diff --git a/.github/workflows/test.yml b/.github/workflows/test.yml
index a7d61b29..9b77dd8d 100644
--- a/.github/workflows/test.yml
+++ b/.github/workflows/test.yml
@@ -64,7 +64,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        ruby: [ruby, head]
+        ruby: [head]
     if: ${{ github.event_name != 'schedule' || github.repository == 'ruby/ruby-bench' }}
     steps:
       - uses: actions/checkout@v3
@@ -85,7 +85,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        ruby: [ruby, head]
+        ruby: [head]
     if: ${{ github.event_name != 'schedule' || github.repository == 'ruby/ruby-bench' }}
     steps:
       - uses: actions/checkout@v3
diff --git a/benchmarks.yml b/benchmarks.yml
index 7c0297d6..5b61fd2f 100644
--- a/benchmarks.yml
+++ b/benchmarks.yml
@@ -91,6 +91,11 @@ graphql-native:
   desc: GraphQL gem parsing a large file, but using a native parser
 knucleotide:
   desc: k-nucleotide from the Computer Language Benchmarks Game - counts nucleotide frequencies using hash tables in parallel using Process.fork
+knucleotide-ractor:
+  desc: k-nucleotide from the Computer Language Benchmarks Game - counts nucleotide frequencies using hash tables in parallel using Ractors
+  ractor: true
+  ractor_only: true
+  default_harness: harness
 lee:
   desc: lee is a circuit-board layout solver, deployed in a plausibly reality-like way
 matmul:
@@ -231,13 +236,20 @@ throw:
   ractor: true
 
 #
-# Ractor-only benchmarks
+# Ractor scaling benchmarks
 #
-ractor/knucleotide:
-  desc: k-nucleotide from the Computer Language Benchmarks Game - counts nucleotide frequencies using hash tables. Counts groups in parallel using Ractors.
-ractor/gvl_release_acquire:
+gvl_release_acquire:
   desc: microbenchmark designed to test how fast the gvl can be acquired and released between ractors.
-ractor/json_parse_float:
+  ractor: true
+  ractor_only: true
+  default_harness: harness-ractor
+json_parse_float:
   desc: test the performance of parsing multiple lists of json floats with ractors.
-ractor/json_parse_string:
+  ractor: true
+  ractor_only: true
+  default_harness: harness-ractor
+json_parse_string:
   desc: test the performance of parsing multiple lists of strings with ractors.
+  ractor: true
+  ractor_only: true
+  default_harness: harness-ractor
diff --git a/benchmarks-ractor/gvl_release_acquire/benchmark.rb b/benchmarks/gvl_release_acquire/benchmark.rb
similarity index 100%
rename from benchmarks-ractor/gvl_release_acquire/benchmark.rb
rename to benchmarks/gvl_release_acquire/benchmark.rb
diff --git a/benchmarks-ractor/json_parse_float/Gemfile b/benchmarks/json_parse_float/Gemfile
similarity index 100%
rename from benchmarks-ractor/json_parse_float/Gemfile
rename to benchmarks/json_parse_float/Gemfile
diff --git a/benchmarks-ractor/json_parse_float/Gemfile.lock b/benchmarks/json_parse_float/Gemfile.lock
similarity index 100%
rename from benchmarks-ractor/json_parse_float/Gemfile.lock
rename to benchmarks/json_parse_float/Gemfile.lock
diff --git a/benchmarks-ractor/json_parse_float/benchmark.rb b/benchmarks/json_parse_float/benchmark.rb
similarity index 100%
rename from benchmarks-ractor/json_parse_float/benchmark.rb
rename to benchmarks/json_parse_float/benchmark.rb
diff --git a/benchmarks-ractor/json_parse_string/Gemfile b/benchmarks/json_parse_string/Gemfile
similarity index 100%
rename from benchmarks-ractor/json_parse_string/Gemfile
rename to benchmarks/json_parse_string/Gemfile
diff --git a/benchmarks-ractor/json_parse_string/Gemfile.lock b/benchmarks/json_parse_string/Gemfile.lock
similarity index 100%
rename from benchmarks-ractor/json_parse_string/Gemfile.lock
rename to benchmarks/json_parse_string/Gemfile.lock
diff --git a/benchmarks-ractor/json_parse_string/benchmark.rb b/benchmarks/json_parse_string/benchmark.rb
similarity index 100%
rename from benchmarks-ractor/json_parse_string/benchmark.rb
rename to benchmarks/json_parse_string/benchmark.rb
diff --git a/benchmarks-ractor/knucleotide/benchmark.rb b/benchmarks/knucleotide-ractor/benchmark.rb
similarity index 71%
rename from benchmarks-ractor/knucleotide/benchmark.rb
rename to benchmarks/knucleotide-ractor/benchmark.rb
index ca5b9cd9..41d265d6 100644
--- a/benchmarks-ractor/knucleotide/benchmark.rb
+++ b/benchmarks/knucleotide-ractor/benchmark.rb
@@ -2,9 +2,11 @@
 # https://salsa.debian.org/benchmarksgame-team/benchmarksgame/
 #
 # k-nucleotide benchmark - Ractor implementation
-# Mirrors the Process.fork version structure as closely as possible
+# Mirrors the Process.fork version: spawns 7 ractors (one per task)
 
