diff --git a/R/climate_indices.R b/R/climate_indices.R
index f299888..1f9fddb 100644
--- a/R/climate_indices.R
+++ b/R/climate_indices.R
@@ -53,6 +53,27 @@ climate_indices <- function() {
       description = paste("Leading mode of North Pacific SST variability.",
                           "Included for completeness; of limited relevance to",
                           "Atlantic shelf systems.")
+    ),
+    LCR = list(
+      label = "Labrador Current retroflection",
+      source = "Jutras et al. 2023, Nature Communications",
+      url = paste0("https://static-content.springer.com/esm/",
+                   "art%3A10.1038%2Fs41467-023-38321-y/MediaObjects/",
+                   "41467_2023_38321_MOESM5_ESM.csv"),
+      format = "decimal_year_csv",
+      reference = paste("Jutras M, Dufour CO, Mucci A, Talbot LC (2023)",
+                        "Large-scale control of the retroflection of the",
+                        "Labrador Current. Nature Communications 14:2623.",
+                        "doi:10.1038/s41467-023-38321-y"),
+      description = paste("How much of the Labrador Current turns eastward at",
+                          "the Grand Banks instead of continuing southwest along",
+                          "the shelf. Positive values mean stronger retroflection,",
+                          "so less cold, fresh, oxygen-rich Labrador water reaches",
+                          "the Scotian Shelf and Gulf of Maine. Unlike the",
+                          "atmospheric indices here it describes a current rather",
+                          "than a pressure or temperature pattern, which makes it",
+                          "the more direct predictor of shelf water properties.",
+                          "Covers 1993-2014 only.")
     )
   )
 }
@@ -68,7 +89,12 @@ index_dictionary <- function() {
   dictionary <- do.call(rbind, lapply(names(catalog), function(name) {
     entry <- catalog[[name]]
     data.frame(name = name, label = entry$label, source = entry$source,
-               url = entry$url, description = entry$description,
+               url = entry$url,
+               # Indices published with a paper carry its citation. The
+               # operational ones from NOAA have no single paper to point at,
+               # so this is empty for them rather than invented.
+               reference = entry$reference %||% NA_character_,
+               description = entry$description,
                stringsAsFactors = FALSE)
   }))
   class(dictionary) <- c("datamatch_index_dictionary", "data.frame")
@@ -85,7 +111,16 @@ print.datamatch_index_dictionary <- function(x, ...) {
   cat(strrep("-", 62), "\n", sep = "")
   print(flat[c("name", "label", "source")], row.names = FALSE, right = FALSE)
   cat("\nThese have no spatial dimension: one value per month, basin-wide.\n")
-  cat("Sources: as.data.frame(index_dictionary())$url\n")
+
+  cited <- flat[!is.na(flat$reference), ]
+  if (nrow(cited) > 0) {
+    cat("\nCite when used:\n")
+    for (i in seq_len(nrow(cited))) {
+      cat("  ", cited$name[i], ": ", cited$reference[i], "\n", sep = "")
+    }
+  }
+
+  cat("\nSources: as.data.frame(index_dictionary())$url\n")
   invisible(x)
 }
 
@@ -97,6 +132,18 @@ print.datamatch_index_dictionary <- function(x, ...) {
 #' The published files use fixed-width year-by-month tables with provider-specific
 #' missing-value codes, which are parsed here into one row per month.
 #'
+#' @section Citing an index:
+#' `LCR` is the published output of a specific study rather than an operational
+#' product, and should be cited when used:
+#'
+#' > Jutras M, Dufour CO, Mucci A, Talbot LC (2023) Large-scale control of the
+#' > retroflection of the Labrador Current. *Nature Communications* **14**:2623.
+#' > \doi{10.1038/s41467-023-38321-y}
+#'
+#' The series is the source data published with that paper's Figure 3, fetched
+#' from the journal rather than recomputed, so the values are the authors' own.
+#' `as.data.frame(index_dictionary())$reference` carries this at runtime.
