From ad9d0e4e4cfd7a46edda8ff920c4ec83e9dda5f0 Mon Sep 17 00:00:00 2001
From: Carson Richardson <carson.richardson@outlook.com>
Date: Sat, 5 Sep 2026 13:56:43 -0400
Subject: [PATCH 1/2] Misc doc changes

---
 NEWS.md          | 86 ++++++++++++++++++++++++++++++++----------------
 cran-comments.md |  2 +-
 revdep/README.md | 70 +++++++++++++++++++--------------------
 3 files changed, 93 insertions(+), 65 deletions(-)

diff --git a/NEWS.md b/NEWS.md
index db88d49..35f88ac 100644
--- a/NEWS.md
+++ b/NEWS.md
@@ -14,7 +14,7 @@
 - Added explicit `subgroup` mappings for mixed hierarchical displays, allowing
   standalone covariates and contiguous categorical blocks to share a forest
   plot. Subgroup p-values can be delegated as "overall" or "level" via `p_method`
-  in `tidy_forest_model()`
+  in `tidy_forest_model()` (#11)
 - Added covariance-aware subgroup effects for fitted interaction models.
   `tidy_forest_model()` and fitted-model `as_forest_data()` methods now use
   `marginaleffects` to derive average slopes or comparisons from the original
@@ -40,10 +40,10 @@
 
 - Added the `forest_data` S3 class as the package's stable interchange format.
   It stores validated effect-scale, axis, confidence-level, reference-value,
-  source-model, source-package, and source-column metadata.
+  source-model, source-package, and source-column metadata. (#17)
 - Converted `as_forest_data()` into an S3 generic with methods for data frames,
   existing `forest_data` objects, linear and generalized linear models, Cox
-  models, and supported mixed-model classes.
+  models, and supported mixed-model classes. (#19)
 - Refactored `ggforestplot()` to consume the `forest_data` contract instead of
   determining plotting behavior from the original model class.
 - Added `forest_metadata()` for inspecting the semantic and provenance
@@ -95,12 +95,12 @@
 
 - Added mixed-model support for `tidy_forest_model()` and `ggforestplot()`
   when `broom.mixed` is installed. Mixed-model tidiers are restricted to fixed
-  effects so random-effect parameter rows are not plotted as terms.
+  effects so random-effect parameter rows are not plotted as terms. (#21)
 - Added confidence interval truncation support to `ggforestplot()` via
   `ci_limits`, with optional arrowheads for intervals extending beyond the
-  displayed range.
+  displayed range. (#18)
 - Added `bind_forest_models()` to stack multiple fitted models into one
-  grouped forest-plot data frame.
+  grouped forest-plot data frame. (#16)
 
 ## Minor Changes
 
@@ -115,59 +115,87 @@
 - Faceted `ggforestplot()` output now respects factor level order in the
   `facet` column, allowing users to control facet ordering directly from their
   data.
-- Fixed faceted `ggforestplot()` output when users add `scale_y_discrete(limits = ...)`, preserving per-facet row filtering while accepting visible term labels.
+- Fixed faceted `ggforestplot()` output when users add 
+  `scale_y_discrete(limits = ...)`, preserving per-facet row filtering while 
+  accepting visible term labels. (#20)
 
 # ggforestplotR 0.2.1
 
 ## Breaking Changes
 
-- Simplified the reference-line API in `ggforestplot()`: `ref_line` now takes the numeric reference value directly, and `NULL` hides the line.
-- Replaced the older reference-line arguments with `ref_label`, `ref_linetype`, and `ref_color`.
+- Simplified the reference-line API in `ggforestplot()`: `ref_line` now takes 
+  the numeric reference value directly, and `NULL` hides the line.
+- Replaced the older reference-line arguments with `ref_label`, `ref_linetype`, 
+  and `ref_color`.
 
 ## Major Changes
 
-- Added `facet` and `facet_strip_position` as clearer names for grouped plot panels. The older `grouping` and `grouping_strip_position` arguments now warn and continue to work.
-- Added `linewidth` for confidence interval widths. The older `line_size` argument now warns and continues to work.
-- Added deprecation warnings for table-display shortcuts in favor of explicit `columns`, `left_columns`, and `right_columns` selections.
+- Added `facet` and `facet_strip_position` as clearer names for grouped plot 
+  panels. The older `grouping` and `grouping_strip_position` arguments now warn 
+  and continue to work.
+- Added `linewidth` for confidence interval widths. The older `line_size` 
+  argument now warns and continues to work.
+- Added deprecation warnings for table-display shortcuts in favor of explicit 
+  `columns`, `left_columns`, and `right_columns` selections.
 
 ## Minor Changes
 
-- Added `stripe_alpha` to control row stripe transparency across plots, side tables, and split tables.
-- Updated table helpers so they inherit custom y-axis ordering and filtering from the trained forest plot scale.
-- Updated vignettes and tests to use the clearer faceting and geometry argument names.
+- Added `stripe_alpha` to control row stripe transparency across plots, 
+  side tables, and split tables. (#7)
+- Updated table helpers so they inherit custom y-axis ordering and filtering 
+  from the trained forest plot scale. (#9)
+- Updated vignettes and tests to use the clearer faceting and geometry argument 
+  names.
 
 # ggforestplotR 0.2.0
 
 ## Major Changes
 
-- Added separate table formatting controls for point estimates, confidence interval bounds, and p-values via `estimate_digits`, `interval_digits`, and `p_digits`.
-- Added `ref_line`, `ref_line_value`, and `ref_line_label` to standardize reference-line naming while keeping `zero_line` arguments backward compatible.
+- Added separate table formatting controls for point estimates, 
+  confidence interval bounds, and p-values via `estimate_digits`, 
+  `interval_digits`, and `p_digits`.
+- Added `ref_line`, `ref_line_value`, and `ref_line_label` to standardize 
+  reference-line naming while keeping `zero_line` arguments backward compatible.
 - Added `term_labels`, a named-vector relabeling helper for displayed model terms.
-- Expanded `add_forest_table(columns = ...)` and split-table column selection to support arbitrary preserved data columns in addition to built-in columns.
-- Added `column_labels`, a named-vector helper for relabeling displayed table column headers.
-- Added a built-in `ci` table column and `ci_fmt` so users can display point estimates and confidence intervals in separate columns. `conf.low` and `conf.high` are accepted as aliases for `ci`.
+- Expanded `add_forest_table(columns = ...)` and split-table column selection to
+  support arbitrary preserved data columns in addition to built-in columns.
+- Added `column_labels`, a named-vector helper for relabeling displayed table 
+  column headers.
+- Added a built-in `ci` table column and `ci_fmt` so users can display point 
+  estimates and confidence intervals in separate columns. `conf.low` and 
+  `conf.high` are accepted as aliases for `ci`.
 
 ## Minor Changes
 
-- Added model-aware default effect labels and scales, including odds ratios for binomial logit models and hazard ratios for Cox models.
+- Added model-aware default effect labels and scales, including odds ratios for 
+  binomial logit models and hazard ratios for Cox models.
 
 # ggforestplotR 0.1.1
 
 ## Minor Changes
 
-- Added `events` column support across `ggforestplot()`, `as_forest_data()`, `add_forest_table()`, and `add_split_table()`. (Thanks @sritchie73 for the suggestion)
-- Added table header styling controls for font size, font face, and font family in both table helpers. (Thanks @sritchie73 for the suggestion)
-- Split table composition code into dedicated source files to simplify maintenance of the public API.
-- Refined split-table examples and documentation to match the expanded numeric column mapping.
+- Added `events` column support across `ggforestplot()`, `as_forest_data()`, 
+  `add_forest_table()`, and `add_split_table()`. (#2) (#3)
+  (Thanks @sritchie73 for the suggestion)
+- Added table header styling controls for font size, font face, and font family 
+  in both table helpers. (Thanks @sritchie73 for the suggestion)
+- Split table composition code into dedicated source files to simplify 
+  maintenance of the public API.
+- Refined split-table examples and documentation to match the expanded numeric 
+  column mapping.
 
 # ggforestplotR 0.1.0
 
 ## Major Changes
 
-- Refined split-table layout logic, including dynamic panel widths from column counts, alignment controls, and cleaner stripe/background handling.
-- `add_forest_table()` now allows for custom column ordering via the `columns` argument.
-- Converted package documentation to roxygen-generated `NAMESPACE` and `.Rd` files.
-- Reworked package documentation, including pkgdown configuration, README installation guidance, and the vignette set.
+- Refined split-table layout logic, including dynamic panel widths from column 
+  counts, alignment controls, and cleaner stripe/background handling.
+- `add_forest_table()` now allows for custom column ordering via the `columns`
+  argument.
+- Converted package documentation to roxygen-generated `NAMESPACE` and `.Rd` 
+  files.
+- Reworked package documentation, including pkgdown configuration, README 
+  installation guidance, and the vignette set.
 
 # ggforestplotR 0.0.0.9000
 
diff --git a/cran-comments.md b/cran-comments.md
index eaf65e5..023b49b 100644
--- a/cran-comments.md
+++ b/cran-comments.md
@@ -4,7 +4,7 @@
 
 ## R CMD check results
 
-- `R CMD check ggforestplotR_0.3.1.tar.gz --as-cran`
+- `R CMD check ggforestplotR_0.5.0.tar.gz --as-cran`
 - Status: OK
 0 errors | 0 warnings | 0 notes
 
diff --git a/revdep/README.md b/revdep/README.md
index e212009..61a88a6 100644
--- a/revdep/README.md
+++ b/revdep/README.md
@@ -1,43 +1,43 @@
 # Platform
 
-|field    |value                                                                                                                                                   |
-|:--------|:-------------------------------------------------------------------------------------------------------------------------------------------------------|
-|version  |R version 4.6.0 (2026-04-24 ucrt)                                                                                                                       |
-|os       |Windows 11 x64 (build 26200)                                                                                                                            |
-|system   |x86_64, mingw32                                                                                                                                         |
-|ui       |RStudio                                                                                                                                                 |
-|language |(EN)                                                                                                                                                    |
-|collate  |English_United States.utf8                                                                                                                              |
-|ctype    |English_United States.utf8                                                                                                                              |
-|tz       |America/New_York                                                                                                                                        |
-|date     |2026-05-19                                                                                                                                              |
-|rstudio  |2026.04.0+526 Globemaster Allium (desktop)                                                                                                              |
-|pandoc   |3.9 @ D:\pandoc\pandoc-3.9\pandoc.exe                                                                                                                   |
-|quarto   |ERROR: Unknown command "TMPDIR=C:/Users/Carso/AppData/Local/Temp/RtmpoFpwMI/file19fc4e91f46". Did you mean command "remove"? @ D:\Quarto\bin\quarto.exe |
+|field    |value                                      |
+|:--------|:------------------------------------------|
+|version  |R version 4.6.1 (2026-06-24 ucrt)          |
+|os       |Windows 11 x64 (build 26200)               |
+|system   |x86_64, mingw32                            |
+|ui       |RStudio                                    |
+|language |(EN)                                       |
+|collate  |English_United States.utf8                 |
+|ctype    |English_United States.utf8                 |
+|tz       |America/New_York                           |
+|date     |2026-09-02                                 |
+|rstudio  |2026.04.0+526 Globemaster Allium (desktop) |
+|pandoc   |3.9 @ D:\pandoc\pandoc-3.9\pandoc.exe      |
+|quarto   |1.9.35 @ D:\Quarto\bin\quarto.exe          |
 
