diff --git a/tests/testthat/test-true-parameter-linking-rmse.R b/tests/testthat/test-true-parameter-linking-rmse.R
new file mode 100644
index 00000000..20c052dc
--- /dev/null
+++ b/tests/testthat/test-true-parameter-linking-rmse.R
@@ -0,0 +1,144 @@
+# True-parameter linking RMSE / recovery (buyer-visible quality gate).
+#
+# Existing fixtures pin the Kim (2006) FIPC *contract* (anchors copied and
+# held fixed). They do not report recovery error of free new-form items
+# against the generating parameters. This file adds that metric.
+#
+# Scale: simdata() defaults to N(0, 1) theta. autoFIPC() is called with
+# forceNormalZeroOne = TRUE and freeMEAN = FALSE so the linked unique items
+# stay on that same metric. Anchors are fixed to *estimated* old-form
+# values, so unique-item RMSE versus truth includes ordinary calibration
+# error plus the small scale discrepancy of those estimated anchors.
+#
+# This package still orchestrates FIPC in R and delegates estimation to
+# mirt (Chalmers, 2012; Bock & Aitkin, 1981). There is no Rust or GPU
+# numeric core; this test is the recovery gate, not a new estimator.
+#
+# References (APA 7th):
+#   Bock, R. D., & Aitkin, M. (1981). Marginal maximum likelihood
+#     estimation of item parameters: Application of an EM algorithm.
+#     Psychometrika, 46(4), 443-459. https://doi.org/10.1007/BF02293801
+#   Chalmers, R. P. (2012). mirt: A multidimensional item response theory
+#     package for the R environment. Journal of Statistical Software,
+#     48(6), 1-29. https://doi.org/10.18637/jss.v048.i06
+#   Kim, S. (2006). A comparative study of IRT fixed parameter calibration
+#     methods. Journal of Educational Measurement, 43(4), 355-381.
+#     https://doi.org/10.1111/j.1745-3984.2006.00021.x
+
+rmse <- function(est, tru) {
+  sqrt(mean((as.numeric(est) - as.numeric(tru))^2))
+}
+
+item_ad_values <- function(vals, items) {
+  out <- numeric(0)
+  for (it in items) {
+    row <- vals[vals$item == it & vals$name %in% c("a1", "d"), c("name", "value")]
+    row <- row[match(c("a1", "d"), row$name), ]
+    out <- c(out, row$value)
+  }
+  out
+}
+
+item_ad_truth <- function(a, d, idx) {
+  as.numeric(rbind(a[idx, 1], d[idx]))
+}
+
+test_that("FIPC recovers generating parameters at bounded RMSE", {
+  skip_on_cran()
+  skip_if_not_installed("mirt")
+
+  set.seed(20260817)
+  old_item_names <- paste0("old_", 1:8)
+  new_item_names <- paste0("new_", 1:8)
+  old_common_items <- old_item_names[1:5]
+  new_common_items <- new_item_names[1:5]
+  unique_idx <- 6:8
+
+  common_a <- c(0.90, 1.15, 1.30, 0.80, 1.05)
+  common_d <- c(-1.00, -0.35, 0.20, 0.80, -0.50)
+  old_a <- matrix(c(common_a, 0.85, 1.20, 0.70), ncol = 1)
+  old_d <- c(common_d, 0.40, -0.65, 0.25)
+  new_a <- matrix(c(common_a, 1.10, 0.85, 1.25), ncol = 1)
+  new_d <- c(common_d, -0.15, 0.55, -0.40)
+
+  old_data <- as.data.frame(mirt::simdata(
+    a = old_a,
+    d = old_d,
+    itemtype = rep("2PL", length(old_item_names)),
+    N = 2000
+  ))
+  new_data <- as.data.frame(mirt::simdata(
+    a = new_a,
+    d = new_d,
+    itemtype = rep("2PL", length(new_item_names)),
+    N = 2000
+  ))
+  names(old_data) <- old_item_names
+  names(new_data) <- new_item_names
+
+  old_model <- mirt::mirt(
+    old_data,
+    1,
+    itemtype = "2PL",
+    method = "EM",
+    verbose = FALSE,
+    technical = list(NCYCLES = 500)
+  )
+  new_model <- mirt::mirt(
+    new_data,
+    1,
+    itemtype = "2PL",
+    method = "EM",
+    verbose = FALSE,
+    technical = list(NCYCLES = 500)
+  )
+
+  linked <- aFIPC::autoFIPC(
+    newformXData = new_model,
+    oldformYData = old_model,
+    newformCommonItemNames = new_common_items,
+    oldformCommonItemNames = old_common_items,
+    itemtype = "2PL",
+    checkIPD = FALSE,
+    tryEM = TRUE,
+    freeMEAN = FALSE,
+    forceNormalZeroOne = TRUE,
+    confirmCommonItems = TRUE
+  )
+
+  old_values <- mirt::mod2values(old_model)
+  linked_values <- mirt::mod2values(linked$LinkedModel)
+
+  old_anchor_est <- item_ad_values(old_values, old_common_items)
+  linked_anchor_est <- item_ad_values(linked_values, new_common_items)
+  anchor_copy_rmse <- rmse(linked_anchor_est, old_anchor_est)
+
+  old_recovery_rmse <- rmse(
+    old_anchor_est,
+    item_ad_truth(old_a, old_d, seq_along(old_common_items))
+  )
+
+  unique_linked_est <- item_ad_values(linked_values, new_item_names[unique_idx])
+  unique_linked_rmse <- rmse(
+    unique_linked_est,
+    item_ad_truth(new_a, new_d, unique_idx)
+  )
+
+  # testthat 3 expect_lt() has no info=; keep every gate on expect_true()
+  # so CI logs always print the three RMSE numbers.
+  metrics <- sprintf(
+    "anchor_copy_rmse=%.6f old_recovery_rmse=%.4f unique_linked_rmse=%.4f",
+    anchor_copy_rmse,
+    old_recovery_rmse,
+    unique_linked_rmse
+  )
+  expect_true(
+    is.finite(anchor_copy_rmse) &&
+      is.finite(old_recovery_rmse) &&
+      is.finite(unique_linked_rmse),
+    info = metrics
+  )
+  expect_true(anchor_copy_rmse < 1e-6, info = metrics)
+  expect_true(old_recovery_rmse < 0.40, info = metrics)
+  expect_true(unique_linked_rmse < 0.50, info = metrics)
+})