+#'
 #' @param index an index name from [climate_indices()]
 #' @param years years to keep; `NULL` keeps the whole record
 #' @param url override the catalog URL, if a provider has moved the file
@@ -145,6 +192,13 @@ fetch_climate_index <- function(index, years = NULL, url = NULL) {
 #' @return a data frame with `YEAR`, `MONTH`, and the named value column
 #' @keywords internal
 parse_index_table <- function(lines, format, name) {
+  # Not every published index is a year-by-month table. The retroflection index
+  # is a daily series in decimal years, so it is parsed separately rather than
+  # bent into the shape of the CPC and PSL files.
+  if (format == "decimal_year_csv") {
+    return(parse_decimal_year_csv(lines, name))
+  }
+
   lines <- trimws(lines)
   lines <- lines[nzchar(lines)]
 
@@ -196,6 +250,79 @@ parse_index_table <- function(lines, format, name) {
   series
 }
 
+#' Parse a daily index published as decimal years
+#'
+#' The format the Jutras et al. retroflection index is published in: a couple of
+#' figure-caption lines, a `Date,<label>` header, then one row per day with the
+#' date as a decimal year.
+#'
+#' @section Two conversions worth stating:
+#' **Decimal years use a fixed 365-day year here.** The step between consecutive
+#' rows is 1/365 to eleven significant figures, so leap days are not represented
+#' and a date is recovered as day `round(fraction * 365)` of its year. Assuming
+#' 365.25 instead would walk the derived dates off by up to several days over the
+#' record.
+#'
+#' **Daily values are averaged to monthly.** Every other index in this package is
+#' monthly, and [attach_climate_index()] joins on year and month, so a daily
+#' series has nothing to join to. The underlying index is already smoothed with a
+#' 12-month rolling mean, so monthly averaging discards very little.
+#'
+#' Months backed by fewer than half their days are dropped rather than reported.
+#' The record's first and last months are partial by construction, and the fixed
+#' 365-day year rolls the final record into a January of its own — which would
+#' otherwise surface as a confident-looking monthly value resting on a single day.
+#'
+#' @param lines the downloaded file's lines
+#' @param name what to call the value column
+#' @return a data frame with `YEAR`, `MONTH`, and the named value column
+#' @keywords internal
+parse_decimal_year_csv <- function(lines, name) {
+  lines <- trimws(lines)
+  # A data row is two comma-separated numbers. The figure-caption and header
+  # lines fail that, which is steadier than skipping a fixed number of lines.
+  is_data <- grepl("^-?[0-9.]+,-?[0-9.eE+-]+$", lines)
+  rows <- lines[is_data]
+
+  if (length(rows) == 0) {
+    stop("No data rows found in the index file; its format may have changed.",
+         call. = FALSE)
+  }
+
+  fields <- do.call(rbind, strsplit(rows, ","))
+  decimal_year <- as.numeric(fields[, 1])
+  value <- as.numeric(fields[, 2])
+  keep <- !is.na(decimal_year) & !is.na(value)
+  decimal_year <- decimal_year[keep]
+  value <- value[keep]
+
+  year <- floor(decimal_year)
+  date <- as.Date(paste0(year, "-01-01")) + round((decimal_year - year) * 365)
+
+  daily <- data.frame(YEAR = as.integer(format(date, "%Y")),
+                      MONTH = as.integer(format(date, "%m")),
+                      value = value)
+
+  monthly <- stats::aggregate(value ~ YEAR + MONTH, data = daily,
+                              FUN = function(z) c(mean(z), length(z)))
+  monthly <- data.frame(YEAR = monthly$YEAR, MONTH = monthly$MONTH,
+                        value = monthly$value[, 1], days = monthly$value[, 2])
+
+  # A month backed by a handful of days is not that month's value. The record's
+  # first and last months are partial by construction, and the fixed 365-day
+  # year rolls the final record into a January of its own - which would
+  # otherwise appear as a confident-looking value resting on one day. Half a
+  # month is the same bar min_coverage sets by default elsewhere here.