 # Dependencies
 
-|package       |old   |new   |Δ  |
-|:-------------|:-----|:-----|:--|
-|ggforestplotR |0.1.1 |0.2.0 |*  |
-|cli           |3.6.6 |3.6.6 |   |
-|cpp11         |0.5.5 |0.5.5 |   |
-|farver        |2.1.2 |2.1.2 |   |
-|ggplot2       |4.0.3 |4.0.3 |   |
-|glue          |1.8.1 |1.8.1 |   |
-|gtable        |0.3.6 |0.3.6 |   |
-|isoband       |0.3.0 |0.3.0 |   |
-|labeling      |0.4.3 |0.4.3 |   |
-|lifecycle     |1.0.5 |1.0.5 |   |
-|patchwork     |1.3.2 |1.3.2 |   |
-|R6            |2.6.1 |2.6.1 |   |
-|RColorBrewer  |1.1-3 |1.1-3 |   |
-|rlang         |1.2.0 |1.2.0 |   |
-|S7            |0.2.2 |0.2.2 |   |
-|scales        |1.4.0 |1.4.0 |   |
-|vctrs         |0.7.3 |0.7.3 |   |
-|viridisLite   |0.4.3 |0.4.3 |   |
-|withr         |3.0.2 |3.0.2 |   |
+|package       |old   |new |Δ  |
+|:-------------|:-----|:---|:--|
+|ggforestplotR |0.3.1 |NA  |*  |
+|cli           |3.6.6 |NA  |*  |
+|cpp11         |0.5.5 |NA  |*  |
+|farver        |2.1.2 |NA  |*  |
+|ggplot2       |4.0.3 |NA  |*  |
+|glue          |1.8.1 |NA  |*  |
+|gtable        |0.3.6 |NA  |*  |
+|isoband       |0.3.0 |NA  |*  |
+|labeling      |0.4.3 |NA  |*  |
+|lifecycle     |1.0.5 |NA  |*  |
+|patchwork     |1.3.2 |NA  |*  |
+|R6            |2.6.1 |NA  |*  |
+|RColorBrewer  |1.1-3 |NA  |*  |
+|rlang         |1.3.0 |NA  |*  |
+|S7            |0.2.2 |NA  |*  |
+|scales        |1.4.0 |NA  |*  |
+|vctrs         |0.7.3 |NA  |*  |
+|viridisLite   |0.4.3 |NA  |*  |
+|withr         |3.0.3 |NA  |*  |
 
 # Revdeps
 

From b2072fcfd88685e1059c1a2679e419c8d8244b6f Mon Sep 17 00:00:00 2001
From: Carson Richardson <carson.richardson@outlook.com>
Date: Sun, 27 Sep 2026 11:07:33 -0400
Subject: [PATCH 2/2] Implement structured forest rows and table formatters

---
 DESCRIPTION                                   |   4 +-
 NEWS.md                                       |  10 +
 R/add_forest_table.R                          |  17 ++
 R/add_split_table.R                           |  18 ++
 R/as_forest_data.R                            |  17 ++
 R/forest_data.R                               |  18 +-
 R/ggforestplot.R                              |  31 ++-
 R/ggplot_add_scales.R                         |   2 +-
 R/row_types.R                                 |  24 ++
 R/table_formatters.R                          |  62 ++++++
 R/utils.R                                     | 205 +++++++++++++-----
 man/add_forest_table.Rd                       |  11 +
 man/add_split_table.Rd                        |  10 +
 man/as_forest_data.Rd                         |   7 +
 man/ggforestplot.Rd                           |   8 +-
 tests/testthat/helper-structured-rows.R       |  27 +++
 tests/testthat/test-as_forest_data.R          |   5 +-
 tests/testthat/test-row-types.R               | 116 ++++++++++
 tests/testthat/test-subgroup-effects.R        |  10 +-
 tests/testthat/test-subgroup.R                |  26 +--
 tests/testthat/test-table-formatters.R        | 106 +++++++++
 vignettes/ggforestplotR-data-helpers.Rmd      |  34 +++
 .../ggforestplotR-plot-customization.Rmd      |  35 +++
 23 files changed, 707 insertions(+), 96 deletions(-)
 create mode 100644 R/row_types.R
 create mode 100644 R/table_formatters.R
 create mode 100644 tests/testthat/helper-structured-rows.R
 create mode 100644 tests/testthat/test-row-types.R
 create mode 100644 tests/testthat/test-table-formatters.R

diff --git a/DESCRIPTION b/DESCRIPTION
index 2a50bbb..8e1de6d 100644
--- a/DESCRIPTION
+++ b/DESCRIPTION
@@ -1,6 +1,6 @@
 Package: ggforestplotR
 Title: Publication-Ready Forest Plots with 'ggplot2'
-Version: 0.5.0
+Version: 0.5.0.9000
 Authors@R:
     person(given = "Carson",
            family = "Richardson",
@@ -35,4 +35,4 @@ Config/testthat/edition: 3
 Config/Needs/website: pkgdown
 URL: https://thatoneguy006.github.io/ggforestplotR/, https://github.com/thatoneguy006/ggforestplotR
 BugReports: https://github.com/thatoneguy006/ggforestplotR/issues
-Config/roxygen2/version: 8.0.0
+Config/roxygen2/version: 8.1.0
diff --git a/NEWS.md b/NEWS.md
index 5af9112..aa7b4c0 100644
--- a/NEWS.md
+++ b/NEWS.md
@@ -1,3 +1,13 @@
+# ggforestplotR (development version)
+
+- Added semantic forest row types (`estimate`, `reference`, `header`, `summary`,
+  and `spacer`) so tables can contain reference categories and structural rows
+  without fabricated plot geometry.
+- Added vectorized `formatters` to both table helpers for estimates, confidence
+  bounds, p-values, and preserved source columns. Digit settings and text
+  templates remain available.
+- Added `reference_text` to both table helpers to customize reference cells.
+
 # ggforestplotR 0.5.0
 
 ## Breaking Changes
diff --git a/R/add_forest_table.R b/R/add_forest_table.R
index d311368..6427cad 100644
--- a/R/add_forest_table.R
+++ b/R/add_forest_table.R
@@ -13,6 +13,8 @@
                                   p_digits = NULL,
                                   estimate_fmt = NULL,
                                   ci_fmt = NULL,
+                                  formatters = NULL,
+                                  reference_text = "Reference",
                                   text_size = NULL,
                                   header_text_size = NULL,
                                   header_fontface = "bold",
@@ -140,6 +142,8 @@
     p_digits = digits$p_digits,
     estimate_fmt = estimate_fmt,
     ci_fmt = ci_fmt,
+    formatters = formatters,
+    reference_text = reference_text,
     columns = table_columns,
     display_data = display_data
   )
@@ -251,6 +255,13 @@
 #'   `columns` includes `"ci"`. Use `{conf.low}` and `{conf.high}` as
 #'   placeholders. The shorthand `{conf.low, conf.high}` is also supported.
 #'   Defaults to `"({conf.low}, {conf.high})"`.
+#' @param formatters Named list of vectorized functions for scalar table
+#'   values. Keys include `estimate`, `ci`, `conf.low`, `conf.high`, `p`
+#'   (or `p.value`), `n`, `events`, `group`, and preserved source columns.
+#'   Each function receives a vector and returns one value per input.
+#'   Functions take precedence over digit settings; `estimate_fmt` and
+#'   `ci_fmt` compose the resulting strings.
+#' @param reference_text Text shown in estimate cells for reference rows.
 #' @param text_size Text size for table contents. Defaults to `3.2`.
 #' @param header_text_size Header text size for table column labels. Defaults
 #'   to `11`.
@@ -317,6 +328,8 @@ add_forest_table <- function(plot = NULL,
                              p_digits = NULL,
                              estimate_fmt = NULL,
                              ci_fmt = NULL,
+                             formatters = NULL,
+                             reference_text = "Reference",
                              text_size = NULL,
                              header_text_size = NULL,
                              header_fontface = "bold",
@@ -361,6 +374,8 @@ add_forest_table <- function(plot = NULL,
         p_digits = p_digits,
         estimate_fmt = estimate_fmt,
         ci_fmt = ci_fmt,
+        formatters = formatters,
+        reference_text = reference_text,
         text_size = text_size,
         header_text_size = header_text_size,
         header_fontface = header_fontface,
@@ -396,6 +411,8 @@ add_forest_table <- function(plot = NULL,
     p_digits = p_digits,
     estimate_fmt = estimate_fmt,
     ci_fmt = ci_fmt,
+    formatters = formatters,
+    reference_text = reference_text,
     text_size = text_size,
     header_text_size = header_text_size,
     header_fontface = header_fontface,
diff --git a/R/add_split_table.R b/R/add_split_table.R
index 4e1b61b..5a73367 100644
--- a/R/add_split_table.R
+++ b/R/add_split_table.R
@@ -13,6 +13,8 @@
                                  p_digits = NULL,
                                  estimate_fmt = NULL,
                                  ci_fmt = NULL,
+                                 formatters = NULL,
+                                 reference_text = "Reference",
                                  text_size = NULL,
                                  header_text_size = NULL,
                                  header_fontface = "bold",
@@ -144,6 +146,8 @@
     p_digits = digits$p_digits,
     estimate_fmt = estimate_fmt,
     ci_fmt = ci_fmt,
+    formatters = formatters,
+    reference_text = reference_text,
     columns = resolved_left,
     display_data = display_data
   )
@@ -161,6 +165,8 @@
     p_digits = digits$p_digits,
     estimate_fmt = estimate_fmt,
     ci_fmt = ci_fmt,
+    formatters = formatters,
+    reference_text = reference_text,
     columns = resolved_right,
     display_data = display_data
   )
@@ -329,6 +335,12 @@
 #'   table columns include `"ci"`. Use `{conf.low}` and `{conf.high}` as
 #'   placeholders. The shorthand `{conf.low, conf.high}` is also supported.
 #'   Defaults to `"({conf.low}, {conf.high})"`.
+#' @param formatters Named list of vectorized functions for scalar table
+#'   values. Keys include `estimate`, `ci`, `conf.low`, `conf.high`, `p`
+#'   (or `p.value`), `n`, `events`, `group`, and preserved source columns.
+#'   Functions take precedence over digit settings; `estimate_fmt` and
+#'   `ci_fmt` compose the resulting strings.
+#' @param reference_text Text shown in estimate cells for reference rows.
 #' @param text_size Text size for table contents. Defaults to `3.2`.
 #' @param header_text_size Header text size for table column labels. Defaults
 #'   to `11`.
@@ -396,6 +408,8 @@ add_split_table <- function(plot = NULL,
                             p_digits = NULL,
                             estimate_fmt = NULL,
                             ci_fmt = NULL,
+                            formatters = NULL,
+                            reference_text = "Reference",
                             text_size = NULL,
                             header_text_size = NULL,
                             header_fontface = "bold",
@@ -435,6 +449,8 @@ add_split_table <- function(plot = NULL,
         p_digits = p_digits,
         estimate_fmt = estimate_fmt,
         ci_fmt = ci_fmt,
+        formatters = formatters,
+        reference_text = reference_text,
         text_size = text_size,
         header_text_size = header_text_size,
         header_fontface = header_fontface,
@@ -467,6 +483,8 @@ add_split_table <- function(plot = NULL,
     p_digits = p_digits,
     estimate_fmt = estimate_fmt,
     ci_fmt = ci_fmt,
+    formatters = formatters,
+    reference_text = reference_text,
     text_size = text_size,
     header_text_size = header_text_size,
     header_fontface = header_fontface,
diff --git a/R/as_forest_data.R b/R/as_forest_data.R
index 833ebd0..c8ab578 100644
--- a/R/as_forest_data.R
+++ b/R/as_forest_data.R
@@ -13,6 +13,11 @@
 #' @param term_labels Optional named vector used to relabel displayed terms.
 #'   Names should match values in the term column and values are the labels to
 #'   display.
+#' @param row_type Optional column name defining semantic rows: `"estimate"`,
+#'   `"reference"`, `"header"`, `"summary"`, or `"spacer"`. Estimate and
+#'   summary rows require finite estimate and CI values; other rows require
+#'   missing geometry. Structured rows retain source order, so use
+#'   `sort_terms = "none"`. If omitted, all rows are estimates.
 #' @param group Optional column name used for color-grouping multiple
 #'   estimates per row. If this column is a factor, its levels control the
 #'   group legend and vertical dodge order.
@@ -122,6 +127,9 @@ as_forest_data.forest_data <- function(data,
   if (!"label" %in% names(out)) {
     out$label <- out$term
   }
+  if (!"row_type" %in% names(out)) {
+    out$row_type <- rep("estimate", nrow(out))
+  }
   if (!is.null(p_method)) {
     stop(
       "`p_method` is already defined by the `forest_data` metadata.",
@@ -188,6 +196,7 @@ as_forest_data.data.frame <- function(data,
                                       sort_terms = c("none", "descending", "ascending"),
                                       subgroup = NULL,
                                       p_method = c("overall", "level"),
+                                      row_type = NULL,
                                       ...) {
   if (!inherits(data, "data.frame")) {
     stop(
@@ -239,6 +248,7 @@ as_forest_data.data.frame <- function(data,
     conf.low = resolve_column(data, conf.low, "conf.low"),
     conf.high = resolve_column(data, conf.high, "conf.high"),
     label = resolve_column(data, label, "label", required = FALSE),
+    row_type = resolve_column(data, row_type, "row_type", required = FALSE),
     group = resolve_column(data, group, "group", required = FALSE),
     subgroup = resolve_column(data, subgroup, "subgroup", required = FALSE),
     grouping = resolve_column(data, grouping, "grouping", required = FALSE),
@@ -269,6 +279,13 @@ as_forest_data.data.frame <- function(data,
     as.character(data[[cols$label]])
   }
   out$label <- apply_term_labels(out$term, out$label, term_labels)
+  out$row_type <- if (is.null(cols$row_type)) {
+    rep("estimate", nrow(out))
+  } else {
+    normalize_forest_row_type(data[[cols$row_type]])
+  }
+  out$term[out$row_type == "spacer" & is.na(out$term)] <- ""
+  out$label[out$row_type == "spacer" & is.na(out$label)] <- ""
 