-require_relative "../../harness/loader"
+Warning[:experimental] = false
+
+require_relative '../../harness/loader'
 
 def frequency(seq, length)
   frequencies = Hash.new(0)
@@ -25,9 +27,9 @@ def sort_by_freq(seq, length)
   table.sort { |a, b|
     cmp = b[1] <=> a[1]
     cmp == 0 ? a[0] <=> b[0] : cmp
-  }.map! { |seq, count|
+  }.map { |seq, count|
     "#{seq} #{'%.3f' % ((count * 100.0) / n)}"
-  }.join("\n") << "\n\n"
+  }.join("\n") + "\n\n"
 end
 
 def find_seq(seq, s)
@@ -48,19 +50,22 @@ def generate_test_sequence(size)
   full_copies.times { sequence << alu }
   sequence << alu[0, remainder] if remainder > 0
 
-  sequence.upcase.freeze
+  sequence.upcase
 end
 
-# Make sequence shareable for Ractors
-TEST_SEQUENCE = make_shareable(generate_test_sequence(100_000))
+TEST_SEQUENCE = Ractor.make_shareable(generate_test_sequence(100_000))
 
-run_benchmark(5) do |num_ractors, ractor_args|
+run_benchmark(5) do
   freqs = [1, 2]
   nucleos = %w(GGT GGTA GGTATT GGTATTTTAATT GGTATTTTAATTTATAGT)
 
-  # Sequential version - mirrors Process version but without Workers
-  results = []
-  freqs.each { |i| results << sort_by_freq(TEST_SEQUENCE, i) }
-  nucleos.each { |s| results << find_seq(TEST_SEQUENCE, s) }
+  ractors = freqs.map { |i|
+    Ractor.new(TEST_SEQUENCE, i) { |seq, len| sort_by_freq(seq, len) }
+  }
+  ractors += nucleos.map { |s|
+    Ractor.new(TEST_SEQUENCE, s) { |seq, nucleo| find_seq(seq, nucleo) }
+  }
+
+  results = ractors.map(&:value)
   results
 end
diff --git a/burn_in.rb b/burn_in.rb
index ea305eb8..91b1ada1 100755
--- a/burn_in.rb
+++ b/burn_in.rb
@@ -59,14 +59,10 @@ def free_file_path(parent_dir, name_prefix)
   end
 end
 
-def run_benchmark(bench_id, no_yjit, logs_path, run_time, ruby_version)
-  # Determine the path to the benchmark script
-  bench_name = bench_id.sub('ractor/', '')
-  bench_dir, harness = if bench_name == bench_id
-    ['benchmarks', 'harness']
-  else
-    ['benchmarks-ractor', 'harness-ractor']
-  end
+def run_benchmark(bench_name, no_yjit, logs_path, run_time, ruby_version, metadata)
+  bench_dir = 'benchmarks'
+  entry = metadata[bench_name] || {}
+  harness = entry.fetch('default_harness', 'harness')
 
   script_path = File.join(bench_dir, bench_name, 'benchmark.rb')
   if not File.exist?(script_path)
@@ -153,12 +149,12 @@ def run_benchmark(bench_id, no_yjit, logs_path, run_time, ruby_version)
   return false
 end
 
-def test_loop(bench_names, no_yjit, logs_path, run_time, ruby_version)
+def test_loop(bench_names, no_yjit, logs_path, run_time, ruby_version, metadata)
   error_found = false
 
   while true
     bench_name = bench_names.sample()
-    error = run_benchmark(bench_name, no_yjit, logs_path, run_time, ruby_version)
+    error = run_benchmark(bench_name, no_yjit, logs_path, run_time, ruby_version, metadata)
     error_found ||= error
 
     if error_found
@@ -201,12 +197,16 @@ def test_loop(bench_names, no_yjit, logs_path, run_time, ruby_version)
 bench_names = []
 
 if args.categories.include?('ractor-only')
-  # Only include benchmarks with ractor/ prefix (from benchmarks-ractor directory)
-  bench_names = metadata.keys.select { |name| name.start_with?('ractor/') }
+  # Include only benchmarks with ractor_only: true
+  metadata.each do |name, entry|
+    if entry['ractor_only']
+      bench_names << name
+    end
+  end
 elsif args.categories.include?('ractor')
-  # Include both ractor/ prefixed benchmarks and those with ractor: true
+  # Include benchmarks with ractor: true or ractor_only: true
   metadata.each do |name, entry|
-    if name.start_with?('ractor/') || entry['ractor']
+    if entry['ractor'] || entry['ractor_only']
       bench_names << name
     end
   end
@@ -221,10 +221,12 @@ def test_loop(bench_names, no_yjit, logs_path, run_time, ruby_version)
     end
   end
 else
-  # Regular category filtering
+  # Regular category filtering - exclude ractor-only and ractor harness benchmarks
   metadata.each do |name, entry|
     category = entry.fetch('category', 'other')
-    if args.categories.include?(category)
+    is_ractor_only = entry['ractor_only'] ||
+      (entry['ractor'] && entry['default_harness'] == 'harness-ractor')
+    if args.categories.include?(category) && !is_ractor_only
       bench_names << name
     end
   end
@@ -237,7 +239,7 @@ def test_loop(bench_names, no_yjit, logs_path, run_time, ruby_version)
 args.num_procs.times do |i|
   pid = Process.fork do
     run_time = (i < args.num_long_runs)? (3600 * 2):10
-    test_loop(bench_names, args.no_yjit, args.logs_path, run_time, ruby_version)
+    test_loop(bench_names, args.no_yjit, args.logs_path, run_time, ruby_version, metadata)
   end
 end
 