+  full <- as.numeric(lubridate::days_in_month(
+    lubridate::ymd(paste(monthly$YEAR, monthly$MONTH, 1, sep = "-"))))
+  monthly <- monthly[monthly$days >= full / 2, ]
+
+  monthly <- monthly[order(monthly$YEAR, monthly$MONTH), c("YEAR", "MONTH", "value")]
+  names(monthly)[names(monthly) == "value"] <- name
+  rownames(monthly) <- NULL
+  monthly
+}
+
 #' Missing-value code declared in a PSL index file
 #'
 #' PSL files state their own missing-value code on a line after the data, rather
diff --git a/README.Rmd b/README.Rmd
index 0cd604a..5e62448 100644
--- a/README.Rmd
+++ b/README.Rmd
@@ -436,7 +436,7 @@ so every observation in a month receives the same number:
 ```{r indices, eval = FALSE}
 observations <- attach_climate_index(observations, c("NAO", "AMO"))
 
-index_dictionary()          # NAO, AO, AMO, PDO
+index_dictionary()          # NAO, AO, AMO, PDO, LCR
 ```
 
 That makes them a different kind of covariate from local temperature. They carry
@@ -445,6 +445,38 @@ within a region conditions are better. Useful for interannual questions ("was
 this a warm-regime year?"), useless for spatial ones: a model given only indices
 cannot produce a map.
 
+#### The Labrador Current retroflection index
+
+`LCR` is the odd one out and often the most directly useful of the five. The
+other four are atmospheric patterns; this one describes a *current* — how much
+of the Labrador Current turns eastward at the Grand Banks instead of continuing
+southwest along the shelf. Positive values mean stronger retroflection, so less
+cold, fresh, oxygen-rich Labrador water reaching the Scotian Shelf and Gulf of
+Maine. For shelf water properties that is a shorter causal chain than the NAO.
+
+```{r lcr, eval = FALSE}
+lcr <- fetch_climate_index("LCR")        # monthly, 1993-2014
+observations <- attach_climate_index(observations, "LCR")
+```
+
+It is the published output of one study rather than an operational product, so
+**cite it when you use it**:
+
+> Jutras M, Dufour CO, Mucci A, Talbot LC (2023) Large-scale control of the
+> retroflection of the Labrador Current. *Nature Communications* **14**:2623.
+> <https://doi.org/10.1038/s41467-023-38321-y>
+
+`index_dictionary()` prints that citation, and
+`as.data.frame(index_dictionary())$reference` carries it at runtime.
+
+Two limits worth knowing. It **covers 1993–2014 only**, so it cannot be attached
+to recent observations. And the series is fetched from the paper's published
+source data rather than recomputed — the authors derived it by tracking virtual
+particles through GLORYS12V1 with
+[OceanParcels](https://oceanparcels.org/), which is a Lagrangian modelling
+exercise rather than something this package reproduces. Extending the record
+past 2014 means redoing that computation, not calling a different function.
+
 One interaction to know: `attach_climate_index()` joins on year and month, and
 `upscale_time(to = "year")` stamps its output `MONTH = 1`. Attaching an index to
 annual data would therefore give every year January's value. Aggregate the index
diff --git a/README.md b/README.md
index df22606..4658457 100644
--- a/README.md
+++ b/README.md
@@ -508,7 +508,7 @@ so every observation in a month receives the same number:
 ``` r
 observations <- attach_climate_index(observations, c("NAO", "AMO"))
 
-index_dictionary()          # NAO, AO, AMO, PDO
+index_dictionary()          # NAO, AO, AMO, PDO, LCR
 ```
 
 That makes them a different kind of covariate from local temperature. They carry
@@ -517,6 +517,39 @@ within a region conditions are better. Useful for interannual questions ("was
 this a warm-regime year?"), useless for spatial ones: a model given only indices
 cannot produce a map.
