diff --git a/rickshaw-run.py b/rickshaw-run.py
index e91db694..bc7ca193 100755
--- a/rickshaw-run.py
+++ b/rickshaw-run.py
@@ -1268,8 +1268,9 @@ def load_tool_params(self):
 
             tool_id_map[tool_id] = tool_name
 
+            this_tool_dir = os.path.join(self.run["tools-dir"], tool_name)
+
             if tool_name not in self.tools_configs:
-                this_tool_dir = os.path.join(self.run["tools-dir"], tool_name)
                 this_tool_config = os.path.join(this_tool_dir, "rickshaw.json")
                 tool_cfg, err = load_json_file(this_tool_config)
                 if tool_cfg is None:
@@ -1284,12 +1285,101 @@ def load_tool_params(self):
                     sys.exit(1)
                 self.tools_configs[tool_name] = tool_cfg
 
-            self.bench_dirs[tool_id] = os.path.join(self.run["tools-dir"], tool_name)
+            self.apply_tool_multiplex(tool_entry, this_tool_dir)
+
+            self.bench_dirs[tool_id] = this_tool_dir
             for arch in self.required_archs:
                 self.image_ids.setdefault(arch, {}).setdefault(tool_id, {})[tool_entry["userenv"]] = {"image": ""}
 
         save_json_file(os.path.join(self.config_dir, "tool-id-map.json"), tool_id_map)
 