diff --git a/lib/argument_parser.rb b/lib/argument_parser.rb
index 35de87c1..e7a4cf4e 100644
--- a/lib/argument_parser.rb
+++ b/lib/argument_parser.rb
@@ -8,6 +8,7 @@ class ArgumentParser
     :out_path,
     :out_override,
     :harness,
+    :harness_explicit,
     :yjit_opts,
     :categories,
     :name_filters,
@@ -15,6 +16,7 @@ class ArgumentParser
     :rss,
     :graph,
     :no_pinning,
+    :force_pinning,
     :turbo,
     :skip_yjit,
     :with_pre_init,
@@ -52,10 +54,8 @@ def parse(argv)
             name = name.shellsplit.first
           end
           version, *options = version.shellsplit
-          rubies_dir = ENV["RUBIES_DIR"] || "#{ENV["HOME"]}/.rubies"
-          unless executable = ["/opt/rubies/#{version}/bin/ruby", "#{rubies_dir}/#{version}/bin/ruby"].find { |path| File.executable?(path) }
-            abort "Cannot find '#{version}' in /opt/rubies or #{rubies_dir}"
-          end
+          executable = find_chruby_ruby(version)
+          abort "Cannot find '#{version}' in chruby paths" unless executable
           args.executables[name] = [executable, *options]
         end
       end
@@ -94,6 +94,7 @@ def parse(argv)
       opts.on("--harness=HARNESS_DIR", "which harness to use") do |v|
         v = "harness-#{v}" unless v.start_with?('harness')
         args.harness = v
+        args.harness_explicit = true
       end
 
       opts.on("--warmup=N", "the number of warmup iterations for the default harness (default: 15)") do |n|
@@ -140,6 +141,10 @@ def parse(argv)
         args.no_pinning = true
       end
 
+      opts.on("--force-pinning", "force pinning even for benchmarks marked no_pinning") do
+        args.force_pinning = true
+      end
+
       opts.on("--turbo", "don't disable CPU turbo boost") do
         args.turbo = true
       end
@@ -165,6 +170,15 @@ def parse(argv)
 
   private
 
+  def find_chruby_ruby(version)
+    rubies_dir = ENV["RUBIES_DIR"] || "#{ENV["HOME"]}/.rubies"
+    chruby_search_paths(version, rubies_dir).find { |path| File.executable?(path) }
+  end
+
+  def chruby_search_paths(version, rubies_dir)
+    ["/opt/rubies/#{version}/bin/ruby", "#{rubies_dir}/#{version}/bin/ruby"]
+  end
+
   def have_yjit?(ruby)
     ruby_version = `#{ruby} -v --yjit 2> #{File::NULL}`.strip
     ruby_version.downcase.include?("yjit")
@@ -176,6 +190,7 @@ def default_args
       out_path: File.expand_path("./data"),
       out_override: nil,
       harness: "harness",
+      harness_explicit: false,
       yjit_opts: "",
       categories: [],
       name_filters: [],
@@ -183,6 +198,7 @@ def default_args
       rss: false,
       graph: false,
       no_pinning: false,
+      force_pinning: false,
       turbo: false,
       skip_yjit: false,
       with_pre_init: nil,
diff --git a/lib/benchmark_filter.rb b/lib/benchmark_filter.rb
index 3219d711..a9ed5586 100644
--- a/lib/benchmark_filter.rb
+++ b/lib/benchmark_filter.rb
@@ -18,12 +18,21 @@ def match?(name)
   private
 
   def matches_category?(name)
-    return true if @categories.empty?
+    if @categories.empty?
+      return false if ractor_harness_benchmark?(name)
+      return true
+    end
 
     benchmark_categories = get_benchmark_categories(name)
     @categories.intersect?(benchmark_categories)
   end
 
+  def ractor_harness_benchmark?(name)
+    benchmark_metadata = @metadata[name] || {}
+    benchmark_metadata['ractor_only'] ||
+      (benchmark_metadata['ractor'] && benchmark_metadata['default_harness'] == 'harness-ractor')
+  end
+
   def matches_name_filter?(name)
     return true if @name_filters.empty?
 
