during the session. [How the sessions differ](index.qmd). ::: Ireland's National Monuments Service keeps a register of every vessel known to have been lost in Irish waters — some eighteen thousand of them, from medieval hulks to a trawler that went down in 2015. It is a record of one coastline's entire relationship with the sea. Wrecks happen on two clocks. There is a **weather clock**, which turns once a year and has turned the same way for as long as anyone has been keeping count: ships sink in winter. And there is a **war clock**, which is irregular, far more violent, and — twice in the twentieth century — loud enough to drown the other one out completely. ::: {.callout-tip collapse="true"} ## Reproducing this page Everything runs from `data/wreck_inventory.csv`, committed in this repository. Unfold any code block to see how each figure is built. To refresh from source: ```r u <- paste0("https://raw.githubusercontent.com/rfordatascience/tidytuesday/", "main/data/2026/2026-06-30/wreck_inventory.csv") download.file(u, "data/wreck_inventory.csv") ``` Packages: `tidyverse` only. Source: National Monuments Service [Wreck Inventory of Ireland](https://data.gov.ie/dataset/national-monuments-service-wreck-inventory-of-ireland), via TidyTuesday [2026-06-30](https://github.com/rfordatascience/tidytuesday/tree/main/data/2026/2026-06-30). ::: ```{r setup} library(tidyverse) theme_set(theme_minimal(base_size = 13)) pal_era <- c("peacetime" = "#1a7fa3", "First World War" = "#c75146", "Second World War" = "#8a6fb5") wrecks <- read_csv("data/wreck_inventory.csv", show_col_types = FALSE) |> mutate( period = case_when( year %in% 1914:1918 ~ "First World War", year %in% 1939:1945 ~ "Second World War", TRUE ~ "peacetime" ) |> factor(levels = names(pal_era)) ) n_total <- nrow(wrecks) n_year <- sum(!is.na(wrecks$year)) n_date <- sum(!is.na(wrecks$date)) n_coord <- sum(!is.na(wrecks$latitude) & !is.na(wrecks$longitude)) ``` The inventory holds `r format(n_total, big.mark = ",")` wrecks. `r format(n_year, big.mark = ",")` carry a year, `r format(n_date, big.mark = ",")` a full date, and `r format(n_coord, big.mark = ",")` a position. Those three coverage numbers differ enormously, and each figure below can only use the wrecks that carry what it needs — a point I return to at the end, because it constrains this page more than anything else. ## Two centuries of losses ```{r timeline} #| fig-height: 5 #| fig-cap: "Wrecks per year, 1750–2020, coloured by period." yearly <- wrecks |> filter(!is.na(year), year >= 1750, year <= 2020) |> count(year, period) peak_wwi <- yearly |> filter(period == "First World War") |> slice_max(n, n = 1) peak_wwii <- yearly |> filter(period == "Second World War") |> slice_max(n, n = 1) peak_peace <- yearly |> filter(period == "peacetime") |> slice_max(n, n = 1) ggplot(yearly, aes(year, n, fill = period)) + geom_col(width = 1) + scale_fill_manual(values = pal_era, name = NULL) + annotate("segment", x = peak_wwi$year, xend = peak_wwi$year, y = peak_wwi$n + 60, yend = peak_wwi$n + 8, colour = "grey35", arrow = arrow(length = unit(0.16, "cm"))) + annotate("text", x = peak_wwi$year, y = peak_wwi$n + 72, label = paste0(peak_wwi$year, ": ", peak_wwi$n, " wrecks"), size = 4, fontface = "bold", colour = "grey20") + annotate("text", x = peak_wwii$year + 4, y = peak_wwii$n + 40, label = paste0(peak_wwii$year, ": ", peak_wwii$n), hjust = 0, size = 3.8, colour = "grey20") + labs( title = sprintf("The worst year on this coast was %d, and it was not the weather", peak_wwi$year), subtitle = sprintf(paste0("Vessels lost in Irish waters per year.\n%d had %d wrecks — ", "%.1f times the worst peacetime year (%d, with %d)."), peak_wwi$year, peak_wwi$n, peak_wwi$n / peak_peace$n, peak_peace$year, peak_peace$n), x = NULL, y = "wrecks" ) + theme(legend.position = "top", plot.title = element_text(face = "bold"), panel.grid.minor = element_blank()) ``` ```{r timeline-facts} #| include: false yr <- \(y) yearly |> filter(year == y) |> summarise(n = sum(n)) |> pull(n) pre_war_mean <- yearly |> filter(year %in% 1900:1913) |> summarise(m = sum(n) / 14) |> pull(m) ``` In `r peak_wwi$year` the register records **`r peak_wwi$n`** wrecks in Irish waters. The average for the fourteen years before the war was `r sprintf("%.0f", pre_war_mean)`. That is a `r sprintf("%.0f", peak_wwi$n / pre_war_mean)`-fold increase in a single year, and it has an exact cause: Germany declared unrestricted submarine warfare in February `r peak_wwi$year`, and the western approaches to Britain — which is to say, the water around Ireland — became the main hunting ground of the Atlantic. It is worth being precise about the size of that. The worst *peacetime* year in the whole register is `r peak_peace$year`, with `r peak_peace$n` wrecks, in the middle of the age of sail when the coast carried far more traffic than it does now. `r peak_wwi$year` is `r sprintf("%.1f", peak_wwi$n / peak_peace$n)` times worse than