+    def apply_tool_multiplex(self, tool_entry, tool_dir):
+        """If this tool declares a multiplex.json, run its flat tool-params.json
+        params through multiplex.py (unmodified) for validation/transform/preset
+        support identical to benchmarks. No-op if the tool has no multiplex.json
+        (fully backward compatible for tools that haven't opted in).
+        """
+        requirements_file = os.path.join(tool_dir, "multiplex.json")
+        if not os.path.exists(requirements_file):
+            return
+
+        tool_id = tool_entry["tool-id"]
+
+        # WRAP: tool-params.json's flat {"arg","val"} shape -> multiplex's
+        # "sets" shape. Every value becomes a one-element "vals" array, which
+        # makes itertools.product() (multiplex's actual multiplication step)
+        # structurally incapable of producing more than one combination --
+        # this is what makes reusing multiplex.py unmodified safe for tools.
+        #
+        # Pre-filter disabled params ourselves (rather than relying on
+        # multiplex's own enabled-filtering) since multiplex's essentials-
+        # preset-override runs BEFORE its enabled-filtering step and could
+        # otherwise silently resurrect a param the user explicitly disabled.
+        wrapped_params = [
+            {
+                "arg": p["arg"],
+                "vals": [p["val"]],
+                "role": "all",   # tools have no client/server concept; set
+                                 # explicitly rather than relying on multiplex's
+                                 # sanitize_set() default of "client", which
+                                 # could cause spurious dedup mismatches
+                                 # against presets that specify "role": "server"
+            }
+            for p in tool_entry.get("params", [])
+            if p.get("enabled") != "no"
+        ]
+
+        mv_params_doc = {
+            "schema": {"version": "2021.12.02"},
+            "global-options": [],
+            "sets": [{"params": wrapped_params}],
+        }
+        mv_params_file = os.path.join(self.config_dir, f"{tool_id}-mv-params.json")
+        # toolbox.json.save_json_file() unconditionally xz-compresses and
+        # renames to "<file>.xz" with no way to opt out -- multiplex.py's own
+        # file reading has no decompression support, so a plain write is
+        # required here (matching how the benchmark mv-params path writes its
+        # own equivalent file directly rather than via save_json_file()).
+        with open(mv_params_file, "w") as f:
+            json.dump(mv_params_doc, f)
+
+        tool_params_out_file = os.path.join(self.config_dir, f"{tool_id}-tool-params-multiplexed.json")
+        multiplex_cmd = (
+            f"{os.environ.get('MULTIPLEX_HOME', '')}/multiplex.py "
+            f"--input {mv_params_file} --output {tool_params_out_file} "
+            f"--requirements {requirements_file}"
+        )
+        logger.debug("about to run: %s", multiplex_cmd)
+        _, output, rc = run_cmd(multiplex_cmd)
+        if rc != 0:
+            logger.error("[ERROR] multiplex failed with an error for tool '%s' and returned rc=%d", tool_id, rc)
+            logger.error("multiplex output is:\n%s", output)
+            sys.exit(1)
+
+        # UNWRAP: multiplex output -> tool-params.json's flat shape. Only
+        # arg/val survive -- role/id have no meaning for tools and would
+        # violate tool-params.json's additionalProperties:false if kept.
+        multiplexed_sets, err = load_json_file(tool_params_out_file)
+        if multiplexed_sets is None:
+            logger.error("[ERROR] could not load multiplexed tool params for '%s': %s", tool_id, err)
+            sys.exit(1)
+
+        if len(multiplexed_sets) != 1:
+            # Should be structurally impossible given the wrap above -- fail
+            # loudly instead of silently taking [0] if a tool's multiplex.json
+            # ever violates that assumption (e.g. via a misused "include").
+            logger.error(
+                "[ERROR] tool '%s' multiplex.json produced %d parameter combinations; "
+                "tool params must resolve to exactly one combination",
+                tool_id, len(multiplexed_sets),
+            )
+            sys.exit(1)
+
+        tool_entry["params"] = [
+            {"arg": param["arg"], "val": param["val"]}
+            for param in multiplexed_sets[0]
+        ]
+
     def load_utility_params(self):
         for utility in UTILITIES:
             utility_dir_key = f"{utility}-dir"
diff --git a/tests/test_apply_tool_multiplex.py b/tests/test_apply_tool_multiplex.py
new file mode 100644
index 00000000..1a75d246
--- /dev/null
+++ b/tests/test_apply_tool_multiplex.py
@@ -0,0 +1,216 @@
+#!/usr/bin/env python3
+# -*- mode: python; indent-tabs-mode: nil; python-indent-level: 4 -*-
+# vim: autoindent tabstop=4 shiftwidth=4 expandtab softtabstop=4 filetype=python
+
+"""Unit tests for rickshaw-run.py's RunState.apply_tool_multiplex() -- the
+bridge that lets a tool's flat tool-params.json params get validated/
+transformed/preset-applied by multiplex.py, identical to benchmarks, without
+ever risking real parameter multiplication (crucible#653).
+
+toolbox is mocked out rather than required, since rickshaw-run.py imports
+from it at module scope and CI does not check toolbox out for this job.
+multiplex.py itself is not invoked -- run_cmd is mocked so these tests
+exercise only the wrap/unwrap logic in apply_tool_multiplex(), not
+multiplex.py's own internals (which are that project's own responsibility).
+"""
+
+import importlib.machinery
+import importlib.util
+import json
+import logging
+import os
+import re
+import sys
+import tempfile
+import types
+import unittest
+from unittest.mock import MagicMock
+
+
+def import_rickshaw_run():
+    """Load rickshaw-run.py as a module with toolbox mocked out."""
+    mock_fileio = types.ModuleType("toolbox.fileio")
+    mock_fileio.open_write_text_file = lambda *a, **k: None
+
+    mock_json = types.ModuleType("toolbox.json")
+
+    def fake_load_json_file(json_file, uselzma=False):
+        try:
+            with open(json_file, "r") as f:
+                return json.load(f), None
+        except Exception as e:
+            return None, str(e)
+
+    def fake_save_json_file(filename, data, schema_file=None):
+        with open(filename, "w") as f:
+            json.dump(data, f)
+
+    mock_json.load_json_file = fake_load_json_file
+    mock_json.save_json_file = fake_save_json_file
+    mock_json.validate_schema = lambda *a, **k: (True, None)
+
+    mock_jsonsettings = types.ModuleType("toolbox.jsonsettings")
+    mock_jsonsettings.get_json_setting = lambda *a, **k: None