@@ -58,7 +67,8 @@ def get_benchmark_categories(name)
     @category_cache[name] ||= begin
       benchmark_metadata = @metadata[name] || {}
       categories = [benchmark_metadata.fetch('category', 'other')]
-      categories << 'ractor' if benchmark_metadata['ractor']
+      categories << 'ractor' if benchmark_metadata['ractor'] || benchmark_metadata['ractor_only']
+      categories << 'ractor-only' if benchmark_metadata['ractor_only']
       categories
     end
   end
diff --git a/lib/benchmark_runner/cli.rb b/lib/benchmark_runner/cli.rb
index 92206a48..d4c011c2 100644
--- a/lib/benchmark_runner/cli.rb
+++ b/lib/benchmark_runner/cli.rb
@@ -34,8 +34,10 @@ def run
         excludes: args.excludes,
         out_path: args.out_path,
         harness: args.harness,
+        harness_explicit: args.harness_explicit,
         pre_init: args.with_pre_init,
-        no_pinning: args.no_pinning
+        no_pinning: args.no_pinning,
+        force_pinning: args.force_pinning
       )
 
       # Benchmark with and without YJIT
diff --git a/lib/benchmark_suite.rb b/lib/benchmark_suite.rb
index 30c4a853..fd12eddc 100644
--- a/lib/benchmark_suite.rb
+++ b/lib/benchmark_suite.rb
@@ -14,24 +14,23 @@
 # BenchmarkSuite runs a collection of benchmarks and collects their results
 class BenchmarkSuite
   BENCHMARKS_DIR = "benchmarks"
-  RACTOR_BENCHMARKS_DIR = "benchmarks-ractor"
-  RACTOR_ONLY_CATEGORY = ["ractor-only"].freeze
   RACTOR_CATEGORY = ["ractor"].freeze
+  RACTOR_ONLY_CATEGORY = ["ractor-only"].freeze
   RACTOR_HARNESS = "harness-ractor"
 
-  attr_reader :categories, :name_filters, :excludes, :out_path, :harness, :pre_init, :no_pinning, :bench_dir, :ractor_bench_dir
+  attr_reader :categories, :name_filters, :excludes, :out_path, :harness, :harness_explicit, :pre_init, :no_pinning, :force_pinning, :bench_dir
 
-  def initialize(categories:, name_filters:, excludes: [], out_path:, harness:, pre_init: nil, no_pinning: false)
+  def initialize(categories:, name_filters:, excludes: [], out_path:, harness:, harness_explicit: false, pre_init: nil, no_pinning: false, force_pinning: false)
     @categories = categories
     @name_filters = name_filters
     @excludes = excludes
     @out_path = out_path
     @harness = harness
+    @harness_explicit = harness_explicit
     @pre_init = pre_init ? expand_pre_init(pre_init) : nil
     @no_pinning = no_pinning
-    @ractor_only = (categories == RACTOR_ONLY_CATEGORY)
-
-    setup_benchmark_directories
+    @force_pinning = force_pinning
+    @bench_dir = BENCHMARKS_DIR
   end
 
   # Run all the benchmarks and record execution times
@@ -41,14 +40,14 @@ def run(ruby:, ruby_description:)
     bench_failures = {}
 
     benchmark_entries = discover_benchmarks
-    cmd_prefix = base_cmd(ruby_description)
     env = benchmark_env(ruby)
 
     benchmark_entries.each_with_index do |entry, idx|
       puts("Running benchmark \"#{entry.name}\" (#{idx+1}/#{benchmark_entries.length})")
 
       result_json_path = File.join(out_path, "temp#{Process.pid}.json")
-      result = run_single_benchmark(entry.script_path, result_json_path, ruby, cmd_prefix, env)
+      cmd_prefix = base_cmd(ruby_description, entry.name)
+      result = run_single_benchmark(entry.script_path, result_json_path, ruby, cmd_prefix, env, entry.name)
 
       if result[:success]
         bench_data[entry.name] = process_benchmark_result(result_json_path, result[:command])
@@ -62,18 +61,6 @@ def run(ruby:, ruby_description:)
 
   private
 
-  def setup_benchmark_directories
-    if @ractor_only
-      @bench_dir = RACTOR_BENCHMARKS_DIR
-      @ractor_bench_dir = RACTOR_BENCHMARKS_DIR
-      @harness = RACTOR_HARNESS
-      @categories = []
-    else
-      @bench_dir = BENCHMARKS_DIR
-      @ractor_bench_dir = RACTOR_BENCHMARKS_DIR
-    end
-  end
-
   def process_benchmark_result(result_json_path, command)
     JSON.parse(File.read(result_json_path)).tap do |json|
       json["command_line"] = command
@@ -88,47 +75,24 @@ def discover_benchmarks
   end
 
   def discover_all_benchmark_entries
-    main_discovery = BenchmarkDiscovery.new(bench_dir)
-    main_entries = main_discovery.discover
-
-    ractor_entries = if benchmark_ractor_directory?
-      ractor_discovery = BenchmarkDiscovery.new(ractor_bench_dir)
-      ractor_discovery.discover
-    else
-      []
-    end
-
-    { main: main_entries, ractor: ractor_entries }
+    discovery = BenchmarkDiscovery.new(bench_dir)
+    { main: discovery.discover }
   end
 