 
+#### The Labrador Current retroflection index
+
+`LCR` is the odd one out and often the most directly useful of the five. The
+other four are atmospheric patterns; this one describes a *current* — how much
+of the Labrador Current turns eastward at the Grand Banks instead of continuing
+southwest along the shelf. Positive values mean stronger retroflection, so less
+cold, fresh, oxygen-rich Labrador water reaching the Scotian Shelf and Gulf of
+Maine. For shelf water properties that is a shorter causal chain than the NAO.
+
+
+``` r
+lcr <- fetch_climate_index("LCR")        # monthly, 1993-2014
+observations <- attach_climate_index(observations, "LCR")
+```
+
+It is the published output of one study rather than an operational product, so
+**cite it when you use it**:
+
+> Jutras M, Dufour CO, Mucci A, Talbot LC (2023) Large-scale control of the
+> retroflection of the Labrador Current. *Nature Communications* **14**:2623.
+> <https://doi.org/10.1038/s41467-023-38321-y>
+
+`index_dictionary()` prints that citation, and
+`as.data.frame(index_dictionary())$reference` carries it at runtime.
+
+Two limits worth knowing. It **covers 1993–2014 only**, so it cannot be attached
+to recent observations. And the series is fetched from the paper's published
+source data rather than recomputed — the authors derived it by tracking virtual
+particles through GLORYS12V1 with
+[OceanParcels](https://oceanparcels.org/), which is a Lagrangian modelling
+exercise rather than something this package reproduces. Extending the record
+past 2014 means redoing that computation, not calling a different function.
+
 One interaction to know: `attach_climate_index()` joins on year and month, and
 `upscale_time(to = "year")` stamps its output `MONTH = 1`. Attaching an index to
 annual data would therefore give every year January's value. Aggregate the index
diff --git a/man/fetch_climate_index.Rd b/man/fetch_climate_index.Rd
index 44e3b5f..c3e7a0a 100644
--- a/man/fetch_climate_index.Rd
+++ b/man/fetch_climate_index.Rd
@@ -24,6 +24,16 @@ series.
 The published files use fixed-width year-by-month tables with provider-specific
 missing-value codes, which are parsed here into one row per month.
 }
+\section{Citing an index}{
+
+\code{LCR} is the published output of a specific study rather than an operational
+product, and should be cited when used:Jutras M, Dufour CO, Mucci A, Talbot LC (2023) Large-scale control of theretroflection of the Labrador Current. Nature Communications 14:2623.\doi{10.1038/s41467-023-38321-y}
+
+The series is the source data published with that paper's Figure 3, fetched
+from the journal rather than recomputed, so the values are the authors' own.
+\code{as.data.frame(index_dictionary())$reference} carries this at runtime.
+}
+
 \examples{
 \dontrun{
 nao <- fetch_climate_index("NAO", years = 2000:2020)
diff --git a/man/parse_decimal_year_csv.Rd b/man/parse_decimal_year_csv.Rd
new file mode 100644
index 0000000..0e8bd96
--- /dev/null
+++ b/man/parse_decimal_year_csv.Rd
@@ -0,0 +1,41 @@
+% Generated by roxygen2: do not edit by hand
+% Please edit documentation in R/climate_indices.R
+\name{parse_decimal_year_csv}
+\alias{parse_decimal_year_csv}
+\title{Parse a daily index published as decimal years}
+\usage{
+parse_decimal_year_csv(lines, name)
+}
+\arguments{
+\item{lines}{the downloaded file's lines}
+
+\item{name}{what to call the value column}
+}
+\value{
+a data frame with \code{YEAR}, \code{MONTH}, and the named value column
+}
+\description{
+The format the Jutras et al. retroflection index is published in: a couple of
+figure-caption lines, a \verb{Date,<label>} header, then one row per day with the
+date as a decimal year.
+}
+\section{Two conversions worth stating}{
+
+\strong{Decimal years use a fixed 365-day year here.} The step between consecutive
+rows is 1/365 to eleven significant figures, so leap days are not represented
+and a date is recovered as day \code{round(fraction * 365)} of its year. Assuming
+365.25 instead would walk the derived dates off by up to several days over the
+record.
