diff --git a/.gitignore b/.gitignore
index 9623fa5..9121a72 100644
--- a/.gitignore
+++ b/.gitignore
@@ -6,3 +6,4 @@
 /.psc*
 /.purs*
 /.psa*
+package-lock.json
diff --git a/.travis.yml b/.travis.yml
index fdb744f..5b14646 100644
--- a/.travis.yml
+++ b/.travis.yml
@@ -2,9 +2,22 @@ language: node_js
 dist: trusty
 sudo: required
 node_js: stable
+env:
+  - PATH=$HOME/purescript:$PATH
 install:
+  - TAG=$(wget -q -O - https://github.com/purescript/purescript/releases/latest --server-response --max-redirect 0 2>&1 | sed -n -e 's/.*Location:.*tag\///p')
+  - wget -O $HOME/purescript.tar.gz https://github.com/purescript/purescript/releases/download/$TAG/linux64.tar.gz
+  - tar -xvf $HOME/purescript.tar.gz -C $HOME/
+  - chmod a+x $HOME/purescript
   - npm install -g bower
   - npm install
-  - bower install
+  - bower install --production
 script:
   - npm run -s build
+  - bower install
+  - npm run -s test
+after_success:
+- >-
+  test $TRAVIS_TAG &&
+  echo $GITHUB_TOKEN | pulp login &&
+  echo y | pulp publish --no-push
diff --git a/package.json b/package.json
index 01da16a..5a4bff0 100644
--- a/package.json
+++ b/package.json
@@ -2,7 +2,8 @@
   "private": true,
   "scripts": {
     "clean": "rimraf output && rimraf .pulp-cache",
-    "build": "pulp build -- --censor-lib --strict"
+    "build": "pulp build -- --censor-lib --strict",
+    "test": "pulp test"
   },
   "devDependencies": {
     "pulp": "^12.2.0",
diff --git a/src/Performance/Minibench.purs b/src/Performance/Minibench.purs
index 638ad87..d7f678e 100644
--- a/src/Performance/Minibench.purs
+++ b/src/Performance/Minibench.purs
@@ -7,6 +7,9 @@
 module Performance.Minibench
   ( bench
   , benchWith
+  , benchWith'
+  , BenchResult
+  , withUnits
   ) where
 
 import Prelude hiding (min,max)
@@ -46,8 +49,31 @@ withUnits t
 -- |
 -- | To increase benchmark accuracy by forcing garbage collection before the
 -- | benchmark is run, node should be invoked with the '--expose-gc' flag.
-benchWith :: forall a. Int -> (Unit -> a) -> Effect Unit
+benchWith
+  :: forall a
+   . Int
+  -> (Unit -> a)
+  -> Effect Unit
 benchWith n f = do
+  res <- benchWith' n f
+  log ("mean   = " <> withUnits res.mean)
+  log ("stddev = " <> withUnits res.stdDev)
+  log ("min    = " <> withUnits res.min)
+  log ("max    = " <> withUnits res.max)
+
+type BenchResult =
+  { mean :: Number
+  , stdDev :: Number
+  , min :: Number
+  , max :: Number
+  }
+
+benchWith'
+  :: forall a
+   . Int
+  -> (Unit -> a)
+  -> Effect BenchResult
+benchWith' n f = do
   sumRef <- Ref.new 0.0
   sum2Ref <- Ref.new 0.0
   minRef <- Ref.new infinity
@@ -70,10 +96,12 @@ benchWith n f = do
   let n'     = toNumber n
       mean   = sum / n'
       stdDev = sqrt ((sum2 - n' * mean * mean) / (n' - 1.0))
-  log ("mean   = " <> withUnits mean)
-  log ("stddev = " <> withUnits stdDev)
-  log ("min    = " <> withUnits min')
-  log ("max    = " <> withUnits max')
+  pure
+    { mean
+    , stdDev
+    , min: min'
+    , max: max'
+    }
 
 -- | Estimate the running time of a function and print a summary to the console,
 -- | by running the function 1000 times.