   def build_directory_map(all_entries)
-    combined_entries = all_entries[:main] + all_entries[:ractor]
-    combined_entries.each_with_object({}) do |entry, map|
+    all_entries[:main].each_with_object({}) do |entry, map|
       map[entry.name] = entry.directory
     end
   end
 
   def filter_benchmarks(all_entries, directory_map)
-    main_benchmarks = filter_entries(
+    filter_entries(
       all_entries[:main],
       categories: categories,
       name_filters: name_filters,
       excludes: excludes,
       directory_map: directory_map
     )
-
-    if benchmark_ractor_directory?
-      ractor_benchmarks = filter_entries(
-        all_entries[:ractor],
-        categories: [],
-        name_filters: name_filters,
-        excludes: excludes,
-        directory_map: directory_map
-      )
-      main_benchmarks + ractor_benchmarks
-    else
-      main_benchmarks
-    end
   end
 
   def filter_entries(entries, categories:, name_filters:, excludes:, directory_map:)
@@ -142,17 +106,20 @@ def filter_entries(entries, categories:, name_filters:, excludes:, directory_map
     entries.select { |entry| filter.match?(entry.name) }
   end
 
-  def run_single_benchmark(script_path, result_json_path, ruby, cmd_prefix, env)
+  def run_single_benchmark(script_path, result_json_path, ruby, cmd_prefix, env, benchmark_name)
     # Fix for jruby/jruby#7394 in JRuby 9.4.2.0
     script_path = File.expand_path(script_path)
 
     # Set up the environment for the benchmarking command
     ENV["RESULT_JSON_PATH"] = result_json_path
 
+    # Use per-benchmark default_harness if set, otherwise use global harness
+    benchmark_harness = benchmark_harness_for(benchmark_name)
+
     # Set up the benchmarking command
     cmd = cmd_prefix + [
       *ruby,
-      "-I", harness,
+      "-I", benchmark_harness,
       *pre_init,
       script_path,
     ].compact
@@ -163,6 +130,14 @@ def run_single_benchmark(script_path, result_json_path, ruby, cmd_prefix, env)
     result
   end
 
+  def benchmark_harness_for(benchmark_name)
+    return harness if harness_explicit
+
+    benchmark_meta = benchmarks_metadata[benchmark_name] || {}
+    default = ractor_category_run? ? RACTOR_HARNESS : harness
+    benchmark_meta.fetch('default_harness', default)
+  end
+
   def benchmark_env(ruby)
     # When the Ruby running this script is not the first Ruby in PATH, shell commands
     # like `bundle install` in a child process will not use the Ruby being benchmarked.
@@ -187,23 +162,19 @@ def benchmarks_metadata
     @benchmarks_metadata ||= YAML.load_file('benchmarks.yml')
   end
 
-  def benchmark_ractor_directory?
-    categories == RACTOR_CATEGORY
-  end
-
   # Check if running on Linux
   def linux?
     @linux ||= RbConfig::CONFIG['host_os'] =~ /linux/
   end
 
   # Set up the base command with CPU pinning if needed
-  def base_cmd(ruby_description)
+  def base_cmd(ruby_description, benchmark_name)
     if linux?
       cmd = setarch_prefix
 
       # Pin the process to one given core to improve caching and reduce variance on CRuby
       # Other Rubies need to use multiple cores, e.g., for JIT threads
-      if ruby_description.start_with?('ruby ') && !no_pinning
+      if ruby_description.start_with?('ruby ') && should_pin?(benchmark_name)
         # The last few cores of Intel CPU may be slow E-Cores, so avoid using the last one.
         cpu = [(Etc.nprocessors / 2) - 1, 0].max
         cmd.concat(["taskset", "-c", "#{cpu}"])
@@ -215,6 +186,19 @@ def base_cmd(ruby_description)
     end
   end
 
+  def should_pin?(benchmark_name)
+    return false if no_pinning
+    return true if force_pinning
+    return false if ractor_category_run?
+
+    benchmark_meta = benchmarks_metadata[benchmark_name] || {}
+    !benchmark_meta["no_pinning"]
+  end
+
+  def ractor_category_run?
+    categories == RACTOR_CATEGORY || categories == RACTOR_ONLY_CATEGORY
+  end
+
   # Generate setarch prefix for Linux
   def setarch_prefix
     # Disable address space randomization (for determinism)
diff --git a/test/argument_parser_test.rb b/test/argument_parser_test.rb
index 7008dcef..d19ae240 100644
--- a/test/argument_parser_test.rb
+++ b/test/argument_parser_test.rb
@@ -103,15 +103,11 @@ def setup_mock_ruby(path)
           ruby_path = File.join(tmpdir, 'opt/rubies/3.2.0/bin/ruby')
           setup_mock_ruby(ruby_path)
 
-          File.stub :executable?, ->(path) {
-            if path == "/opt/rubies/3.2.0/bin/ruby"
-              File.exist?(ruby_path) && File.stat(ruby_path).executable?
-            end
-          } do
-            parser = ArgumentParser.new
+          parser = ArgumentParser.new
+          parser.stub :chruby_search_paths, ->(version, rubies_dir) { [ruby_path] } do
             args = parser.parse(['--chruby=ruby-3.2.0::3.2.0'])
 