   out$group <- if (is.null(cols$group)) {
     NA_character_
diff --git a/R/forest_data.R b/R/forest_data.R
index 1b691a4..a2e7d49 100644
--- a/R/forest_data.R
+++ b/R/forest_data.R
@@ -170,17 +170,23 @@ validate_forest_metadata <- function(data, metadata) {
       identical(metadata$axis_transform, "log10")
   )
 
+  geometry <- if ("row_type" %in% names(data)) {
+    data$row_type %in% forest_geometry_row_types()
+  } else {
+    rep(TRUE, nrow(data))
+  }
+
   bounded_columns <- c("estimate", "conf.low", "conf.high")
   if (identical(scale, "probability") &&
       any(vapply(bounded_columns, function(column) {
-        any(data[[column]] < 0 | data[[column]] > 1)
+        any(data[[column]][geometry] < 0 | data[[column]][geometry] > 1)
       }, logical(1)))) {
     stop("Probability estimates and intervals must be between 0 and 1.", call. = FALSE)
   }
 
   if (identical(scale, "risk_difference") &&
       any(vapply(bounded_columns, function(column) {
-        any(data[[column]] < -1 | data[[column]] > 1)
+        any(data[[column]][geometry] < -1 | data[[column]][geometry] > 1)
       }, logical(1)))) {
     stop("Risk-difference estimates and intervals must be between -1 and 1.", call. = FALSE)
   }
@@ -200,11 +206,13 @@ forest_axis_label <- function(metadata) {
   }
 }
 
-set_forest_metadata <- function(data, metadata) {
+set_forest_metadata <- function(data, metadata, validate = TRUE) {
   if (is.null(metadata$p_method)) {
     metadata$p_method <- "overall"
   }
-  validate_forest_metadata(data, metadata)
+  if (isTRUE(validate)) {
+    validate_forest_metadata(data, metadata)
+  }
   attr(data, "forest_meta") <- metadata
 
   # Transitional mirrors for code written against ggforestplotR <= 0.3.1.
@@ -283,7 +291,7 @@ strip_forest_data_class <- function(x) {
     return(strip_forest_data_class(out))
   }
 
-  set_forest_metadata(out, metadata)
+  set_forest_metadata(out, metadata, validate = nrow(out) > 0L)
 }
 
 #' @export
diff --git a/R/ggforestplot.R b/R/ggforestplot.R
index 848de81..ae183aa 100644
--- a/R/ggforestplot.R
+++ b/R/ggforestplot.R
@@ -16,6 +16,10 @@
 #' @param group Optional column name used for color-grouping estimates. If this
 #'   column is a factor, its levels control the group legend and vertical dodge
 #'   order.
+#' @param row_type Optional data-frame column mapping for semantic row types.
+#'   `"estimate"` and `"summary"` rows draw points and intervals;
+#'   `"reference"`, `"header"`, and `"spacer"` rows occupy display space
+#'   without geometry. Existing `forest_data` objects define this internally.
 #' @param subgroup Optional column name defining hierarchical subgroup blocks.
 #'   Missing or empty values identify ordinary standalone estimates. Each
 #'   non-empty subgroup must form one contiguous block within a facet. This is
@@ -139,7 +143,8 @@ ggforestplot <- function(data,
                          ref_linetype = 2,
                          ref_color = "grey60",
                          subgroup = NULL,
-                         p_method = c("overall", "level")) {
+                         p_method = c("overall", "level"),
+                         row_type = NULL) {
   ref_line_missing <- missing(ref_line)
   conf_level_missing <- missing(conf.level)
   p_method_missing <- missing(p_method)
@@ -167,6 +172,12 @@ ggforestplot <- function(data,
   if (inherits(data, "forest_data") && !p_method_missing) {
     stop("`p_method` is already defined by the `forest_data` metadata.", call. = FALSE)
   }
+  if (!is.null(row_type) && !is.data.frame(data)) {
+    stop("`row_type` is only a mapping for data-frame input.", call. = FALSE)
+  }
+  if (!is.null(row_type) && inherits(data, "forest_data")) {
+    stop("`row_type` is already defined by the `forest_data` object.", call. = FALSE)
+  }
 
   forest_data <- if (inherits(data, "forest_data")) {
     as_forest_data(
@@ -186,6 +197,7 @@ ggforestplot <- function(data,
       term_labels = term_labels,
       group = group,
       subgroup = subgroup,
+      row_type = row_type,
       grouping = facet,
       separate_groups = separate_groups,
       n = n,
@@ -251,7 +263,9 @@ ggforestplot <- function(data,
   display_data <- build_forest_plot_data(forest_data)
   forest_data <- display_data$forest_data
   plot_data <- display_data$plot_data
-  estimate_data <- plot_data[plot_data$row_type == "estimate", , drop = FALSE]
+  geometry_data <- plot_data[
+    plot_data$row_type %in% forest_geometry_row_types(), , drop = FALSE
+  ]
   stripe_data <- display_data$stripe_data
   separator_data <- display_data$separator_data
   plot_stripe_data <- stripe_data
@@ -263,7 +277,7 @@ ggforestplot <- function(data,
     plot_x_limits <- ci_limits
   } else if (isTRUE(plot_exponentiate)) {
     plot_x_limits <- default_plot_background_limits(
-      estimate_data,
+      geometry_data,
       exponentiate = plot_exponentiate,
       include_zero = draw_ref_line,
       ref_line = ref_line
@@ -276,11 +290,11 @@ ggforestplot <- function(data,
   }
 
   ci_plot_data <- build_ci_plot_data(
-    estimate_data,
+    geometry_data,
     ci_limits = ci_limits,
     exponentiate = plot_exponentiate
   )
-  has_groups <- has_table_values(estimate_data, "group")
+  has_groups <- has_table_values(geometry_data, "group")
   dodge <- ggplot2::position_dodge(width = dodge_width)
   point_mapping <- if (has_groups) {
     ggplot2::aes(
@@ -312,12 +326,7 @@ ggforestplot <- function(data,
   }
 
   p <- ggplot2::ggplot(ci_plot_data, point_mapping)
-  header_data <- plot_data[
-    plot_data$row_type == "subgroup_header",
-    ,
-    drop = FALSE
-  ]
-  if (nrow(header_data) > 0L) {
+  if (any(plot_data$row_type != "estimate")) {
     p <- p + ggplot2::geom_blank(
       data = plot_data,
       mapping = ggplot2::aes(y = .data$row_key),
diff --git a/R/ggplot_add_scales.R b/R/ggplot_add_scales.R
index ec9bb74..6f2beed 100644
--- a/R/ggplot_add_scales.R
+++ b/R/ggplot_add_scales.R
@@ -212,7 +212,7 @@ is_ggforestplot_faceted_y_scale <- function(scale, plot) {
   display_data <- forest_full_display_data(state)
   has_subgroup_headers <- !is.null(display_data) &&
     "row_type" %in% names(display_data) &&
-    any(display_data$row_type == "subgroup_header")
+    any(display_data$row_type != "estimate")
 