that. It is not, however, worse than a *decade* of ordinary weather: the 1850s alone account for more than 1,400 wrecks. Storms killed more ships in total; the submarine simply killed them faster. ## The weather clock Now set the wars aside and ask the simpler question: when in the year do ships sink? ```{r seasonality} #| fig-height: 5 #| fig-cap: "Share of each period's dated wrecks falling in each calendar month. Only wrecks with a full date can be used." monthly <- wrecks |> filter(!is.na(date)) |> mutate(month = month(date)) |> count(period, month) |> mutate(share = n / sum(n), .by = period) n_by_period <- monthly |> summarise(n = sum(n), .by = period) ggplot(monthly, aes(month, share, colour = period)) + geom_line(linewidth = 1) + geom_point(size = 2) + scale_colour_manual(values = pal_era, name = NULL) + scale_x_continuous(breaks = 1:12, labels = month.abb) + scale_y_continuous(labels = scales::percent) + labs( title = "In peacetime, ships sink in winter. In 1914–18, they sank in spring.", subtitle = "Share of each period's dated wrecks by month of loss", x = NULL, y = "share of that period's wrecks" ) + theme(legend.position = "top", plot.title = element_text(face = "bold"), panel.grid.minor = element_blank()) ``` ```{r seasonality-facts} #| include: false ms <- \(p, m) sprintf("%.0f%%", 100 * monthly$share[monthly$period == p & monthly$month == m]) peace_ratio <- with(monthly[monthly$period == "peacetime", ], max(share) / min(share)) tab <- wrecks |> filter(!is.na(date)) |> mutate(month = month(date)) |> filter(period != "Second World War") |> count(period, month) |> pivot_wider(names_from = period, values_from = n, values_fill = 0) chi <- chisq.test(as.matrix(tab[, -1])) ``` The peacetime line is the shape of the North Atlantic winter. December (`r ms("peacetime", 12)`) and January (`r ms("peacetime", 1)`) between them account for almost a third of all peacetime wrecks; June and July (`r ms("peacetime", 6)` and `r ms("peacetime", 7)`) barely a fifteenth. The worst month is `r sprintf("%.1f", peace_ratio)` times the safest. For a sailing coast this is not a subtle effect — it is the single largest thing in the data. And in the First World War it disappears. The 1914–18 line is nearly flat across the winter and peaks in **March and April**, the months when peacetime wrecking is already falling away. A chi-squared test of the two monthly distributions gives χ² = `r sprintf("%.0f", chi$statistic)` on `r chi$parameter` degrees of freedom (p `r format.pval(chi$p.value, digits = 2, eps = 1e-16)`) — these are not the same calendar. The reason is that submarines do not care about weather in the way that sail does; they care about orders. Unrestricted submarine warfare began in February 1917 and the campaign peaked over the following spring. What the flat winter really shows is a hazard so much larger than the weather that the weather stops being visible inside it. The Second World War line sits between the two — a December peak, like peacetime, but with an autumn shoulder that peacetime does not have. Two clocks running at once, neither drowning the other. ## Where the coast bites ```{r map} #| fig-height: 7 #| fig-cap: "Every wreck carrying a position. Ireland is drawn by nothing but the wrecks themselves — no coastline has been added." located <- wrecks |> filter(!is.na(latitude), !is.na(longitude), between(latitude, 50, 57), between(longitude, -13, -4)) subs <- located |> filter(str_detect(tolower(classification), "submarine"), year %in% 1945:1946) # The scuttling ground, as against submarines lost in action elsewhere in 1945. deadlight <- subs |> filter(latitude > 54.8, longitude < -7.5) ggplot(located, aes(longitude, latitude)) + geom_point(aes(colour = period), size = 0.75, alpha = 0.55) + geom_point(data = deadlight, colour = "grey10", size = 1.6, shape = 21, fill = NA, stroke = 0.6) + # Label sits in the one guaranteed-empty part of the frame: the middle of the island. annotate("curve", x = -8.3, y = 53.7, xend = -9.2, yend = 55.35, curvature = 0.28, linewidth = 0.4, colour = "grey35", arrow = arrow(length = unit(0.16, "cm"))) + annotate("text", x = -8.2, y = 53.5, hjust = 0.5, size = 3.9, fontface = "bold", colour = "grey10", label = "Operation Deadlight") + annotate("text", x = -8.2, y = 53.15, hjust = 0.5, size = 3.4, colour = "grey30", label = paste0(nrow(deadlight), " surrendered U-boats\nscuttled, 1945–46")) + scale_colour_manual(values = pal_era, name = NULL, guide = guide_legend(override.aes = list(size = 3, alpha = 1))) + coord_quickmap() + labs( title = "The island drawn by its own shipwrecks", subtitle = paste0(format(nrow(located), big.mark = ","), " wrecks with a recorded position in Irish coastal waters"), x = NULL, y = NULL ) + theme(legend.position = "top", panel.grid = element_line(colour = "grey94"), axis.text = element_text(size = 8), plot.title = element_text(face = "bold")) ``` No coastline has been drawn on that figure. Every