-            assert_equal '/opt/rubies/3.2.0/bin/ruby', args.executables['ruby-3.2.0'].first
+            assert_equal ruby_path, args.executables['ruby-3.2.0'].first
           end
         end
       end
@@ -126,9 +122,11 @@ def setup_mock_ruby(path)
           ENV['HOME'] = tmpdir
 
           parser = ArgumentParser.new
-          args = parser.parse(['--chruby=my-ruby::3.3.0'])
+          parser.stub :chruby_search_paths, ->(version, rd) { ["#{rd}/#{version}/bin/ruby"] } do
+            args = parser.parse(['--chruby=my-ruby::3.3.0'])
 
-          assert_equal ruby_path, args.executables['my-ruby'].first
+            assert_equal ruby_path, args.executables['my-ruby'].first
+          end
         end
       end
 
@@ -143,20 +141,11 @@ def setup_mock_ruby(path)
 
           ENV['HOME'] = tmpdir
 
-          File.stub :executable?, ->(path) {
-            case path
-            when "/opt/rubies/3.2.0/bin/ruby"
-              File.exist?(opt_ruby) && File.stat(opt_ruby).executable?
-            when "#{tmpdir}/.rubies/3.2.0/bin/ruby"
-              File.exist?(home_ruby) && File.stat(home_ruby).executable?
-            else
-              File.method(:executable?).super_method.call(path)
-            end
-          } do
-            parser = ArgumentParser.new
+          parser = ArgumentParser.new
+          parser.stub :chruby_search_paths, ->(version, rd) { [opt_ruby, home_ruby] } do
             args = parser.parse(['--chruby=test::3.2.0'])
 
-            assert_equal '/opt/rubies/3.2.0/bin/ruby', args.executables['test'].first
+            assert_equal opt_ruby, args.executables['test'].first
           end
         end
       end
@@ -171,9 +160,11 @@ def setup_mock_ruby(path)
           ENV['RUBIES_DIR'] = custom_rubies
 
           parser = ArgumentParser.new
-          args = parser.parse(['--chruby=custom::3.4.0'])
+          parser.stub :chruby_search_paths, ->(version, rd) { ["#{rd}/#{version}/bin/ruby"] } do
+            args = parser.parse(['--chruby=custom::3.4.0'])
 
-          assert_equal ruby_path, args.executables['custom'].first
+            assert_equal ruby_path, args.executables['custom'].first
+          end
         end
       end
 
@@ -183,10 +174,11 @@ def setup_mock_ruby(path)
           ENV['HOME'] = tmpdir
 
           parser = ArgumentParser.new
-
-          assert_raises(SystemExit) do
-            capture_io do
-              parser.parse(['--chruby=nonexistent::nonexistent-version-999'])
+          parser.stub :chruby_search_paths, ->(version, rd) { ["#{rd}/#{version}/bin/ruby"] } do
+            assert_raises(SystemExit) do
+              capture_io do
+                parser.parse(['--chruby=nonexistent::nonexistent-version-999'])
+              end
             end
           end
         end
@@ -202,10 +194,12 @@ def setup_mock_ruby(path)
           ENV['HOME'] = tmpdir
 
           parser = ArgumentParser.new
-          args = parser.parse(['--chruby=yjit::3.2.0 --yjit'])
+          parser.stub :chruby_search_paths, ->(version, rd) { ["#{rd}/#{version}/bin/ruby"] } do
+            args = parser.parse(['--chruby=yjit::3.2.0 --yjit'])
 
-          assert_equal ruby_path, args.executables['yjit'].first
-          assert_equal '--yjit', args.executables['yjit'].last
+            assert_equal ruby_path, args.executables['yjit'].first
+            assert_equal '--yjit', args.executables['yjit'].last
+          end
         end
       end
 
@@ -219,10 +213,12 @@ def setup_mock_ruby(path)
           ENV['HOME'] = tmpdir
 
           parser = ArgumentParser.new
-          args = parser.parse(['--chruby=3.2.0 --yjit'])
+          parser.stub :chruby_search_paths, ->(version, rd) { ["#{rd}/#{version}/bin/ruby"] } do
+            args = parser.parse(['--chruby=3.2.0 --yjit'])
 
-          assert args.executables.key?('3.2.0')
-          assert_equal ruby_path, args.executables['3.2.0'].first
+            assert args.executables.key?('3.2.0')
+            assert_equal ruby_path, args.executables['3.2.0'].first
+          end
         end
       end
 
@@ -238,12 +234,14 @@ def setup_mock_ruby(path)
           ENV['HOME'] = tmpdir
 
           parser = ArgumentParser.new
-          args = parser.parse(['--chruby=ruby32::3.2.0;ruby33::3.3.0 --yjit'])
+          parser.stub :chruby_search_paths, ->(version, rd) { ["#{rd}/#{version}/bin/ruby"] } do
+            args = parser.parse(['--chruby=ruby32::3.2.0;ruby33::3.3.0 --yjit'])
 