+
+\strong{Daily values are averaged to monthly.} Every other index in this package is
+monthly, and \code{\link[=attach_climate_index]{attach_climate_index()}} joins on year and month, so a daily
+series has nothing to join to. The underlying index is already smoothed with a
+12-month rolling mean, so monthly averaging discards very little.
+
+Months backed by fewer than half their days are dropped rather than reported.
+The record's first and last months are partial by construction, and the fixed
+365-day year rolls the final record into a January of its own — which would
+otherwise surface as a confident-looking monthly value resting on a single day.
+}
+
+\keyword{internal}
diff --git a/tests/testthat/test-climate-indices.R b/tests/testthat/test-climate-indices.R
index 690d740..8688280 100644
--- a/tests/testthat/test-climate-indices.R
+++ b/tests/testthat/test-climate-indices.R
@@ -19,6 +19,70 @@ psl_lines <- c(
   "  Calculated at NOAA PSL1"
 )
 
+# The retroflection index, in the shape Nature publishes it: two caption lines, a
+# header, then daily rows dated as decimal years on a fixed 365-day year.
+decimal_year_lines <- function(start = 1993, months = 1:12, days_in_last = 31) {
+  rows <- unlist(lapply(months, function(m) {
+    first <- as.Date(paste(start, m, 1, sep = "-"))
+    n <- if (m == max(months)) days_in_last else
+      as.integer(lubridate::days_in_month(first))
+    doy <- as.integer(format(first, "%j")) + seq_len(n) - 1
+    sprintf("%.11f,%.4f", start + (doy - 1) / 365, m / 100)
+  }))
+  c("Figure 3,", ",", "Date,Retroflection index", rows)
+}
+
+test_that("a decimal-year CSV parses to monthly means", {
+  series <- parse_index_table(decimal_year_lines(), format = "decimal_year_csv",
+                              name = "LCR")
+
+  expect_setequal(names(series), c("YEAR", "MONTH", "LCR"))
+  expect_equal(nrow(series), 12)
+  # Each month's daily values are all month/100, so the monthly mean is too.
+  expect_equal(series$LCR, (1:12) / 100)
+})
+
+test_that("a month backed by too few days is dropped, not reported", {
+  # December carries a single day here. Averaging it would produce a confident
+  # December value resting on one observation, which is the failure mode
+  # min_coverage guards against elsewhere in the package.
+  series <- parse_index_table(decimal_year_lines(days_in_last = 1),
+                              format = "decimal_year_csv", name = "LCR")
+
+  expect_equal(nrow(series), 11)
+  expect_false(12 %in% series$MONTH)
+
+  # Half a month is enough to keep.
+  half <- parse_index_table(decimal_year_lines(days_in_last = 16),
+                            format = "decimal_year_csv", name = "LCR")
+  expect_true(12 %in% half$MONTH)
+})
+
+test_that("decimal years are read on the 365-day convention the file uses", {
+  # The published file steps by exactly 1/365 between days. Reading it as 365.25
+  # would walk the derived dates off by several days across the record, moving
+  # values into neighbouring months.
+  series <- parse_index_table(decimal_year_lines(), format = "decimal_year_csv",
+                              name = "LCR")
+
+  expect_setequal(series$MONTH, 1:12)
+  expect_true(all(series$YEAR == 1993))
+})
+
+test_that("the retroflection index carries its citation", {
+  # It is the published output of one study rather than an operational product,
+  # so the reference has to travel with it.
+  entry <- climate_indices()$LCR
+
+  expect_match(entry$reference, "Jutras")
+  expect_match(entry$reference, "10\\.1038/s41467-023-38321-y")
+
+  dictionary <- as.data.frame(index_dictionary())
+  expect_match(dictionary$reference[dictionary$name == "LCR"], "Nature Communications")
+  # The operational indices have no single paper, and must not have one invented.
+  expect_true(all(is.na(dictionary$reference[dictionary$name %in% c("NAO", "AO")])))
+})
+
 test_that("a CPC table parses to one row per month", {
   series <- parse_index_table(cpc_lines, format = "cpc_table", name = "NAO")