diff --git a/.github/scripts/benchmark.R b/.github/scripts/benchmark.R
new file mode 100644
index 000000000..72ed6da60
--- /dev/null
+++ b/.github/scripts/benchmark.R
@@ -0,0 +1,105 @@
+# -- 1. Load PR version --
+message("--- Loading PR version ---")
+pkgload::load_all(".")
+pr_version <- as.character(packageVersion("cards"))
+message("PR version: ", pr_version)
+
+pr_ard_summary <- cards::ard_summary
+pr_data <- cards::ADSL
+
+pr_ard_summary(data = pr_data, variables = AGE)
+pkgload::unload("cards")
+
+# -- 2. Install and load main version --
+message("--- Installing main version ---")
+remotes::install_github("insightsengineering/cards", quiet = TRUE)
+library(cards)
+main_version <- as.character(packageVersion("cards"))
+message("Main version: ", main_version)
+
+main_ard_summary <- cards::ard_summary
+main_data <- cards::ADSL
+
+# -- 3. Multi-round benchmarks --
+set.seed(42)
+n_rounds <- 5L
+
+message("--- Running ard_summary benchmarks (", n_rounds, " rounds) ---")
+bench_rounds <- lapply(seq_len(n_rounds), function(r) {
+  message("  Round ", r)
+  res <- bench::mark(
+    `ard_summary (main)` = main_ard_summary(data = main_data, variables = AGE),
+    `ard_summary (pr)` = pr_ard_summary(data = pr_data, variables = AGE),
+    iterations = 100,
+    check = FALSE
+  )
+  data.frame(
+    expression = as.character(res$expression),
+    median_s = as.numeric(res$median),
+    round = r
+  )
+}) |>
+  do.call(what = rbind)
+
+# -- 4. Build comparison table with confidence intervals --
+build_comparison <- function(rounds_df) {
+  rounds_df$group <- sub(" \\((main|pr)\\)$", "", rounds_df$expression)
+  rounds_df$version <- ifelse(
+    grepl("(main)", rounds_df$expression, fixed = TRUE), "main", "pr"
+  )
+
+  groups <- unique(rounds_df$group)
+  rows <- lapply(groups, function(g) {
+    main_medians <- rounds_df$median_s[rounds_df$group == g & rounds_df$version == "main"]
+    pr_medians <- rounds_df$median_s[rounds_df$group == g & rounds_df$version == "pr"]
+
+    ratios <- pr_medians / main_medians
+    mean_ratio <- mean(ratios)
+    diff_pct <- (mean_ratio - 1) * 100
+
+    n <- length(ratios)
+    se <- sd(ratios) / sqrt(n)
+    t_crit <- qt(0.975, df = n - 1)
+    ci_lo <- (mean_ratio - t_crit * se - 1) * 100
+    ci_hi <- (mean_ratio + t_crit * se - 1) * 100
+
+    if (ci_hi < 0) {
+      verdict <- paste0("\U2705 ", round(diff_pct, 1), "%")
+    } else if (ci_lo > 0) {
+      verdict <- paste0("\U274C +", round(diff_pct, 1), "%")
+    } else {
+      sign_chr <- ifelse(diff_pct >= 0, "+", "")
+      verdict <- paste0("\U2796 ", sign_chr, round(diff_pct, 1), "%")
+    }
+
+    data.frame(
+      expression = g,
+      main = paste0(round(mean(main_medians) * 1000, 2), "ms"),
+      pr = paste0(round(mean(pr_medians) * 1000, 2), "ms"),
+      change = verdict,
+      ci = paste0("[", round(ci_lo, 1), "%, ", round(ci_hi, 1), "%]")
+    )
+  })
+  do.call(rbind, rows)
+}
+
+bench_tab <- build_comparison(bench_rounds)
+
+header <- paste0(
+  "## Performance Benchmark\n\n",
+  "Comparing **main** (`", main_version, "`) vs **PR** (`", pr_version, "`)\n\n",
+  "Each benchmark runs ", n_rounds, " independent rounds. ",
+  "The **change** column shows the mean % difference (negative = faster).\n",
+  "The **95% CI** column shows the confidence interval on the change. ",
+  "If the CI excludes 0%, the result is flagged as a real improvement (\U2705) or regression (\U274C).\n\n"
+)
+
+bench_section <- paste0(
+  "### ard_summary()\n\n",
+  paste(knitr::kable(bench_tab, format = "markdown"), collapse = "\n"),
+  "\n"
+)
+
+report <- paste0(header, bench_section)
+writeLines(report, "bench_report.md")
+cat(report)
diff --git a/.github/workflows/R-CMD-check.yaml b/.github/workflows/R-CMD-check.yaml
index 6a9cd5dbd..c024cd1e1 100644
--- a/.github/workflows/R-CMD-check.yaml
+++ b/.github/workflows/R-CMD-check.yaml
@@ -31,8 +31,8 @@ jobs:
           - {os: ubuntu-latest,   r: 'release'}
           - {os: ubuntu-latest,   r: 'oldrel-1'}
           - {os: ubuntu-latest,   r: 'oldrel-2'}
-          # - {os: ubuntu-latest,   r: 'oldrel-3'}
-          # - {os: ubuntu-latest,   r: 'oldrel-4'}
+          - {os: ubuntu-latest,   r: 'oldrel-3'}
+          - {os: ubuntu-latest,   r: 'oldrel-4'}
 
     env:
       GITHUB_PAT: ${{ secrets.GITHUB_TOKEN }}
diff --git a/.github/workflows/benchmark.yaml b/.github/workflows/benchmark.yaml
new file mode 100644
index 000000000..b400ac832
--- /dev/null
+++ b/.github/workflows/benchmark.yaml
@@ -0,0 +1,41 @@
+name: Performance Benchmark
+
+on:
+  pull_request:
+    branches: [main]
+
+jobs:
+  benchmark:
+    if: startsWith(github.event.pull_request.title, 'perf')
+    runs-on: ubuntu-latest
+    env:
+      GITHUB_PAT: ${{ secrets.GITHUB_TOKEN }}
+    permissions:
+      pull-requests: write
+    steps:
+      - name: Checkout PR
+        uses: actions/checkout@v4
+
+      - name: Setup R
+        uses: r-lib/actions/setup-r@v2
+
+      - name: Install Dependencies & Cache
+        uses: r-lib/actions/setup-r-dependencies@v2
+        with:
+          dependencies: '"hard"'
+          extra-packages: |
+            any::bench
+            any::knitr
+            any::remotes
+            any::pkgload
+          cache: true
+          cache-version: 1
+
+      - name: Run Benchmarks
+        run: Rscript .github/scripts/benchmark.R
+
+      - name: Post Comment
+        uses: mshick/add-pr-comment@v2
+        with:
+          message-path: bench_report.md
+          refresh-main-comment: true