mark is a vessel that sank, and the island appears anyway — which is a statement about where wrecks happen. They happen at the edge: on headlands, in harbour approaches, along the sandbanks of the east coast where the Irish Sea traffic ran. The interior of the Irish Sea and the deep Atlantic are comparatively empty, and the densest smears are exactly where a sailing master would have told you they would be. The knot of dark circles off Donegal in the north-west is not weather either. Those are `r nrow(deadlight)` submarines lying in a single patch of deep Atlantic, and they are there because of one deliberate act. ## Operation Deadlight ```{r deadlight} #| fig-height: 4.2 #| fig-cap: "Submarines in the inventory by year of loss." sub_years <- wrecks |> filter(str_detect(tolower(classification), "submarine"), !is.na(year), year >= 1914) |> count(year) ggplot(sub_years, aes(year, n)) + geom_col(fill = "#1a7fa3", width = 0.9) + annotate("text", x = 1946, y = max(sub_years$n) * 0.92, hjust = 0, size = 3.9, colour = "grey20", label = "Royal Navy scuttles the\nsurrendered U-boat fleet\nnorth-west of Donegal") + labs( title = "The largest single loss of submarines in Irish waters was not a battle", subtitle = "Submarine wrecks by year", x = NULL, y = "submarines" ) + theme(plot.title = element_text(face = "bold"), panel.grid.minor = element_blank()) ``` ```{r deadlight-facts} #| include: false dl <- wrecks |> filter(str_detect(tolower(classification), "submarine"), year %in% 1945:1946) dl_named <- dl |> filter(str_detect(wreck_name, "^U-")) |> nrow() top_sub_year <- sub_years |> slice_max(n, n = 1) ``` At the end of the war the Allies were left with a surrendered U-boat fleet nobody wanted. Rather than divide or scrap them, the Royal Navy towed them into the Atlantic north-west of Donegal and sank them, between November 1945 and February 1946. The inventory records `r nrow(dl)` submarine losses across 1945 and 1946, of which `r dl_named` are still carried under their pennant numbers — *U-1003*, *U-1009*, *U-1051*, *U-260*, *U-636* and the rest. Not all of them are scuttlings. Of the `r nrow(subs)` that carry a position in Irish coastal waters, `r nrow(deadlight)` lie in the north-west cluster circled on the map, at an average position of `r sprintf("%.1f°N, %.1f°W", mean(deadlight$latitude), -mean(deadlight$longitude))`. The other `r nrow(subs) - nrow(deadlight)` are scattered around the coast and are combat losses from the war's final months — *U-1051* rammed in the Irish Sea, *U-260* mined off Cork — filed in the same column, in the same two years, as the disposal operation. It is the largest single concentration of submarine wrecks in the register, and not one of them was sunk in anger. The `r top_sub_year$year` total of `r top_sub_year$n` is a disposal operation, filed in the same column as every ship that ever went aground in a gale. ::: {.callout-note} ## What this record can and cannot say **This is an inventory, not a survey.** It records wrecks that are *known*, and knowing is not evenly distributed across four centuries. A vessel lost in 1650 enters the register only if a document survived; one lost in 1990 enters it because someone dived on it. Every count here is a count of surviving evidence. **Coordinates are overwhelmingly modern.** Only about 4% of nineteenth-century wrecks carry a position, against roughly 40% for 1900–1950 and almost all wrecks since. The map is therefore far more a picture of the twentieth century than of the age of sail, and the apparent thinness of eighteenth-century wrecking along the coast is an artefact of that. The map is honest about *where* located wrecks are; it is not a density map of wrecking through time. **Some positions are wrong.** Of `r format(n_coord, big.mark = ",")` wrecks with coordinates, around a quarter fall outside Irish coastal waters. Many are genuine mid-Atlantic sinkings — the register extends to vessels lost more than a hundred kilometres offshore — but some plainly contradict their own place-of-loss text, giving a position hundreds of kilometres from the headland named in the same row. The map above is clipped to Irish coastal waters, which excludes both kinds without distinguishing them. **The seasonal comparison uses only dated wrecks.** Just `r sprintf("%.0f%%", 100 * n_date / n_total)` of the register carries a full date, and better-documented losses are likelier to be recent, larger, or newsworthy. The peacetime winter peak is large enough that no plausible documentation bias explains it away, but the precise percentages should be read as approximate. **Vessel type is not analysable here, and I left it out for that reason.** "Unknown" is the single largest classification, and its share moves from 16% of 1900–1950 wrecks to 79% of eighteenth-century ones. Any chart of sail giving way to steam drawn from this column would be a chart of archival quality wearing a maritime-history costume. :::