diff --git a/.Rbuildignore b/.Rbuildignore
index 46c7c80..76855d9 100644
--- a/.Rbuildignore
+++ b/.Rbuildignore
@@ -1,3 +1,4 @@
+^data-raw$
 ^.*\.Rproj$
 ^\.Rproj\.user$
 ^_pkgdown\.yml$
diff --git a/R/adjust.R b/R/adjust.R
new file mode 100644
index 0000000..b5f0853
--- /dev/null
+++ b/R/adjust.R
@@ -0,0 +1,45 @@
+# R/adjust.R
+# Inflation adjustment helpers using bundled CPIAUCSL annual averages.
+# The `cpi_annual` tibble (year, cpi) is stored as internal data in
+# R/sysdata.rda and built via data-raw/ on package update.
+
+#' Return the bundled annual CPI table.
+#'
+#' @return Tibble with columns `year` (integer) and `cpi` (numeric, CPIAUCSL
+#'   annual average, 1982-84 = 100).
+#' @noRd
+.cpi_table <- function() {
+  cpi_annual
+}
+
+#' Inflate (or deflate) an amount vector between two years.
+#'
+#' Converts nominal amounts in `from_year` dollars to real amounts in
+#' `to_year` dollars using the bundled CPIAUCSL annual average index.
+#' Multiplies by `cpi[to_year] / cpi[from_year]`.
+#'
+#' @param amt Numeric vector of nominal amounts.
+#' @param from_year Integer or integer-like vector of source years (one per
+#'   element of `amt`, or length 1).
+#' @param to_year Integer target year (scalar).
+#' @return Numeric vector of real amounts, same length as `amt`.
+#' @noRd
+.inflate <- function(amt, from_year, to_year) {
+  cpi <- .cpi_table()
+  from_year <- as.integer(from_year)
+  to_year   <- as.integer(to_year)
+  if (length(to_year) != 1L) {
+    cli::cli_abort("`to_year` must be a scalar.")
+  }
+  if (!(to_year %in% cpi$year)) {
+    cli::cli_abort("CPI unavailable for to_year = {to_year}. Supported: {min(cpi$year)}-{max(cpi$year)}.")
+  }
+  missing_years <- setdiff(from_year[!is.na(from_year)], cpi$year)
+  if (length(missing_years) > 0) {
+    cli::cli_abort("CPI unavailable for from_year value(s): {missing_years}.")
+  }
+  lookup <- stats::setNames(cpi$cpi, as.character(cpi$year))
+  cpi_from <- unname(lookup[as.character(from_year)])
+  cpi_to   <- unname(lookup[as.character(to_year)])
+  amt * cpi_to / cpi_from
+}
diff --git a/R/sysdata.rda b/R/sysdata.rda
new file mode 100644
index 0000000..8865593
Binary files /dev/null and b/R/sysdata.rda differ
diff --git a/data-raw/cpi_annual.R b/data-raw/cpi_annual.R
new file mode 100644
index 0000000..a4aa912
--- /dev/null
+++ b/data-raw/cpi_annual.R
@@ -0,0 +1,34 @@
+# data-raw/cpi_annual.R
+#
+# Refresh the bundled CPIAUCSL annual-average table from FRED.
+# Run interactively when the CPI series needs to be extended forward:
+#   Rscript data-raw/cpi_annual.R
+#
+# Source: FRED series CPIAUCSL (Consumer Price Index for All Urban Consumers,
+# All Items, 1982-84 = 100), monthly. Annual mean computed here. The bundled
+# artifact is R/sysdata.rda (loaded automatically by the package).
+
+stopifnot(requireNamespace("utils", quietly = TRUE),
+          requireNamespace("tibble", quietly = TRUE),
+          requireNamespace("usethis", quietly = TRUE))
+
+fred_url <- "https://fred.stlouisfed.org/graph/fredgraph.csv?id=CPIAUCSL"
+raw <- utils::read.csv(url(fred_url), stringsAsFactors = FALSE)
+date_col <- intersect(c("observation_date", "DATE", "date"), names(raw))[1]
+stopifnot(length(date_col) == 1L, !is.na(date_col))
+raw$year <- as.integer(format(as.Date(raw[[date_col]]), "%Y"))
+raw$CPIAUCSL <- suppressWarnings(as.numeric(raw$CPIAUCSL))
+raw <- raw[!is.na(raw$CPIAUCSL), ]
+
+annual <- stats::aggregate(
+  raw$CPIAUCSL, by = list(year = raw$year), FUN = mean, na.rm = TRUE
+)
+names(annual)[2] <- "cpi"
+
+cpi_annual <- tibble::as_tibble(annual)
+cpi_annual$cpi <- round(cpi_annual$cpi, 4)
+
+message(sprintf("CPI range: %d-%d (%d years)",
+                min(cpi_annual$year), max(cpi_annual$year), nrow(cpi_annual)))
+
+usethis::use_data(cpi_annual, internal = TRUE, overwrite = TRUE)
diff --git a/tests/testthat/test-adjust.R b/tests/testthat/test-adjust.R
new file mode 100644
index 0000000..b10e3e6
--- /dev/null
+++ b/tests/testthat/test-adjust.R
@@ -0,0 +1,43 @@
+test_that("cpi_annual internal data is available with expected shape", {
+  cpi <- uscogdata:::.cpi_table()
+  expect_true(is.data.frame(cpi))
+  expect_named(cpi, c("year", "cpi"))
+  expect_true(nrow(cpi) > 70L)
+  expect_true(all(c(2000L, 2010L, 2021L) %in% cpi$year))
+})
+
+test_that(".inflate matches a known anchor: 2000 in 2021$ ≈ 1.57x nominal", {
+  # FRED CPIAUCSL annual: 2000=172.19, 2021=270.97 → ratio ≈ 1.574
+  result <- uscogdata:::.inflate(100, from_year = 2000, to_year = 2021)
+  expect_equal(result, 157.4, tolerance = 0.5)
+})
+
+test_that(".inflate is vectorized over from_year", {
+  result <- uscogdata:::.inflate(
+    amt       = c(100, 100, 100),
+    from_year = c(2000, 2010, 2021),
+    to_year   = 2021
+  )
+  expect_length(result, 3L)
+  expect_equal(result[3], 100, tolerance = 1e-6)  # same year → identity
+  expect_gt(result[1], 150)  # 2000 inflated to 2021 ≈ 157
+  expect_lt(result[2], 130)  # 2010 inflated to 2021 ≈ 124
+  expect_gt(result[2], 115)
+})
+
+test_that(".inflate errors on unknown from_year or to_year", {
+  expect_error(
+    uscogdata:::.inflate(100, from_year = 1900, to_year = 2021),
+    "CPI unavailable for from_year"
+  )
+  expect_error(
+    uscogdata:::.inflate(100, from_year = 2000, to_year = 2200),
+    "CPI unavailable for to_year"
+  )
+})
+
+test_that(".inflate preserves NA amounts", {
+  result <- uscogdata:::.inflate(c(100, NA, 200), from_year = 2000, to_year = 2021)
+  expect_true(is.na(result[2]))
+  expect_false(any(is.na(result[c(1, 3)])))
+})