   if (!isTRUE(state$has_groupings) && !isTRUE(has_subgroup_headers)) {
     return(FALSE)
diff --git a/R/row_types.R b/R/row_types.R
new file mode 100644
index 0000000..1753984
--- /dev/null
+++ b/R/row_types.R
@@ -0,0 +1,24 @@
+forest_row_types <- function() {
+  c("estimate", "reference", "header", "summary", "spacer")
+}
+
+forest_geometry_row_types <- function() {
+  c("estimate", "summary")
+}
+
+normalize_forest_row_type <- function(x) {
+  if (!is.character(x) && !is.factor(x)) {
+    stop("`row_type` must contain character or factor values.", call. = FALSE)
+  }
+  values <- tolower(trimws(as.character(x)))
+  bad <- unique(values[is.na(values) | !values %in% forest_row_types()])
+  if (length(bad) > 0L) {
+    quoted <- ifelse(is.na(bad), "NA", sprintf('"%s"', bad))
+    stop(sprintf(
+      "Unsupported `row_type` value(s): %s. Allowed values are: %s.",
+      paste(quoted, collapse = ", "),
+      paste(sprintf('"%s"', forest_row_types()), collapse = ", ")
+    ), call. = FALSE)
+  }
+  values
+}
diff --git a/R/table_formatters.R b/R/table_formatters.R
new file mode 100644
index 0000000..d67264f
--- /dev/null
+++ b/R/table_formatters.R
@@ -0,0 +1,62 @@
+normalize_table_formatters <- function(formatters, data = NULL) {
+  if (is.null(formatters)) {
+    return(list())
+  }
+  if (!is.list(formatters) || is.null(names(formatters)) ||
+      anyNA(names(formatters)) || any(!nzchar(names(formatters)))) {
+    stop("`formatters` must be a named list of functions with non-empty names.", call. = FALSE)
+  }
+  keys <- names(formatters)
+  keys[keys == "p.value"] <- "p"
+  if (anyDuplicated(keys)) {
+    stop("`formatters` names must be unique after alias normalization.", call. = FALSE)
+  }
+  source_names <- if (inherits(data, "forest_data")) {
+    names(forest_source_columns(data))
+  } else if (!is.null(data)) {
+    names(data)
+  } else {
+    character()
+  }
+  allowed <- unique(c(
+    "estimate", "ci", "conf.low", "conf.high", "p", "n", "events", "group",
+    setdiff(source_names, c("term", "label"))
+  ))
+  unknown <- setdiff(keys, allowed)
+  if (length(unknown) > 0L) {
+    stop(sprintf("Unknown formatter target(s): %s.", paste(unknown, collapse = ", ")), call. = FALSE)
+  }
+  if (!all(vapply(formatters, is.function, logical(1)))) {
+    stop("Every `formatters` entry must be a function.", call. = FALSE)
+  }
+  names(formatters) <- keys
+  formatters
+}
+
+resolve_table_formatter <- function(formatters, key) {
+  formatter <- formatters[[key]]
+  if (is.null(formatter) && key %in% c("conf.low", "conf.high")) {
+    formatter <- formatters[["ci"]]
+  }
+  formatter
+}
+
+apply_table_formatter <- function(values, formatter = NULL, fallback = as.character,
+                                  key = "value") {
+  result <- tryCatch(
+    if (is.null(formatter)) fallback(values) else formatter(values),
+    error = function(e) {
+      stop(sprintf("Formatter for `%s` failed: %s", key, conditionMessage(e)),
+           call. = FALSE)
+    }
+  )
+  if (!is.atomic(result)) {
+    stop(sprintf("Formatter for `%s` must return an atomic vector.", key), call. = FALSE)
+  }
+  if (length(result) != length(values)) {
+    stop(sprintf("Formatter for `%s` must return one value per input value.", key), call. = FALSE)
+  }
+  result <- as.character(result)
+  result[is.na(result)] <- ""
+  result
+}
diff --git a/R/utils.R b/R/utils.R
index c7699c5..ab6dd5a 100644
--- a/R/utils.R
+++ b/R/utils.R
@@ -22,7 +22,8 @@ resolve_column <- function(data, column, arg, required = TRUE) {
 forest_display_reserved_columns <- function() {
   c(
     "row_key", "grouping_panel", "row_type", "display_label",
-    ".forest_source_row", ".display_identity"
+    ".forest_source_row", ".display_identity",
+    ".forest_generated_subgroup_header"
   )
 }
 
@@ -49,14 +50,26 @@ validate_forest_data <- function(data, exponentiate = FALSE) {
     )
   }
 
+  row_type <- if ("row_type" %in% names(data)) {
+    normalize_forest_row_type(data$row_type)
+  } else {
+    rep("estimate", nrow(data))
+  }
+  geometry <- row_type %in% forest_geometry_row_types()
   numeric_cols <- c("estimate", "conf.low", "conf.high")
 
+  if (!any(geometry)) {
+    stop("Forest data must contain at least one `estimate` or `summary` row.", call. = FALSE)
+  }
   for (col in numeric_cols) {
-    if (anyNA(data[[col]])) {
-      stop(sprintf("Column `%s` cannot contain missing values.", col), call. = FALSE)
+    if (!is.numeric(data[[col]]) ||
+        any(!is.finite(data[[col]][geometry]))) {
+      stop(sprintf("Column `%s` must contain finite numeric values for estimate and summary rows.", col), call. = FALSE)
     }
-
-    if (isTRUE(exponentiate) && any(data[[col]] <= 0)) {
+    if (any(!is.na(data[[col]][!geometry]))) {
+      stop("Reference, header, and spacer rows cannot contain estimate or confidence-interval values.", call. = FALSE)
+    }
+    if (isTRUE(exponentiate) && any(data[[col]][geometry] <= 0)) {
       stop(
         "Ratio estimates and logarithmic axes require strictly positive `estimate`, `conf.low`, and `conf.high` values.",
         call. = FALSE
@@ -64,10 +77,18 @@ validate_forest_data <- function(data, exponentiate = FALSE) {
     }
   }
 
-  if (any(data$conf.low > data$conf.high)) {
+  if (any(data$conf.low[geometry] > data$conf.high[geometry])) {
     stop("`conf.low` cannot be greater than `conf.high`.", call. = FALSE)
   }
 
+  labels <- if ("label" %in% names(data)) data$label else data$term
+  for (values in list(data$term, labels)) {
+    if (any(!is.na(row_type) & row_type != "spacer" &
+            (is.na(values) | !nzchar(trimws(as.character(values)))))) {
+      stop("Non-spacer rows require non-empty term and label values.", call. = FALSE)
+    }
+  }
+
   invisible(data)
 }
 
@@ -353,6 +374,13 @@ sort_forest_data <- function(data, sort_terms = c("none", "descending", "ascendi
   has_subgroups <- "subgroup" %in% names(data) &&
     any(!is.na(data$subgroup) & nzchar(data$subgroup))
 
+  has_structured_rows <- "row_type" %in% names(data) &&
+    any(data$row_type != "estimate")
+
+  if (isTRUE(has_structured_rows)) {
+    stop("`sort_terms` must be \"none\" when structured `row_type` values are used so row order is preserved.", call. = FALSE)
+  }
+
   if (isTRUE(has_subgroups)) {
     stop(
       "`sort_terms` must be \"none\" when `subgroup` is used so source row order is preserved.",
@@ -573,17 +601,19 @@ collapse_grouped_values <- function(formatted,
 }
 
 format_forest_p_values <- function(values, group = NULL, digits = 2, p_digits = digits,
-                                   force_group_labels = FALSE, align_groups = FALSE) {
+                                   force_group_labels = FALSE, align_groups = FALSE,
+                                   formatter = NULL) {
   p_digits <- resolve_table_digits(digits = digits, p_digits = p_digits)$p_digits
   values <- as.numeric(values)
   eps <- 10^(-p_digits)
-  formatted <- ifelse(
+  fallback <- function(values) ifelse(
     is.na(values),
     "",
     ifelse(values < eps, paste0("<", sprintf(paste0("%.", p_digits, "f"), eps)),
       sprintf(paste0("%.", p_digits, "f"), values)
     )
   )
+  formatted <- apply_table_formatter(values, formatter, fallback, key = "p")
   collapse_grouped_values(
     formatted,
     group,
@@ -598,7 +628,10 @@ format_forest_estimates <- function(estimate, conf.low, conf.high,
                                     interval_digits = digits,
                                     estimate_fmt = NULL,
                                     force_group_labels = FALSE,
-                                    align_groups = FALSE) {
+                                    align_groups = FALSE,
+                                    estimate_formatter = NULL,
+                                    conf_low_formatter = NULL,
+                                    conf_high_formatter = NULL) {
   digits <- resolve_table_digits(
     digits = digits,
     estimate_digits = estimate_digits,
@@ -611,9 +644,12 @@ format_forest_estimates <- function(estimate, conf.low, conf.high,
     stop("`estimate_fmt` must be a single character string.", call. = FALSE)
   }
 
-  estimate_text <- sprintf(paste0("%.", digits$estimate_digits, "f"), estimate)
-  conf_low_text <- sprintf(paste0("%.", digits$interval_digits, "f"), conf.low)
-  conf_high_text <- sprintf(paste0("%.", digits$interval_digits, "f"), conf.high)
+  estimate_text <- apply_table_formatter(estimate, estimate_formatter,
+    function(x) sprintf(paste0("%.", digits$estimate_digits, "f"), x), "estimate")
+  conf_low_text <- apply_table_formatter(conf.low, conf_low_formatter,
+    function(x) sprintf(paste0("%.", digits$interval_digits, "f"), x), "conf.low")
+  conf_high_text <- apply_table_formatter(conf.high, conf_high_formatter,
+    function(x) sprintf(paste0("%.", digits$interval_digits, "f"), x), "conf.high")
 
   formatted <- vapply(
     seq_along(estimate_text),
@@ -645,7 +681,9 @@ format_forest_intervals <- function(conf.low, conf.high,
                                     interval_digits = digits,
                                     ci_fmt = NULL,
                                     force_group_labels = FALSE,
-                                    align_groups = FALSE) {
+                                    align_groups = FALSE,
+                                    conf_low_formatter = NULL,
+                                    conf_high_formatter = NULL) {
   digits <- resolve_table_digits(
     digits = digits,
     interval_digits = interval_digits
@@ -657,8 +695,10 @@ format_forest_intervals <- function(conf.low, conf.high,
     stop("`ci_fmt` must be a single character string.", call. = FALSE)
   }
 
