diff --git a/src/bench/builtin.rs b/src/bench/builtin.rs
index bbe6716..ed0d30c 100644
--- a/src/bench/builtin.rs
+++ b/src/bench/builtin.rs
@@ -1,19 +1,42 @@
 use std::sync::{Arc, Mutex};
+use std::thread::JoinHandle;
 
 use libpt::log::info;
 
 use crate::error::{BenchError, Error, WResult};
-use crate::game::{Game, GameBuilder};
+use crate::game::{self, Game, GameBuilder};
 use crate::solve::Solver;
 use crate::wlist::WordList;
 
 use super::{Benchmark, Report};
 
-#[derive(Debug, Clone)]
+use rayon::prelude::*;
+
+#[derive(Debug)]
 pub struct BuiltinBenchmark<'wl, WL: WordList, SL: Solver<'wl, WL>> {
     solver: SL,
     builder: GameBuilder<'wl, WL>,
     report: Arc<Mutex<Report>>,
+    benchth: Option<JoinHandle<()>>,
+}
+impl<'wl, WL, SL> BuiltinBenchmark<'wl, WL, SL>
+where
+    WL: WordList,
+    WL: 'wl,
+    SL: Solver<'wl, WL>,
+    SL: 'wl,
+    SL: Send,
+{
+    #[inline]
+    fn inner_bench(outside_data: (Arc<Mutex<Report>>, GameBuilder<'wl, WL>, SL), _index: usize) {
+        todo!()
+    }
+    fn run_bench(n: usize, report: Arc<Mutex<Report>>, builder: GameBuilder<'wl, WL>, solver: SL) {
+        (0..n)
+            .into_par_iter()
+            .map(|idx| Self::inner_bench((report.clone(), builder.clone(), solver.clone()), idx));
+        report.lock().expect("lock is poisoned").finalize();
+    }
 }
 
 impl<'wl, WL, SL> Benchmark<'wl, WL, SL> for BuiltinBenchmark<'wl, WL, SL>
@@ -22,25 +45,36 @@ where
     WL: 'wl,
     SL: Solver<'wl, WL>,
     SL: 'wl,
+    SL: Send,
 {
     fn build(
-        wordlist: &'wl WL,
+        _wordlist: &'wl WL,
         solver: SL,
         builder: GameBuilder<'wl, WL>,
-        threads: usize
+        threads: usize,
     ) -> crate::error::WResult<Self> {
         info!("using {threads} threads for benchmarking");
-        rayon::ThreadPoolBuilder::new().num_threads(threads).build_global().unwrap();
+        rayon::ThreadPoolBuilder::new()
+            .num_threads(threads)
+            .build_global()
+            .unwrap();
         Ok(Self {
             solver,
             report: Arc::new(Mutex::new(Report::new(builder.build()?))),
             builder,
+            benchth: None,
         })
     }
-    fn solver(&self) -> &SL {
+    fn solver(&self) -> SL {
+        self.solver.clone()
+    }
+    fn builder(&'wl self) -> game::GameBuilder<'wl, WL> {
+        self.builder.clone()
+    }
+    fn solver_ref(&'wl self) -> &'wl SL {
         &self.solver
     }
-    fn builder(&'wl self) -> &'wl crate::game::GameBuilder<'wl, WL> {
+    fn builder_ref(&'wl self) -> &'wl game::GameBuilder<'wl, WL> {
         &self.builder
     }
 
@@ -51,4 +85,21 @@ where
     fn report(&'wl self) -> super::Report {
         self.report.lock().expect("lock is poisoned").clone()
     }
+
+    fn start(&'wl mut self, n: usize) -> WResult<()> {
+        let report = self.report_mutex(); // FIXME: needs to keep self borrowed for some reason?
+        let solver = self.solver();
+        let builder = self.builder();
+        let benchth = std::thread::spawn(move || Self::run_bench(n, report, builder, solver));
+
+        self.benchth = Some(benchth);
+        Ok(())
+    }
+
+    fn is_finished(&self) -> Option<bool> {
+        match &self.benchth {
+            Some(th) => Some(th.is_finished()),
+            None => None,
+        }
+    }
 }
diff --git a/src/bench/mod.rs b/src/bench/mod.rs
index ae0234e..658aa1b 100644
--- a/src/bench/mod.rs
+++ b/src/bench/mod.rs
@@ -19,7 +19,7 @@ pub mod builtin;
 /// Default amount of games to play for a [Benchmark]
 pub const DEFAULT_N: usize = 50;
 