+
+    mock_logging_mod = types.ModuleType("toolbox.logging")
+    mock_logging_mod.setup_logging = lambda *a, **k: None
+
+    mock_roadblock = types.ModuleType("toolbox.roadblock")
+    mock_roadblock.do_roadblock = lambda *a, **k: (0, None)
+    mock_roadblock.ROADBLOCK_EXITS = {
+        "success": 0, "input": 2, "timeout": 3,
+        "abort": 4, "heartbeat_timeout": 5, "abort_waiting": 6,
+    }
+
+    mock_run = types.ModuleType("toolbox.run")
+    mock_run.run_cmd = MagicMock()
+
+    mock_toolbox = types.ModuleType("toolbox")
+    mock_toolbox.fileio = mock_fileio
+    mock_toolbox.json = mock_json
+    mock_toolbox.jsonsettings = mock_jsonsettings
+    mock_toolbox.logging = mock_logging_mod
+    mock_toolbox.roadblock = mock_roadblock
+    mock_toolbox.run = mock_run
+
+    mod_name = "rickshaw_run_under_test_multiplex"
+    sys.modules.pop(mod_name, None)
+
+    mocks = {
+        "toolbox": mock_toolbox,
+        "toolbox.fileio": mock_fileio,
+        "toolbox.json": mock_json,
+        "toolbox.jsonsettings": mock_jsonsettings,
+        "toolbox.logging": mock_logging_mod,
+        "toolbox.roadblock": mock_roadblock,
+        "toolbox.run": mock_run,
+    }
+    saved = {key: sys.modules.get(key) for key in mocks}
+    sys.modules.update(mocks)
+
+    try:
+        script_path = os.path.join(os.path.dirname(__file__), "..", "rickshaw-run.py")
+        loader = importlib.machinery.SourceFileLoader(mod_name, script_path)
+        spec = importlib.util.spec_from_loader(mod_name, loader)
+        mod = importlib.util.module_from_spec(spec)
+        sys.modules[mod_name] = mod
+        spec.loader.exec_module(mod)
+    finally:
+        for key, val in saved.items():
+            if val is None:
+                sys.modules.pop(key, None)
+            else:
+                sys.modules[key] = val
+
+    mod.logger = logging.getLogger("test_apply_tool_multiplex")
+    return mod
+
+
+def fake_multiplex(sets):
+    """Build a run_cmd side_effect simulating multiplex.py: reads the --input
+    file this code wrote, writes back `sets` (a list of param lists) to the
+    --output path, and returns rc=0. Pass rc= to simulate a failure instead.
+    """
+
+    def _side_effect(cmd):
+        match = re.search(r"--output (\S+)", cmd)
+        output_file = match.group(1)
+        with open(output_file, "w") as f:
+            json.dump(sets, f)
+        return (cmd, "", 0)
+
+    return _side_effect
+
+
+class TestApplyToolMultiplex(unittest.TestCase):
+    def setUp(self):
+        self.mod = import_rickshaw_run()
+        self.state = self.mod.RunState()
+        self.state.config_dir = tempfile.mkdtemp()
+        self.tool_dir = tempfile.mkdtemp()
+        self.run_cmd = self.mod.run_cmd
+        self.run_cmd.reset_mock()
+
+    def _write_multiplex_json(self):
+        with open(os.path.join(self.tool_dir, "multiplex.json"), "w") as f:
+            json.dump({"validations": {}}, f)
+
+    def test_no_multiplex_json_is_noop(self):
+        tool_entry = {"tool-id": "kernel", "params": [{"arg": "interval", "val": "10"}]}
+        self.state.apply_tool_multiplex(tool_entry, self.tool_dir)
+        self.assertEqual(tool_entry["params"], [{"arg": "interval", "val": "10"}])
+        self.run_cmd.assert_not_called()
+
+    def test_wraps_and_unwraps_round_trip(self):
+        self._write_multiplex_json()
+        self.run_cmd.side_effect = fake_multiplex(
+            [[{"arg": "interval", "val": "10", "role": "all"}]]
+        )
+        tool_entry = {"tool-id": "kernel", "params": [{"arg": "interval", "val": "10"}]}
+        self.state.apply_tool_multiplex(tool_entry, self.tool_dir)
+
+        # role must not leak into the unwrapped result
+        self.assertEqual(tool_entry["params"], [{"arg": "interval", "val": "10"}])
+
+        # verify the wrapped input actually sent to multiplex used a
+        # one-element vals array and role "all", not multiplex's "client" default
+        cmd = self.run_cmd.call_args[0][0]
+        input_file = re.search(r"--input (\S+)", cmd).group(1)
+        with open(input_file) as f:
+            wrapped = json.load(f)
+        self.assertEqual(
+            wrapped["sets"][0]["params"],
+            [{"arg": "interval", "vals": ["10"], "role": "all"}],
+        )
+
+    def test_disabled_param_excluded_from_wrap(self):
+        self._write_multiplex_json()
+        self.run_cmd.side_effect = fake_multiplex([[]])
+        tool_entry = {
+            "tool-id": "kernel",
+            "params": [{"arg": "interval", "val": "10", "enabled": "no"}],
+        }
+        self.state.apply_tool_multiplex(tool_entry, self.tool_dir)
+
+        cmd = self.run_cmd.call_args[0][0]
+        input_file = re.search(r"--input (\S+)", cmd).group(1)
+        with open(input_file) as f:
+            wrapped = json.load(f)
+        self.assertEqual(wrapped["sets"][0]["params"], [])
+
+    def test_multiplex_failure_exits(self):
+        self._write_multiplex_json()
+        self.run_cmd.return_value = ("cmd", "some error", 1)
+        tool_entry = {"tool-id": "kernel", "params": [{"arg": "interval", "val": "10"}]}
+        with self.assertRaises(SystemExit):
+            self.state.apply_tool_multiplex(tool_entry, self.tool_dir)
+
+    def test_empty_set_fail_without_defaults_preset_exits(self):
+        # Regression lock-in: EC_EMPTY_SET_FAIL (rc=6) is a real, intentional
+        # backward-compatibility trap -- a tool entry with zero params (legal
+        # today per schema/tool-params.json, meaning "use the *-start
+        # script's own bash defaults") starts hard-failing here unless the
+        # tool's multiplex.json defines a "defaults" preset. This test locks
+        # in that rickshaw-run treats it as fatal (via the generic rc!=0
+        # handling) rather than "fixing it away" silently in the future.
+        self._write_multiplex_json()
+        self.run_cmd.return_value = ("cmd", "empty param set", 6)
+        tool_entry = {"tool-id": "kernel", "params": []}
+        with self.assertRaises(SystemExit):
+            self.state.apply_tool_multiplex(tool_entry, self.tool_dir)
+
+    def test_more_than_one_combination_is_rejected(self):
+        self._write_multiplex_json()
+        self.run_cmd.side_effect = fake_multiplex(
+            [
+                [{"arg": "interval", "val": "10", "role": "all"}],
+                [{"arg": "interval", "val": "20", "role": "all"}],
+            ]
+        )
+        tool_entry = {"tool-id": "kernel", "params": [{"arg": "interval", "val": "10"}]}
+        with self.assertRaises(SystemExit):
+            self.state.apply_tool_multiplex(tool_entry, self.tool_dir)
+
+
+if __name__ == "__main__":
+    unittest.main()