-  conf_low_text <- sprintf(paste0("%.", digits$interval_digits, "f"), conf.low)
-  conf_high_text <- sprintf(paste0("%.", digits$interval_digits, "f"), conf.high)
+  conf_low_text <- apply_table_formatter(conf.low, conf_low_formatter,
+    function(x) sprintf(paste0("%.", digits$interval_digits, "f"), x), "conf.low")
+  conf_high_text <- apply_table_formatter(conf.high, conf_high_formatter,
+    function(x) sprintf(paste0("%.", digits$interval_digits, "f"), x), "conf.high")
 
   formatted <- vapply(
     seq_along(conf_low_text),
@@ -687,9 +727,10 @@ format_forest_intervals <- function(conf.low, conf.high,
 format_forest_table_values <- function(values,
                                        group = NULL,
                                        force_group_labels = FALSE,
-                                       align_groups = FALSE) {
-  formatted <- as.character(values)
-  formatted[is.na(formatted)] <- ""
+                                       align_groups = FALSE,
+                                       formatter = NULL,
+                                       key = "value") {
+  formatted <- apply_table_formatter(values, formatter, as.character, key)
   collapse_grouped_values(
     formatted,
     group,
@@ -856,7 +897,7 @@ assign_hierarchical_row_keys <- function(data, has_groupings) {
 
   factor_levels <- if (
     "row_type" %in% names(data) &&
-      any(data$row_type == "subgroup_header")
+      any(data$row_type != "estimate")
   ) {
     rev(all_levels)
   } else {
@@ -939,8 +980,12 @@ expand_subgroup_display_rows <- function(data, has_groupings) {
   display_data <- as_forest_display_frame(data)
   p_method <- forest_p_method(display_data)
   display_data$.forest_source_row <- seq_len(nrow(display_data))
-  display_data$row_type <- "estimate"
+  if (!"row_type" %in% names(display_data)) {
+    display_data$row_type <- "estimate"
+  }
+  display_data$.forest_generated_subgroup_header <- FALSE
   display_data$display_label <- display_data$label
+  display_data$display_label[display_data$row_type == "spacer"] <- ""
 
   has_subgroups <- "subgroup" %in% names(display_data) &&
     any(!is.na(display_data$subgroup) & nzchar(display_data$subgroup))
@@ -975,7 +1020,8 @@ expand_subgroup_display_rows <- function(data, has_groupings) {
         header$label <- subgroup
         header$subgroup <- subgroup
         header$grouping_panel <- display_data$grouping_panel[row_index]
-        header$row_type <- "subgroup_header"
+        header$row_type <- "header"
+        header$.forest_generated_subgroup_header <- TRUE
         header$display_label <- subgroup
 
         block_end <- position
@@ -1043,7 +1089,8 @@ subgroup_header_rows <- function(data, header_row) {
 format_subgroup_header_p_values <- function(data,
                                             header_row,
                                             p_digits,
-                                            align_groups) {
+                                            align_groups,
+                                            formatter = NULL) {
   block <- subgroup_header_rows(data, header_row)
 
   if (nrow(block) == 0L) {
@@ -1055,7 +1102,7 @@ format_subgroup_header_p_values <- function(data,
       length(unique(block_values)) == 1L) {
     return(format_forest_p_values(
       block_values[[1L]],
-      p_digits = p_digits
+      p_digits = p_digits, formatter = formatter
     ))
   }
 
@@ -1064,7 +1111,7 @@ format_subgroup_header_p_values <- function(data,
   if (!isTRUE(has_groups)) {
     values <- block$p.value[!is.na(block$p.value)]
     value <- if (length(values) == 0L) NA_real_ else values[[1L]]
-    return(format_forest_p_values(value, p_digits = p_digits))
+    return(format_forest_p_values(value, p_digits = p_digits, formatter = formatter))
   }
 
   group_key <- ifelse(
@@ -1082,7 +1129,8 @@ format_subgroup_header_p_values <- function(data,
     values,
     group = group_levels,
     p_digits = p_digits,
-    align_groups = align_groups
+    align_groups = align_groups,
+    formatter = formatter
   )
 }
 
@@ -1214,14 +1262,23 @@ build_forest_plot_data <- function(data) {
   plot_data <- expand_subgroup_display_rows(forest_data, has_groupings)
   plot_data$.display_identity <- encode_display_identity(
     plot_data$row_type,
-    ifelse(plot_data$row_type == "estimate", plot_data$subgroup, NA_character_),
+    ifelse(plot_data$.forest_generated_subgroup_header |
+             plot_data$row_type %in% c("estimate", "reference", "summary"),
+           plot_data$subgroup, NA_character_),
     plot_data$label
   )
+  unique_source_rows <- plot_data$row_type == "spacer" |
+    (plot_data$row_type == "header" &
+       !plot_data$.forest_generated_subgroup_header)
+  plot_data$.display_identity[unique_source_rows] <- paste0(
+    plot_data$.display_identity[unique_source_rows],
+    "|source:", plot_data$.forest_source_row[unique_source_rows]
+  )
   plot_data <- assign_row_keys(plot_data, has_groupings)
 
-  estimate_rows <- plot_data$row_type == "estimate"
-  source_rows <- plot_data$.forest_source_row[estimate_rows]
-  source_keys <- as.character(plot_data$row_key[estimate_rows])
+  original_rows <- !is.na(plot_data$.forest_source_row)
+  source_rows <- plot_data$.forest_source_row[original_rows]
+  source_keys <- as.character(plot_data$row_key[original_rows])
   forest_data$row_key <- factor(
     source_keys[match(seq_len(nrow(forest_data)), source_rows)],
     levels = levels(plot_data$row_key)
@@ -1386,6 +1443,8 @@ build_forest_table_data <- function(data,
                                     p_digits = NULL,
                                     estimate_fmt = NULL,
                                     ci_fmt = NULL,
+                                    formatters = NULL,
+                                    reference_text = "Reference",
                                     column_labels = NULL,
                                     columns = NULL,
                                     display_data = data) {
@@ -1395,6 +1454,11 @@ build_forest_table_data <- function(data,
     interval_digits = interval_digits,
     p_digits = p_digits
   )
+  if (!is.character(reference_text) || length(reference_text) != 1L ||
+      is.na(reference_text)) {
+    stop("`reference_text` must be a single non-missing character string.", call. = FALSE)
+  }
+  formatters <- normalize_table_formatters(formatters, data)
   if (is.null(conf.level) && inherits(data, "forest_data")) {
     conf.level <- forest_metadata(data)$conf_level
   }
@@ -1437,8 +1501,8 @@ build_forest_table_data <- function(data,
   row_label_data_columns <- unique(c(
     "term", "label", mapped_row_label_columns
   ))
-  has_subgroup_headers <- "row_type" %in% names(display_data) &&
-    any(display_data$row_type == "subgroup_header")
+  has_subgroup_headers <- ".forest_generated_subgroup_header" %in%
+    names(display_data) && any(display_data$.forest_generated_subgroup_header)
   visible_data_columns <- names(display_data)[
     !startsWith(names(display_data), "..source..") &
       !names(display_data) %in% forest_display_reserved_columns()
@@ -1476,9 +1540,15 @@ build_forest_table_data <- function(data,
     } else {
       "estimate"
     }
-    row_types[[row_key]] <- row_type
-    is_header <- identical(row_type, "subgroup_header")
-    is_child <- !is_header &&
+    row_types[[i]] <- row_type
+    is_header <- identical(row_type, "header")
+    is_generated_header <- is_header &&
+      ".forest_generated_subgroup_header" %in% names(rd) &&
+      isTRUE(rd$.forest_generated_subgroup_header[[1L]])
+    is_reference <- identical(row_type, "reference")
+    is_spacer <- identical(row_type, "spacer")
+    is_geometry <- row_type %in% forest_geometry_row_types()
+    is_child <- !is_header && !is_spacer &&
       any(!is.na(rd$subgroup) & nzchar(rd$subgroup))
     term_text <- if ("display_label" %in% names(rd)) {
       rd$display_label[[1L]]
@@ -1490,7 +1560,7 @@ build_forest_table_data <- function(data,
     } else {
       rd$group
     }
-    row_group_values[[i]] <- if (is_header && identical(p_method, "overall")) {
+    row_group_values[[i]] <- if (is_generated_header && identical(p_method, "overall")) {
       subgroup_header_group_values(data, rd)
     } else {
       as.character(rd$group)
@@ -1499,50 +1569,67 @@ build_forest_table_data <- function(data,
       row_key = row_key,
       grouping_panel = rd$grouping_panel[1L],
       term_text = term_text,
-      group_text = if (is_header) "" else format_forest_table_values(
-        group_values, rd$group, align_groups = align_groups
+      group_text = if (is_header || is_spacer) "" else format_forest_table_values(
+        group_values, rd$group, align_groups = align_groups,
+        formatter = formatters[["group"]], key = "group"
       ),
-      n_text = if (is_header) "" else format_forest_table_values(
-        rd$n, rd$group, align_groups = align_groups
+      n_text = if (is_header || is_spacer) "" else format_forest_table_values(
+        rd$n, rd$group, align_groups = align_groups,
+        formatter = formatters[["n"]], key = "n"
       ),
-      events_text = if (is_header) "" else format_forest_table_values(
-        rd$events, rd$group, align_groups = align_groups
+      events_text = if (is_header || is_spacer) "" else format_forest_table_values(
+        rd$events, rd$group, align_groups = align_groups,
+        formatter = formatters[["events"]], key = "events"
       ),
-      estimate_text = if (is_header) "" else format_forest_estimates(
+      estimate_text = if (is_reference) collapse_grouped_values(
+        rep(reference_text, nrow(rd)), rd$group, align_groups = align_groups
+      ) else if (!is_geometry) "" else format_forest_estimates(
         rd$estimate, rd$conf.low, rd$conf.high, rd$group,
         estimate_digits = digits$estimate_digits,
         interval_digits = digits$interval_digits,
         estimate_fmt = estimate_fmt,
-        align_groups = align_groups
+        align_groups = align_groups,
+        estimate_formatter = formatters[["estimate"]],
+        conf_low_formatter = resolve_table_formatter(formatters, "conf.low"),
+        conf_high_formatter = resolve_table_formatter(formatters, "conf.high")
       ),
-      estimate_value_text = if (is_header) "" else format_forest_estimates(
+      estimate_value_text = if (is_reference) collapse_grouped_values(
+        rep(reference_text, nrow(rd)), rd$group, align_groups = align_groups
+      ) else if (!is_geometry) "" else format_forest_estimates(
         rd$estimate, rd$conf.low, rd$conf.high, rd$group,
         estimate_digits = digits$estimate_digits,
         interval_digits = digits$interval_digits,
         estimate_fmt = if (is.null(estimate_fmt)) "{estimate}" else estimate_fmt,
-        align_groups = align_groups
+        align_groups = align_groups,
+        estimate_formatter = formatters[["estimate"]],
+        conf_low_formatter = resolve_table_formatter(formatters, "conf.low"),
+        conf_high_formatter = resolve_table_formatter(formatters, "conf.high")
       ),
-      ci_text = if (is_header) "" else format_forest_intervals(
+      ci_text = if (!is_geometry) "" else format_forest_intervals(
         rd$conf.low, rd$conf.high, rd$group,
         interval_digits = digits$interval_digits,
         ci_fmt = ci_fmt,
-        align_groups = align_groups
+        align_groups = align_groups,
+        conf_low_formatter = resolve_table_formatter(formatters, "conf.low"),
+        conf_high_formatter = resolve_table_formatter(formatters, "conf.high")
       ),
-      p_text = if (is_header && identical(p_method, "overall")) {
+      p_text = if (is_generated_header && identical(p_method, "overall")) {
         format_subgroup_header_p_values(
           data,
           rd,
           p_digits = digits$p_digits,
-          align_groups = align_groups
+          align_groups = align_groups,
+          formatter = formatters[["p"]]
         )
-      } else if (is_header ||
+      } else if (is_reference || is_spacer || is_generated_header ||
                  (is_child && identical(p_method, "overall"))) {
         ""
       } else {
         format_forest_p_values(
           rd$p.value, rd$group,
           p_digits = digits$p_digits,
-          align_groups = align_groups
+          align_groups = align_groups,
+          formatter = formatters[["p"]]
         )
       },
       stringsAsFactors = FALSE
@@ -1550,8 +1637,12 @@ build_forest_table_data <- function(data,
 
     for (extra in extra_columns) {
       storage_field <- extra_storage_lookup[[extra]]
+      if (is_spacer) {
+        row_parts[[i]][[storage_field]] <- ""
+        next
+      }
       if (isTRUE(has_subgroup_headers) && extra %in% row_label_data_columns) {
-        if (is_header) {
+        if (is_generated_header) {
           row_parts[[i]][[storage_field]] <- term_text
           next
         }
@@ -1564,7 +1655,8 @@ build_forest_table_data <- function(data,
         label_text <- format_forest_table_values(
           label_values,
           rd$group,
-          align_groups = FALSE
+          align_groups = FALSE,
+          formatter = formatters[[extra]], key = extra
         )
         if (is_child) {
           label_text <- paste0(
@@ -1576,7 +1668,7 @@ build_forest_table_data <- function(data,
         next
       }
 
