diff --git a/.github/workflows/bench.yml b/.github/workflows/bench.yml
index 333580e4..1015edda 100644
--- a/.github/workflows/bench.yml
+++ b/.github/workflows/bench.yml
@@ -5,7 +5,7 @@ on:
   - cron: '36 1 * * SUN'  # M H d m w (Sundays at 01:36)
 jobs:
   asvfull:
-    # if: (github.event_name == 'push' && startsWith(github.event.ref, 'refs/tags')) || github.event_name == 'schedule'
+    if: (github.event_name == 'push' && startsWith(github.event.ref, 'refs/tags')) || github.event_name == 'schedule'
     name: Full
     runs-on: ubuntu-latest
     steps:
diff --git a/benchmarks/README.md b/benchmarks/README.md
new file mode 100644
index 00000000..a4e2e528
--- /dev/null
+++ b/benchmarks/README.md
@@ -0,0 +1,24 @@
+# Benchmarks
+
+These benchmarks serve two purposes:
+
+1. Thorough performance tests against regression
+    - `tqdm`
+    - `tqdm(miniters=manually_optimised, smoothing=0)`
+    - `no-progress` (empty loop without progress wrapper)
+2. Compare `tqdm`'s speed to popular alternatives
+    - [`rich.progress`](https://pypi.org/project/rich)
+    - [`progressbar2`](https://pypi.org/project/progressbar2)
+    - [`alive-progress`](https://pypi.org/project/alive-progress)
+
+Performance graphs are available at <https://tqdm.github.io/tqdm/>
+
+## Running
+
+These benchmarks are run automatically for all releases and pull requests.
+
+To run locally:
+
+- conda/pip install `virtualenv` and `asv`
+- clone this repository
+- run `asv --help` in the repository root (one directory above this file)