diff --git a/.github/workflows/reusable-clang-tidy.yml b/.github/workflows/reusable-clang-tidy.yml
index e66909ffad..68ddea3ea2 100644
--- a/.github/workflows/reusable-clang-tidy.yml
+++ b/.github/workflows/reusable-clang-tidy.yml
@@ -96,10 +96,10 @@ jobs:
           set -o pipefail
           run-clang-tidy -j ${{ steps.nproc.outputs.nproc }} -p "${BUILD_DIR}" -quiet -fix -allow-no-checks ${TARGETS} 2>&1 | tee "${OUTPUT_FILE}"
 
-      - name: Print errors
+      - name: Print filtered clang-tidy errors
         if: ${{ steps.run_clang_tidy.outcome != 'success' }}
         run: |
-          sed '/error\||/!d' "${OUTPUT_FILE}"
+          bin/filter-clang-tidy.py "${OUTPUT_FILE}"
 
       - name: Upload clang-tidy output
         if: ${{ github.event.repository.visibility == 'public' && steps.run_clang_tidy.outcome != 'success' }}
@@ -143,12 +143,12 @@ jobs:
           \`\`\`
           EOF
 
-      - name: Append clang-tidy output to issue body (filter for errors and warnings)
+      - name: Append filtered clang-tidy output to issue body
         if: ${{ steps.run_clang_tidy.outcome != 'success' }}
         run: |
           if [ -f "${OUTPUT_FILE}" ]; then
-              # Extract lines containing 'error:', 'warning:', or 'note:'
-              grep -E '(error:|warning:|note:)' "${OUTPUT_FILE}" >"${FILTERED_OUTPUT_FILE}" || true
+              # Filter to the unique errors with their source context.
+              bin/filter-clang-tidy.py "${OUTPUT_FILE}" >"${FILTERED_OUTPUT_FILE}" || true
 
               # If filtered output is empty, use original (might be a different error format)
               if [ ! -s "${FILTERED_OUTPUT_FILE}" ]; then
diff --git a/bin/filter-clang-tidy.py b/bin/filter-clang-tidy.py
new file mode 100755
index 0000000000..204a4e36f3
--- /dev/null
+++ b/bin/filter-clang-tidy.py
@@ -0,0 +1,102 @@
+#!/usr/bin/env python3
+
+"""
+Reduce run-clang-tidy output to its unique errors.
+
+It does two things:
+
+  1. Filters the raw output down to diagnostics and their source-context lines
+     (the indented "  103 | ..." / "      | ^" lines clang-tidy prints),
+     matching the "path:line:col: error:" diagnostic shape.
+
+  2. Deduplicates. The same diagnostic in a header is reported once per
+     translation unit that includes it, so identical error blocks are collapsed
+     to their first occurrence.
+
+An "error block" is an "error:" line together with the indented context lines
+and any "note:" lines that follow it (up to the next "error:" line). Blocks are
+compared as a whole, so an error stays attached to its own context, and
+first-occurrence order is preserved.
+
+The deduplicated output goes to stdout; a summary of unique error counts per
+check is printed to stderr.
+
+Usage:
+  bin/filter-clang-tidy.py [INPUT_FILE]            # read from file, or
+  run-clang-tidy ... | bin/filter-clang-tidy.py    # read from stdin
+"""
+
+import re
+import sys
+from collections import Counter
+
+# A clang-tidy diagnostic line looks like "path:line:col: error: msg [check]".
+# Matching on that shape (rather than a loose "error" substring) avoids treating
+# progress lines whose paths contain "error" as diagnostics, e.g.
+#   [284/850][0.7s] /nix/.../clang-tidy ... src/.../error.cpp
+DIAG_RE = re.compile(r":\d+:\d+: (?:error|warning|note):")
+ERROR_RE = re.compile(r":\d+:\d+: error:")
+CHECK_RE = re.compile(r" error: .*\[([^\],]+)")
+
+
+def filter_and_dedup(lines: list[str]) -> list[str]:
+    """Keep diagnostics with their context, then drop duplicate error blocks."""
+    blocks: list[str] = []
+    seen: set[str] = set()
+    current: list[str] = []
+
+    def flush() -> None:
+        if not current:
+            return
+        block = "".join(current)
+        if block not in seen:
+            seen.add(block)
+            blocks.append(block)
+
+    for line in lines:
+        # Keep only diagnostics and their indented source-context lines; drop
+        # progress/status output and blank lines.
+        if not (DIAG_RE.search(line) or line[:1] in (" ", "\t")):
+            continue
+        # An "error:" line starts a new block; its context and any following
+        # "note:" lines (and their context) belong to it.
+        if ERROR_RE.search(line):
+            flush()
+            current = []
+        current.append(line)
+    flush()
+
+    return blocks
+
+
+def summarize(blocks: list[str]) -> Counter[str]:
+    """Count unique errors per check name (e.g. "bugprone-branch-clone")."""
+    counts: Counter[str] = Counter()
+    for block in blocks:
+        # The error line is the first line of the block.
+        match = CHECK_RE.search(block.splitlines()[0])
+        if match:
+            counts[match.group(1)] += 1
+    return counts
+
+
+def main() -> int:
+    if len(sys.argv) > 1 and sys.argv[1] != "-":
+        with open(sys.argv[1], encoding="utf-8") as f:
+            lines = f.readlines()
+    else:
+        lines = sys.stdin.readlines()
+
+    blocks = filter_and_dedup(lines)
+    # Blank line between blocks so distinct errors are easy to tell apart.
+    sys.stdout.write("\n".join(blocks))
+
+    print("\nUnique errors per check:", file=sys.stderr)
+    for check, count in summarize(blocks).most_common():
+        print(f"{count:>4} {check}", file=sys.stderr)
+
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())