-      if (is_header) {
+      if (is_generated_header) {
         row_parts[[i]][[storage_field]] <- ""
         next
       }
@@ -1589,7 +1681,8 @@ build_forest_table_data <- function(data,
       row_parts[[i]][[storage_field]] <- format_forest_table_values(
         values,
         rd$group,
-        align_groups = align_groups
+        align_groups = align_groups,
+        formatter = formatters[[extra]], key = extra
       )
     }
   }
@@ -1660,7 +1753,7 @@ build_forest_table_data <- function(data,
     long_part <- data.frame(
       row_key = table_rows$row_key,
       grouping_panel = table_rows$grouping_panel,
-      row_type = unname(row_types[as.character(table_rows$row_key)]),
+      row_type = unname(row_types[match(as.character(table_rows$row_key), row_levels)]),
       column_key = key,
       column_position = NA_real_,
       text = table_rows[[column_field_lookup[[key]]]],
diff --git a/man/add_forest_table.Rd b/man/add_forest_table.Rd
index f572a35..3e05166 100644
--- a/man/add_forest_table.Rd
+++ b/man/add_forest_table.Rd
@@ -20,6 +20,8 @@ add_forest_table(
   p_digits = NULL,
   estimate_fmt = NULL,
   ci_fmt = NULL,
+  formatters = NULL,
+  reference_text = "Reference",
   text_size = NULL,
   header_text_size = NULL,
   header_fontface = "bold",
@@ -95,6 +97,15 @@ shorthand \verb{\{conf.low, conf.high\}} is also supported. Defaults to
 placeholders. The shorthand \verb{\{conf.low, conf.high\}} is also supported.
 Defaults to \code{"({conf.low}, {conf.high})"}.}
 
+\item{formatters}{Named list of vectorized functions for scalar table
+values. Keys include \code{estimate}, \code{ci}, \code{conf.low}, \code{conf.high}, \code{p}
+(or \code{p.value}), \code{n}, \code{events}, \code{group}, and preserved source columns.
+Each function receives a vector and returns one value per input.
+Functions take precedence over digit settings; \code{estimate_fmt} and
+\code{ci_fmt} compose the resulting strings.}
+
+\item{reference_text}{Text shown in estimate cells for reference rows.}
+
 \item{text_size}{Text size for table contents. Defaults to \code{3.2}.}
 
 \item{header_text_size}{Header text size for table column labels. Defaults
diff --git a/man/add_split_table.Rd b/man/add_split_table.Rd
index 85c2dfa..08f6509 100644
--- a/man/add_split_table.Rd
+++ b/man/add_split_table.Rd
@@ -20,6 +20,8 @@ add_split_table(
   p_digits = NULL,
   estimate_fmt = NULL,
   ci_fmt = NULL,
+  formatters = NULL,
+  reference_text = "Reference",
   text_size = NULL,
   header_text_size = NULL,
   header_fontface = "bold",
@@ -96,6 +98,14 @@ table columns include \code{"ci"}. Use \code{{conf.low}} and \code{{conf.high}}
 placeholders. The shorthand \verb{\{conf.low, conf.high\}} is also supported.
 Defaults to \code{"({conf.low}, {conf.high})"}.}
 
+\item{formatters}{Named list of vectorized functions for scalar table
+values. Keys include \code{estimate}, \code{ci}, \code{conf.low}, \code{conf.high}, \code{p}
+(or \code{p.value}), \code{n}, \code{events}, \code{group}, and preserved source columns.
+Functions take precedence over digit settings; \code{estimate_fmt} and
+\code{ci_fmt} compose the resulting strings.}
+
+\item{reference_text}{Text shown in estimate cells for reference rows.}
+
 \item{text_size}{Text size for table contents. Defaults to \code{3.2}.}
 
 \item{header_text_size}{Header text size for table column labels. Defaults
diff --git a/man/as_forest_data.Rd b/man/as_forest_data.Rd
index 4183692..e30dbd0 100644
--- a/man/as_forest_data.Rd
+++ b/man/as_forest_data.Rd
@@ -49,6 +49,7 @@ as_forest_data(data, ...)
   sort_terms = c("none", "descending", "ascending"),
   subgroup = NULL,
   p_method = c("overall", "level"),
+  row_type = NULL,
   ...
 )
 
@@ -238,6 +239,12 @@ the canonical \code{p.value} column for either the omnibus interaction test or
 row-level effect tests, according to \code{p_method}, so they can share a table
 column with ordinary coefficient p-values.}
 
+\item{row_type}{Optional column name defining semantic rows: \code{"estimate"},
+\code{"reference"}, \code{"header"}, \code{"summary"}, or \code{"spacer"}. Estimate and
+summary rows require finite estimate and CI values; other rows require
+missing geometry. Structured rows retain source order, so use
+\code{sort_terms = "none"}. If omitted, all rows are estimates.}
+
 \item{conf.int}{Logical; model methods require \code{TRUE} because forest data
 include confidence-interval columns.}
 
diff --git a/man/ggforestplot.Rd b/man/ggforestplot.Rd
index 19ea8e3..eda0c91 100644
--- a/man/ggforestplot.Rd
+++ b/man/ggforestplot.Rd
@@ -45,7 +45,8 @@ ggforestplot(
   ref_linetype = 2,
   ref_color = "grey60",
   subgroup = NULL,
-  p_method = c("overall", "level")
+  p_method = c("overall", "level"),
+  row_type = NULL
 )
 }
 \arguments{
@@ -168,6 +169,11 @@ inference inside the plotting layer.}
 \code{"overall"} displays one p-value on each subgroup header; \code{"level"}
 displays p-values on the subgroup estimate rows. Fitted-model subgroup
 data inherit this choice from \code{\link[=tidy_forest_model]{tidy_forest_model()}}.}
+
+\item{row_type}{Optional data-frame column mapping for semantic row types.
+\code{"estimate"} and \code{"summary"} rows draw points and intervals;
+\code{"reference"}, \code{"header"}, and \code{"spacer"} rows occupy display space
+without geometry. Existing \code{forest_data} objects define this internally.}
 }
 \value{
 A \code{ggplot} object. Use standard \code{ggplot2} functions such as
diff --git a/tests/testthat/helper-structured-rows.R b/tests/testthat/helper-structured-rows.R
new file mode 100644
index 0000000..b9072a5
--- /dev/null
+++ b/tests/testthat/helper-structured-rows.R
@@ -0,0 +1,27 @@
+structured_rows <- function() {
+  data.frame(
+    term = c("Section", "Female", "Male", NA, NA, "Overall"),
+    estimate = c(NA, NA, 1.42, NA, NA, 1.18),
+    conf.low = c(NA, NA, 1.10, NA, NA, 1.04),
+    conf.high = c(NA, NA, 1.84, NA, NA, 1.34),
+    row_type = c("header", "reference", "estimate", "spacer", "spacer", "summary"),
+    p.value = c(0.03, NA, 0.008, NA, NA, 0.01),
+    n = c(NA, 60, 55, NA, NA, 115),
+    extra = c("Head", "Ref", "Male", NA, NA, "All")
+  )
+}
+
+make_structured_plot <- function(raw = structured_rows(), ...) {
+  ggforestplot(raw, row_type = "row_type", p.value = "p.value",
+               n = "n", ref_line = NULL, ...)
+}
+
+as_structured_data <- function(raw, ...) {
+  as_forest_data(raw, term = "term", estimate = "estimate",
+                 conf.low = "conf.low", conf.high = "conf.high", ...)
+}
+
+table_cell <- function(spec, key, type) {
+  cells <- spec$table_data
+  cells$text[cells$column_key == key & cells$row_type == type]
+}
diff --git a/tests/testthat/test-as_forest_data.R b/tests/testthat/test-as_forest_data.R
index 114cba4..2cb98a4 100644
--- a/tests/testthat/test-as_forest_data.R
+++ b/tests/testthat/test-as_forest_data.R
@@ -25,14 +25,15 @@ test_that("as_forest_data standardizes coefficient columns", {
   )
 
