diff --git a/examples/clocked_perft.py b/examples/clocked_perft.py
new file mode 100644
index 000000000..a58fe8eaf
--- /dev/null
+++ b/examples/clocked_perft.py
@@ -0,0 +1,70 @@
+"""Example: clock legal-move generation and perft counting for a position.
+
+The single most recycled hot loop in chess engines, RL-for-Universal-
+Compiler-Optimization pipelines and chessy is the legal-move generator
+plus the perft* node counters (the same "how many nodes can this machine
+walk" benchmark the RL-for-Universal-Compiler-Optimization README measures
+when tuning offline inference budgets). This example is deliberately tiny:
+copy the FEN, get canonical node counts (perft values that every engine
+test suite checks against), and get a real wall-clock nodes/sec figure for
+this machine and this python-chess build.
+
+Run:
+    python clocked_perft.py                    # start position, depth 4
+    python clocked_perft.py "<FEN>" 3          # custom position / depth
+
+Output (one parseable line plus per-depth details):
+    total_nodes_per_s=xxx  total_wall_s=yyy  total_nodes=zzz  perft=<dict>
+    d1=<a>  d2=<b>  ...  dN=<z>
+
+perft(W) is defined as the number of distinct lines of play of exactly W
+plies from the given position; the well-known reference values are:
+    start:  d1=20 d2=400 d3=8902 d4=197281 d5=4865609
+    "KiwiPete" (k1q1...):  d1=48 d2=2039 d3=97862 d4=4085603
+Those are exactly the numbers python-chess engines and UCI benches print
+to prove the node counter matches the canonical test suite.
+"""
+import sys
+import time
+
+import chess
+
+
+def perft(board, depth):
+    """Return the depth-bounded node count for exactly `depth` plies."""
+    if depth == 0:
+        return 1
+    nodes = 0
+    for move in board.legal_moves:
+        board.push(move)
+        nodes += perft(board, depth - 1)
+        board.pop()
+    return nodes
+
+
+def clocked_perft(fen, depth):
+    """Return (perft_wall_s, total_nodes, {d: n}) with a per-depth breakdown."""
+    board = chess.Board(fen)
+    t0 = time.perf_counter()
+    per_depth = {}
+    for d in range(1, depth + 1):
+        per_depth[d] = perft(board, d)
+    wall = time.perf_counter() - t0
+    total = sum(per_depth.values())
+    return wall, total, per_depth
+
+
+def main():
+    fen = sys.argv[1] if len(sys.argv) > 1 else chess.STARTING_FEN
+    depth = int(sys.argv[2]) if len(sys.argv) > 2 else 4
+    wall, total, per_depth = clocked_perft(fen, depth)
+    nps = total / wall if wall else 0.0
+    parts = ["total_nodes_per_s=%.1f" % nps,
+             "total_wall_s=%.3f" % wall,
+             "total_nodes=%d" % total,
+             "perft=" + str(per_depth)]
+    print("  ".join(parts))
+
+
+if __name__ == "__main__":
+    main()