-          assert_equal 2, args.executables.size
-          assert_equal ruby_path_32, args.executables['ruby32'].first
-          assert_equal ruby_path_33, args.executables['ruby33'].first
-          assert_equal '--yjit', args.executables['ruby33'].last
+            assert_equal 2, args.executables.size
+            assert_equal ruby_path_32, args.executables['ruby32'].first
+            assert_equal ruby_path_33, args.executables['ruby33'].first
+            assert_equal '--yjit', args.executables['ruby33'].last
+          end
         end
       end
     end
@@ -589,10 +587,12 @@ def setup_mock_ruby(path)
           parser = ArgumentParser.new(ruby_executable: mock_ruby)
 
           parser.stub :have_yjit?, true do
-            args = parser.parse(['--chruby=test::3.2.0'])
+            parser.stub :chruby_search_paths, ->(version, rd) { ["#{rd}/#{version}/bin/ruby"] } do
+              args = parser.parse(['--chruby=test::3.2.0'])
 
-            assert_equal 1, args.executables.size
-            assert_equal ruby_path, args.executables['test'].first
+              assert_equal 1, args.executables.size
+              assert_equal ruby_path, args.executables['test'].first
+            end
           end
         end
       end
diff --git a/test/benchmark_filter_test.rb b/test/benchmark_filter_test.rb
index 5b623d10..1abdfada 100644
--- a/test/benchmark_filter_test.rb
+++ b/test/benchmark_filter_test.rb
@@ -47,6 +47,23 @@
       assert_equal true, filter.match?('ractor_bench')
     end
 
+    it 'excludes ractor harness benchmarks from default runs' do
+      metadata = @metadata.merge('ractor_harness_bench' => { 'category' => 'other', 'ractor' => true, 'default_harness' => 'harness-ractor' })
+      filter = BenchmarkFilter.new(categories: [], name_filters: [], excludes: [], metadata: metadata)
+
+      assert_equal true, filter.match?('fib')
+      assert_equal true, filter.match?('ractor_bench') # ractor: true without harness-ractor runs in default
+      assert_equal false, filter.match?('ractor_harness_bench') # ractor: true with harness-ractor excluded
+    end
+
+    it 'includes ractor harness benchmarks in ractor category' do
+      metadata = @metadata.merge('ractor_harness_bench' => { 'category' => 'other', 'ractor' => true, 'default_harness' => 'harness-ractor' })
+      filter = BenchmarkFilter.new(categories: ['ractor'], name_filters: [], excludes: [], metadata: metadata)
+
+      assert_equal true, filter.match?('ractor_harness_bench')
+      assert_equal true, filter.match?('ractor_bench')
+    end
+
     it 'handles regex filters' do
       filter = BenchmarkFilter.new(categories: [], name_filters: ['/rails/'], excludes: [], metadata: @metadata)
 
diff --git a/test/benchmark_suite_test.rb b/test/benchmark_suite_test.rb
index cd9eee3b..91443987 100644
--- a/test/benchmark_suite_test.rb
+++ b/test/benchmark_suite_test.rb
@@ -14,7 +14,6 @@
 
     # Create mock benchmarks directory structure
     FileUtils.mkdir_p('benchmarks')
-    FileUtils.mkdir_p('benchmarks-ractor')
     FileUtils.mkdir_p('harness')
 
     # Create a simple benchmark file
@@ -73,7 +72,7 @@
       assert_equal true, suite.no_pinning
     end
 
-    it 'sets bench_dir to BENCHMARKS_DIR by default' do
+    it 'sets bench_dir to BENCHMARKS_DIR' do
       suite = BenchmarkSuite.new(
         categories: ['micro'],
         name_filters: [],
@@ -82,37 +81,100 @@
       )
 
       assert_equal 'benchmarks', suite.bench_dir
-      assert_equal 'benchmarks-ractor', suite.ractor_bench_dir
       assert_equal 'harness', suite.harness
       assert_equal ['micro'], suite.categories
     end
 
-    it 'sets bench_dir to ractor directory and updates harness when ractor-only category is used' do
+    it 'keeps bench_dir as BENCHMARKS_DIR when ractor category is used' do
       suite = BenchmarkSuite.new(
-        categories: ['ractor-only'],
+        categories: ['ractor'],
         name_filters: [],
         out_path: @out_path,
         harness: 'harness'
       )
 
-      assert_equal 'benchmarks-ractor', suite.bench_dir
-      assert_equal 'benchmarks-ractor', suite.ractor_bench_dir
-      assert_equal 'harness-ractor', suite.harness
-      assert_equal [], suite.categories
+      assert_equal 'benchmarks', suite.bench_dir
+      assert_equal 'harness', suite.harness
+      assert_equal ['ractor'], suite.categories
     end
 