   expect_equal(
-    names(out)[seq_len(12)],
+    names(out)[seq_len(13)],
     c(
-      "term", "estimate", "conf.low", "conf.high", "label", "group",
+      "term", "estimate", "conf.low", "conf.high", "label", "row_type", "group",
       "subgroup", "grouping", "separate_groups", "n", "events", "p.value"
     )
   )
   expect_true(all(c("variable", "beta", "lower", "upper", "cohort", "section", "block", "sample_size", "event_count") %in% names(out)))
   expect_equal(as.character(out$term), c("Age", "BMI"))
+  expect_true(all(out$row_type == "estimate"))
   expect_equal(as.character(out$group), c("A", "A"))
   expect_equal(as.character(out$grouping), c("Clinical", "Clinical"))
   expect_equal(as.character(out$separate_groups), c("Anthropometrics", "Anthropometrics"))
diff --git a/tests/testthat/test-row-types.R b/tests/testthat/test-row-types.R
new file mode 100644
index 0000000..646ed97
--- /dev/null
+++ b/tests/testthat/test-row-types.R
@@ -0,0 +1,116 @@
+test_that("row types normalize and require appropriate geometry", {
+  raw <- structured_rows()
+  raw$row_type[2] <- " Reference "
+  out <- as_structured_data(raw, row_type = "row_type")
+  expect_equal(out$row_type[2], "reference")
+  expect_equal(out$label[4:5], c("", ""))
+  expect_equal(unname(forest_metadata(out)$column_mapping[["row_type"]]), "row_type")
+
+  raw$row_type[2] <- "ref"
+  expect_error(as_structured_data(raw, row_type = "row_type"), "Unsupported.*ref")
+  raw$row_type[2] <- NA_character_
+  expect_error(as_structured_data(raw, row_type = "row_type"), "Unsupported.*NA")
+  raw$row_type[2] <- "reference"
+  raw$estimate[2] <- 1
+  expect_error(as_structured_data(raw, row_type = "row_type"), "cannot contain estimate")
+  raw$estimate[2] <- NA_real_
+  raw$conf.low[3] <- NA_real_
+  expect_error(as_structured_data(raw, row_type = "row_type"), "conf.low.*finite")
+  raw$conf.low[3] <- 1.1
+  raw$estimate[6] <- NA_real_
+  expect_error(as_structured_data(raw, row_type = "row_type"), "estimate.*finite")
+  raw <- structured_rows()
+  expect_error(as_structured_data(raw[1:2, ], row_type = "row_type"),
+               "at least one")
+  expect_error(as_structured_data(raw, row_type = "row_type",
+                              sort_terms = "ascending"), "sort_terms.*none")
+  raw$term[2] <- ""
+  expect_error(as_structured_data(raw, row_type = "row_type"), "non-empty")
+})
+
+test_that("structured rows preserve order and draw geometry only where required", {
+  p <- make_structured_plot()
+  display <- p$ggforestplotR_state$display_data
+  expect_equal(display$row_type, structured_rows()$row_type)
+  expect_equal(length(unique(as.character(display$row_key))), 6L)
+  expect_equal(rev(levels(display$row_key)), as.character(display$row_key))
+  expect_equal(display$display_label[4:5], c("", ""))
+  expect_true(all(!is.na(p$ggforestplotR_state$forest_data$row_key)))
+  built <- ggplot2::ggplot_build(p)
+  point_index <- which(vapply(p$layers, function(x) inherits(x$geom, "GeomPoint"), logical(1)))
+  expect_equal(nrow(built$data[[point_index]]), 2L)
+  expect_true(any(vapply(p$layers, function(x) inherits(x$geom, "GeomBlank"), logical(1))))
+})
+
+test_that("reference, header, summary, and spacer table cells follow semantics", {
+  p <- make_structured_plot()
+  spec <- build_forest_table_data(
+    p$ggforestplotR_state$forest_data,
+    display_data = p$ggforestplotR_state$display_data,
+    columns = c("term", "n", "estimate", "ci", "p", "extra")
+  )
+  expect_equal(table_cell(spec, "estimate", "reference"), "Reference")
+  expect_equal(table_cell(spec, "ci", "reference"), "")
+  expect_equal(table_cell(spec, "p", "reference"), "")
+  expect_equal(table_cell(spec, "n", "reference"), "60")
+  expect_equal(table_cell(spec, "extra", "reference"), "Ref")
+  expect_equal(table_cell(spec, "estimate", "summary"), "1.18")
+  expect_equal(table_cell(spec, "p", "header"), "0.030")
+  expect_equal(table_cell(spec, "extra", "header"), "Head")
+  expect_true(all(spec$table_data$text[spec$table_data$row_type == "spacer"] == ""))
+  spec <- build_forest_table_data(
+    p$ggforestplotR_state$forest_data,
+    display_data = p$ggforestplotR_state$display_data,
+    columns = "estimate", reference_text = "Ref."
+  )
+  expect_equal(table_cell(spec, "estimate", "reference"), "Ref.")
+})
+
+test_that("subgroup reference children are indented and generated headers stay marked", {
+  raw <- structured_rows()[2:3, ]
+  raw$subgroup <- "Sex"
+  p <- ggforestplot(raw, row_type = "row_type", subgroup = "subgroup", ref_line = NULL)
+  display <- p$ggforestplotR_state$display_data
+  expect_equal(display$row_type, c("header", "reference", "estimate"))
+  expect_equal(display$.forest_generated_subgroup_header, c(TRUE, FALSE, FALSE))
+  expect_equal(display$display_label[2], "   Female")
+  expect_equal(display$display_label[3], "   Male")
+})
+
+test_that("semantic scale checks apply to geometry rows only", {
+  raw <- structured_rows()
+  expect_s3_class(as_structured_data(raw, row_type = "row_type",
+                                      estimate_scale = "ratio"), "forest_data")
+  raw$estimate[3] <- 0
+  expect_error(as_structured_data(raw, row_type = "row_type",
+                                  estimate_scale = "ratio"), "strictly positive")
+  raw <- structured_rows()
+  raw$estimate[3] <- 1.2
+  expect_error(as_structured_data(raw, row_type = "row_type",
+                                  estimate_scale = "probability"), "between 0 and 1")
+  raw <- structured_rows()
+  raw$estimate[3] <- 1.2
+  expect_error(as_structured_data(raw, row_type = "row_type",
+                                  estimate_scale = "risk_difference"), "between -1 and 1")
+})
+
+test_that("grouped estimates still share rows while reference rows remain separate", {
+  raw <- data.frame(
+    term = c("Female", "Female", "Female", "Female"),
+    estimate = c(NA, NA, 1.2, 1.4),
+    conf.low = c(NA, NA, 1.0, 1.2),
+    conf.high = c(NA, NA, 1.4, 1.6),
+    type = c("reference", "reference", "estimate", "estimate"),
+    model = c("A", "B", "A", "B")
+  )
+  p <- ggforestplot(raw, row_type = "type", group = "model", ref_line = NULL)
+  display <- p$ggforestplotR_state$display_data
+  expect_equal(length(unique(as.character(display$row_key))), 2L)
+  expect_equal(length(unique(as.character(display$row_key[1:2]))), 1L)
+  expect_equal(length(unique(as.character(display$row_key[3:4]))), 1L)
+  spec <- build_forest_table_data(
+    p$ggforestplotR_state$forest_data,
+    display_data = display, columns = "estimate"
+  )
+  expect_equal(table_cell(spec, "estimate", "reference"), "Reference\nReference")
+})
diff --git a/tests/testthat/test-subgroup-effects.R b/tests/testthat/test-subgroup-effects.R
index 7e06522..5d402ac 100644
--- a/tests/testthat/test-subgroup-effects.R
+++ b/tests/testthat/test-subgroup-effects.R
@@ -199,7 +199,7 @@ test_that("interaction and covariate p-values share the canonical column", {
 
   plot <- ggforestplot(out)
   display_data <- plot$ggforestplotR_state$display_data
-  header <- display_data$row_type == "subgroup_header"
+  header <- display_data$row_type == "header"
   children <- display_data$row_type == "estimate" &
     !is.na(display_data$subgroup)
   standalone <- display_data$row_type == "estimate" &
@@ -232,7 +232,7 @@ test_that("level p-values stay on subgroup estimate rows", {
 
   plot <- ggforestplot(out)
   display_data <- plot$ggforestplotR_state$display_data
-  header <- display_data$row_type == "subgroup_header"
+  header <- display_data$row_type == "header"
   children <- display_data$row_type == "estimate" &
     !is.na(display_data$subgroup)
   standalone <- display_data$row_type == "estimate" &
@@ -297,12 +297,12 @@ test_that("model-derived rows stay aligned with plots and tables", {
 
   expect_equal(
     display_data$row_type,
-    c("subgroup_header", rep("estimate", 3L))
+    c("header", rep("estimate", 3L))
   )
   expect_equal(display_data$display_label[[1L]], "cyl")
   expect_equal(trimws(display_data$display_label[-1L]), c("4", "6", "8"))
   expect_true(all(!is.na(
-    display_data$p.value[display_data$row_type == "subgroup_header"]
+    display_data$p.value[display_data$row_type == "header"]
   )))
 
   table_spec <- build_forest_table_data(
@@ -312,7 +312,7 @@ test_that("model-derived rows stay aligned with plots and tables", {
   )
   table_data <- table_spec$table_data
   header_key <- as.character(
-    display_data$row_key[display_data$row_type == "subgroup_header"]
+    display_data$row_key[display_data$row_type == "header"]
   )
   header_cells <- table_data[
     as.character(table_data$row_key) == header_key,
diff --git a/tests/testthat/test-subgroup.R b/tests/testthat/test-subgroup.R
index acc2ef4..d1a5543 100644
--- a/tests/testthat/test-subgroup.R
+++ b/tests/testthat/test-subgroup.R
@@ -59,7 +59,7 @@ subgroup_is_indented <- function(display_label, original_label) {
 }
 
 expect_subgroup_header_blanks <- function(table_data, display_data) {
-  header_rows <- display_data$row_type == "subgroup_header"
+  header_rows <- display_data$row_type == "header"
   header_keys <- as.character(display_data$row_key[header_rows])
   expected_labels <- stats::setNames(
     display_data$display_label[header_rows],
@@ -107,7 +107,7 @@ test_that("subgroup is an explicit canonical mapping and is not inferred", {
 
   expect_equal(as.character(explicit$subgroup), raw$subgroup_name)
   expect_equal(nrow(explicit), nrow(raw))
-  expect_false("row_type" %in% names(explicit))
+  expect_true(all(explicit$row_type == "estimate"))
   expect_equal(
     unname(forest_metadata(explicit)$column_mapping[["subgroup"]]),
     "subgroup_name"
@@ -130,8 +130,8 @@ test_that("mixed standalone and subgroup rows preserve source order", {
   expect_equal(
     display_data$row_type,
     c(
-      "estimate", "subgroup_header", "estimate", "estimate",
-      "estimate", "subgroup_header", "estimate", "estimate"
+      "estimate", "header", "estimate", "estimate",
+      "estimate", "header", "estimate", "estimate"
     )
   )
   expect_equal(
@@ -173,7 +173,7 @@ test_that("p_method level keeps p-values on subgroup children", {
     p_method = "level"
   )
   display_data <- p$ggforestplotR_state$display_data
-  header_rows <- display_data$row_type == "subgroup_header"
+  header_rows <- display_data$row_type == "header"
   child_rows <- display_data$row_type == "estimate" &
     !is.na(display_data$subgroup) & nzchar(display_data$subgroup)
   standalone_rows <- display_data$row_type == "estimate" & !child_rows
@@ -205,13 +205,13 @@ test_that("subgroup hierarchy handles boundary compositions", {
   expect_equal(
     grouped_display$row_type,
     c(
-      "subgroup_header", "estimate", "estimate",
-      "subgroup_header", "estimate", "estimate"
+      "header", "estimate", "estimate",
+      "header", "estimate", "estimate"
     )
   )
   expect_equal(
     grouped_display$display_label[
-      grouped_display$row_type == "subgroup_header"
+      grouped_display$row_type == "header"
     ],
     c("Race", "Sex")
   )
@@ -251,7 +251,7 @@ test_that("point and confidence interval geoms exclude subgroup headers", {
   p <- subgroup_plot()
   display_data <- p$ggforestplotR_state$display_data
   header_keys <- as.character(
-    display_data$row_key[display_data$row_type == "subgroup_header"]
+    display_data$row_key[display_data$row_type == "header"]
   )
   point_indices <- subgroup_layer_indices(p, "GeomPoint")
   interval_indices <- subgroup_layer_indices(p, "GeomErrorbar")
@@ -297,7 +297,7 @@ test_that("subgroup forest table preserves hierarchy and row alignment", {
     display_data$display_label
   )
 