-pub trait Benchmark<'wl, WL, SL>: Clone + Sized + Debug + Sync
+pub trait Benchmark<'wl, WL, SL>: Sized + Debug + Sync
 where
     WL: WordList,
     WL: 'wl,
@@ -30,18 +30,24 @@ where
         wordlist: &'wl WL,
         solver: SL,
         builder: GameBuilder<'wl, WL>,
-        threads: usize
+        threads: usize,
     ) -> crate::error::WResult<Self>;
-    fn builder(&'wl self) -> &'wl GameBuilder<'wl, WL>;
+    fn builder(&'wl self) -> GameBuilder<'wl, WL>;
+    fn builder_ref(&'wl self) -> &'wl GameBuilder<'wl, WL>;
     fn make_game(&'wl self) -> WResult<Game<'wl, WL>> {
-        Ok(self.builder().build()?)
+        Ok(self.builder_ref().build()?)
     }
-    fn solver(&'wl self) -> &'wl SL;
+    fn solver(&'wl self) -> SL;
+    fn solver_ref(&'wl self) -> &'wl SL;
     fn play(&'wl self) -> WResult<GuessResponse> {
-        self.solver().play(&mut self.make_game()?)
+        self.solver_ref().play(&mut self.make_game()?)
     }
+    fn start(&'wl mut self, n: usize) -> WResult<()>;
+    fn is_finished(&self) -> Option<bool>;
     // TODO: add some interface to get reports while the benchmark runs
     // TODO: make the benchmark optionally multithreaded
+    // NOTE: This is blocking, use start to let it run in another thread
+    #[deprecated]
     fn bench(&'wl self, n: usize) -> WResult<Report> {
         let report = self.report_mutex();
         let this = std::sync::Arc::new(self);
diff --git a/src/bench/report.rs b/src/bench/report.rs
index ad3ad31..c6b0c58 100644
--- a/src/bench/report.rs
+++ b/src/bench/report.rs
@@ -142,8 +142,12 @@ impl Display for Report {
     fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
         write!(
             f,
-            "n: {}, win_ratio: {:.2}%, avg_score: {:.4} steps until finish, avgerage time per game: {}μs, \n\
-            rating: {:.4}, full time until completion: {}ms
+            "n: {}\n\
+            win_ratio: {:.2}%\n\
+            avg_score: {:.4}\n\
+            avgerage time per game: {}μs\n\
+            full time until completion: {}ms\n\
+            rating: {:.4}\n\
             ",
             self.n(),
             self.avg_win() * 100.0,
diff --git a/src/bin/bench/cli.rs b/src/bin/bench/cli.rs
index e2772cc..0480bd7 100644
--- a/src/bin/bench/cli.rs
+++ b/src/bin/bench/cli.rs
@@ -11,7 +11,8 @@ use wordle_analyzer::bench::builtin::BuiltinBenchmark;
 use wordle_analyzer::bench::report::Report;
 use wordle_analyzer::bench::{Benchmark, DEFAULT_N};
 use wordle_analyzer::error::WResult;
-use wordle_analyzer::solve::{BuiltinSolverNames, Solver};
+use wordle_analyzer::game::GameBuilder;
+use wordle_analyzer::solve::{AnyBuiltinSolver, BuiltinSolverNames, Solver};
 use wordle_analyzer::wlist::builtin::BuiltinWList;
 
 use wordle_analyzer::{self, game};
@@ -54,24 +55,22 @@ fn main() -> anyhow::Result<()> {
     trace!("dumping CLI: {:#?}", cli);
 
     let wl = BuiltinWList::default();
-    let builder = game::Game::builder(&wl)
+    let builder: GameBuilder<'_, BuiltinWList> = game::Game::builder(&wl)
         .length(cli.length)
         .max_steps(cli.max_steps)
         .precompute(cli.precompute);
-    let solver = cli.solver.to_solver(&wl);
-    let bench = Arc::new(BuiltinBenchmark::build(&wl, solver, builder, cli.threads)?);
-    let bench_running = bench.clone();
+    let solver: AnyBuiltinSolver<'_, BuiltinWList> = cli.solver.to_solver(&wl);
+    let mut bench = BuiltinBenchmark::build(&wl, solver, builder, cli.threads)?;
     trace!("{bench:#?}");
     let n = cli.n;
-    let bench_th: std::thread::JoinHandle<WResult<Report>> =
-        std::thread::spawn(move || bench_running.bench(n));
+    bench.start(n)?;
 
-    while !bench_th.is_finished() {
+    while !bench.is_finished()? {
         println!("{}", bench.report());
     }
 
-    // finished report
-    println!("{}", bench_th.join().expect("thread go boom")?);
 
+    // FIXME: Rustc thinks wl is borrowed at this point, but it is not!!!!! Or at least it should
+    // not be
     Ok(())
 }