-    it 'keeps bench_dir as BENCHMARKS_DIR when ractor category is used' do
+    it 'tracks harness_explicit flag' do
+      suite_explicit = BenchmarkSuite.new(
+        categories: [],
+        name_filters: [],
+        out_path: @out_path,
+        harness: 'custom-harness',
+        harness_explicit: true
+      )
+      assert_equal true, suite_explicit.harness_explicit
+
+      suite_auto = BenchmarkSuite.new(
+        categories: [],
+        name_filters: [],
+        out_path: @out_path,
+        harness: 'harness-ractor'
+      )
+      assert_equal false, suite_auto.harness_explicit
+    end
+  end
+
+  describe '#benchmark_harness_for' do
+    before do
+      @metadata_with_harness = {
+        'simple' => { 'category' => 'micro' },
+        'custom_harness_bench' => { 'category' => 'other', 'default_harness' => 'harness' }
+      }
+      File.write('benchmarks.yml', YAML.dump(@metadata_with_harness))
+    end
+
+    it 'returns default_harness when set and harness not explicit' do
+      suite = BenchmarkSuite.new(
+        categories: [],
+        name_filters: [],
+        out_path: @out_path,
+        harness: 'harness-ractor',
+        harness_explicit: false
+      )
+
+      assert_equal 'harness', suite.send(:benchmark_harness_for, 'custom_harness_bench')
+      assert_equal 'harness-ractor', suite.send(:benchmark_harness_for, 'simple')
+    end
+
+    it 'ignores default_harness when harness is explicit' do
+      suite = BenchmarkSuite.new(
+        categories: [],
+        name_filters: [],
+        out_path: @out_path,
+        harness: 'custom-harness',
+        harness_explicit: true
+      )
+
+      assert_equal 'custom-harness', suite.send(:benchmark_harness_for, 'custom_harness_bench')
+      assert_equal 'custom-harness', suite.send(:benchmark_harness_for, 'simple')
+    end
+  end
+
+  describe '#ractor_category_run?' do
+    it 'returns true for ractor category' do
       suite = BenchmarkSuite.new(
         categories: ['ractor'],
         name_filters: [],
         out_path: @out_path,
+        harness: 'harness-ractor'
+      )
+
+      assert_equal true, suite.send(:ractor_category_run?)
+    end
+
+    it 'returns false for other categories' do
+      suite = BenchmarkSuite.new(
+        categories: ['micro'],
+        name_filters: [],
+        out_path: @out_path,
         harness: 'harness'
       )
 
-      assert_equal 'benchmarks', suite.bench_dir
-      assert_equal 'benchmarks-ractor', suite.ractor_bench_dir
-      assert_equal 'harness', suite.harness
-      assert_equal ['ractor'], suite.categories
+      assert_equal false, suite.send(:ractor_category_run?)
     end
   end
 
@@ -232,9 +294,9 @@
       assert_empty bench_failures
     end
 
-    it 'handles ractor-only category' do
+    it 'handles ractor category with ractor benchmarks' do
       # Create a ractor benchmark
-      File.write('benchmarks-ractor/ractor_test.rb', <<~RUBY)
+      File.write('benchmarks/ractor_test.rb', <<~RUBY)
         require 'json'
         result = {
           'warmup' => [0.001],
@@ -244,8 +306,14 @@
         File.write(ENV['RESULT_JSON_PATH'], JSON.generate(result))
       RUBY
 
+      metadata = {
+        'ractor_test' => { 'category' => 'other', 'ractor' => true },
+        'simple' => { 'category' => 'micro' }
+      }
+      File.write('benchmarks.yml', YAML.dump(metadata))
+
       suite = BenchmarkSuite.new(
-        categories: ['ractor-only'],
+        categories: ['ractor'],
         name_filters: [],
         out_path: @out_path,
         harness: 'harness',
@@ -257,41 +325,10 @@
         bench_data, bench_failures = suite.run(ruby: [RbConfig.ruby], ruby_description: 'ruby 3.2.0')
       end
 
-      # When ractor-only is specified, it should use benchmarks-ractor directory
+      # Should only include ractor benchmarks when ractor category specified
       assert_includes bench_data, 'ractor_test'
+      refute_includes bench_data, 'simple'
       assert_empty bench_failures
-
-      # harness should be updated to harness-ractor
-      assert_equal 'harness-ractor', suite.harness
-    end
-
-    it 'includes both regular and ractor benchmarks with ractor category' do
-      File.write('benchmarks-ractor/ractor_bench.rb', <<~RUBY)
-        require 'json'
-        result = {
-          'warmup' => [0.001],
-          'bench' => [0.001],
-          'rss' => 10485760
-        }
-        File.write(ENV['RESULT_JSON_PATH'], JSON.generate(result))
-      RUBY
-
-      suite = BenchmarkSuite.new(
-        categories: ['ractor'],
-        name_filters: [],
-        out_path: @out_path,
-        harness: 'harness',
-        no_pinning: true
-      )
-
-      bench_data = nil
-      capture_io do
-        bench_data, _ = suite.run(ruby: [RbConfig.ruby], ruby_description: 'ruby 3.2.0')
-      end
-
-      # With ractor category, both directories should be scanned
-      # but we need appropriate filters
-      assert_instance_of Hash, bench_data
     end
 
     it 'expands pre_init when provided' do