-  header_rows <- display_data$row_type == "subgroup_header"
+  header_rows <- display_data$row_type == "header"
   child_rows <- display_data$row_type == "estimate" &
     !is.na(display_data$subgroup) & nzchar(display_data$subgroup)
   standalone_rows <- display_data$row_type == "estimate" & !child_rows
@@ -472,7 +472,7 @@ test_that("subgroup parent p-values respect facets and estimate groups", {
     display_data = p$ggforestplotR_state$display_data
   )
   parent_p <- spec$table_data[
-    spec$table_data$row_type == "subgroup_header" &
+    spec$table_data$row_type == "header" &
       spec$table_data$column_key == "p",
     ,
     drop = FALSE
@@ -578,7 +578,7 @@ test_that("source columns are not shadowed by subgroup display internals", {
     display_data$row_key[display_data$row_type == "estimate"]
   )
   header_keys <- as.character(
-    display_data$row_key[display_data$row_type == "subgroup_header"]
+    display_data$row_key[display_data$row_type == "header"]
   )
 
   for (column in internal_names) {
@@ -641,7 +641,7 @@ test_that("custom y limits support header-only subgroup views", {
   expect_length(split_tables, 2L)
   expect_true(all(vapply(
     split_tables,
-    function(table) all(table$data$row_type == "subgroup_header"),
+    function(table) all(table$data$row_type == "header"),
     logical(1)
   )))
 })
diff --git a/tests/testthat/test-table-formatters.R b/tests/testthat/test-table-formatters.R
new file mode 100644
index 0000000..d7af4dd
--- /dev/null
+++ b/tests/testthat/test-table-formatters.R
@@ -0,0 +1,106 @@
+test_that("formatter specifications and results are validated", {
+  raw <- structured_rows()
+  p <- make_structured_plot(raw)
+  data <- p$ggforestplotR_state$forest_data
+  display <- p$ggforestplotR_state$display_data
+  build <- function(formatters) build_forest_table_data(
+    data, display_data = display, columns = c("estimate", "p"),
+    formatters = formatters
+  )
+  expect_error(build(list(function(x) x)), "named list")
+  expect_error(build("bad"), "named list")
+  expect_error(build(stats::setNames(list(identity), "")), "non-empty")
+  expect_error(build(list(p = identity, p.value = identity)), "unique")
+  expect_error(build(list(p = "bad")), "function")
+  expect_error(build(list(unknown = identity)), "Unknown formatter")
+  expect_error(build(list(estimate = function(x) stop("boom"))),
+               "Formatter for `estimate` failed: boom")
+  expect_error(apply_table_formatter(1:2, function(x) "fixed", key = "estimate"),
+               "one value per input")
+  expect_error(apply_table_formatter(1:2, function(x) list("a", "b"), key = "n"),
+               "atomic vector")
+  expect_error(build_forest_table_data(data, display_data = display,
+                                       reference_text = NA_character_), "reference_text")
+})
+
+test_that("estimate, CI, and p formatters precede templates and digits", {
+  p <- make_structured_plot()
+  args <- list(
+    data = p$ggforestplotR_state$forest_data,
+    display_data = p$ggforestplotR_state$display_data,
+    columns = c("estimate", "ci", "p"),
+    p_digits = 1,
+    estimate_fmt = "{estimate} [{conf.low}, {conf.high}]",
+    ci_fmt = "<{conf.low}|{conf.high}>",
+    formatters = list(
+      estimate = function(x) sprintf("%.1f", x),
+      ci = function(x) sprintf("%.3f", x),
+      conf.low = function(x) paste0("L", sprintf("%.2f", x)),
+      p.value = function(x) sprintf("P%.3f", x)
+    )
+  )
+  spec <- do.call(build_forest_table_data, args)
+  expect_equal(table_cell(spec, "estimate", "estimate"), "1.4 [L1.10, 1.840]")
+  expect_equal(table_cell(spec, "ci", "estimate"), "<L1.10|1.840>")
+  expect_equal(table_cell(spec, "p", "estimate"), "P0.008")
+  expect_equal(table_cell(spec, "p", "header"), "P0.030")
+  expect_equal(table_cell(spec, "estimate", "reference"), "Reference")
+  args$columns <- "estimate"
+  spec <- do.call(build_forest_table_data, args)
+  expect_equal(table_cell(spec, "estimate", "estimate"), "1.4 [L1.10, 1.840]")
+})
+
+test_that("custom source columns and non-geometric values are formatted", {
+  raw <- structured_rows()
+  raw$prevalence <- c(NA, 0.2, 0.33, NA, NA, 0.4)
+  raw$date <- as.Date("2026-01-01") + seq_len(nrow(raw))
+  p <- make_structured_plot(raw)
+  spec <- build_forest_table_data(
+    p$ggforestplotR_state$forest_data,
+    display_data = p$ggforestplotR_state$display_data,
+    columns = c("term", "n", "prevalence", "date"),
+    formatters = list(
+      n = function(x) paste0("N=", x),
+      prevalence = function(x) ifelse(is.na(x), NA_character_,
+                                      paste0(round(100 * x), "%")),
+      date = function(x) format(x, "%Y/%m/%d")
+    )
+  )
+  expect_equal(table_cell(spec, "n", "reference"), "N=60")
+  expect_equal(table_cell(spec, "prevalence", "reference"), "20%")
+  expect_match(table_cell(spec, "date", "reference"), "^2026/")
+  expect_true(all(spec$table_data$text[spec$table_data$row_type == "spacer"] == ""))
+})
+
+test_that("both public table adders accept formatters and reference text", {
+  skip_if_not_installed("patchwork")
+  p <- make_structured_plot()
+  formatter <- list(estimate = function(x) sprintf("%.1f", x))
+  expect_s3_class(
+    p + add_forest_table(columns = c("term", "estimate"),
+                         formatters = formatter, reference_text = "Ref."),
+    "patchwork"
+  )
+  expect_s3_class(
+    p + add_split_table(left_columns = "term", right_columns = "estimate",
+                        formatters = formatter, reference_text = "Ref."),
+    "patchwork"
+  )
+})
+
+test_that("generated subgroup p-values use the custom formatter", {
+  raw <- structured_rows()[2:3, ]
+  raw$subgroup <- "Sex"
+  raw$p.value[2] <- 0.007
+  p <- ggforestplot(raw, row_type = "row_type", subgroup = "subgroup",
+                    p.value = "p.value", ref_line = NULL)
+  spec <- build_forest_table_data(
+    p$ggforestplotR_state$forest_data,
+    display_data = p$ggforestplotR_state$display_data,
+    columns = c("estimate", "p"),
+    formatters = list(p = function(x) sprintf("P%.4f", x))
+  )
+  expect_equal(table_cell(spec, "p", "header"), "P0.0070")
+  expect_equal(table_cell(spec, "p", "reference"), "")
+  expect_equal(table_cell(spec, "p", "estimate"), "")
+})
diff --git a/vignettes/ggforestplotR-data-helpers.Rmd b/vignettes/ggforestplotR-data-helpers.Rmd
index 382e348..21086a7 100644
--- a/vignettes/ggforestplotR-data-helpers.Rmd
+++ b/vignettes/ggforestplotR-data-helpers.Rmd
@@ -26,6 +26,40 @@ library(ggplot2)
 This short article covers the `forest_data` interchange object and the helper
 functions that prepare it before a plot is drawn.
 
+## Structured forest data and reference rows
+
+Map a source column with `row_type =` to place reference categories and
+structural rows alongside estimates. `estimate` and `summary` rows need finite
+estimates and confidence bounds. `reference`, `header`, and `spacer` rows use
+`NA` for all three geometry values. Structured rows keep their source order;
+`sort_terms` must remain `"none"`.
+
+```{r structured-rows}
+structured <- data.frame(
+  term = c("Female", "Male", "Clinical", "Age", "", "Overall"),
+  estimate = c(NA, 1.42, NA, 1.03, NA, 1.18),
+  conf.low = c(NA, 1.10, NA, 1.01, NA, 1.04),
+  conf.high = c(NA, 1.84, NA, 1.05, NA, 1.34),
+  subgroup = c("Sex", "Sex", NA, NA, NA, NA),
+  row_type = c("reference", "estimate", "header", "estimate", "spacer", "summary")
+)
+
+ggforestplot(
+  structured,
+  subgroup = "subgroup",
+  row_type = "row_type",
+  exponentiate = TRUE
+) + add_split_table(
+  left_columns = "term",
+  right_columns = c("estimate", "ci"),
+  reference_text = "Reference"
+)
+```
+
+You can also supply explicit `"header"` and `"spacer"` rows in the source
+table. A mapped `subgroup` creates its own header; reference rows inside the
+subgroup appear as indented children.
+
 ## Use `as_forest_data()` to standardize a coefficient table
 
 `as_forest_data()` converts your column names into the internal structure used
diff --git a/vignettes/ggforestplotR-plot-customization.Rmd b/vignettes/ggforestplotR-plot-customization.Rmd
index a7fa95d..5a18848 100644
--- a/vignettes/ggforestplotR-plot-customization.Rmd
+++ b/vignettes/ggforestplotR-plot-customization.Rmd
@@ -25,6 +25,41 @@ library(ggplot2)
 
 This article focuses on utility and enhanced customization of forest plots and accompanied tables.
 
+## Formatting table values with functions
+
+`formatters` accepts a named list of vectorized functions. Each function
+receives a source vector and returns one value per input. You can format built
+in statistics and any preserved source column. A `ci` formatter applies to
+both confidence bounds unless `conf.low` or `conf.high` overrides it.
+
+```{r function-formatters}
+format_data <- data.frame(
+  term = c("Age", "BMI"),
+  estimate = c(0.31, -0.24),
+  conf.low = c(0.12, -0.40),
+  conf.high = c(0.56, -0.08),
+  p.value = c(0.0004, 0.023),
+  prevalence = c(0.1543, 0.2871)
+)
+
+ggforestplot(format_data, p.value = "p.value") +
+  add_forest_table(
+    columns = c("term", "prevalence", "estimate", "p"),
+    formatters = list(
+      estimate = function(x) sprintf("%.1f", x),
+      ci = function(x) sprintf("%.3f", x),
+      p = function(x) ifelse(x < 0.001, "<0.001", sprintf("%.3f", x)),
+      prevalence = function(x) sprintf("%.1f%%", 100 * x)
+    ),
+    estimate_fmt = "{estimate} [{conf.low}, {conf.high}]"
+  )
+```
+
+Formatter functions act on scalar source values before `estimate_fmt` and
+`ci_fmt` compose text, and before grouped cells are combined. Functions from
+other packages work when they follow the same vector input and output
+contract.
+
 ## Group rows and control strip placement
 
 `facet` creates section panels, and `facet_strip_position` controls which