diff --git a/models/runner/README.md b/models/runner/README.md
index e8fe3565..b710cb51 100644
--- a/models/runner/README.md
+++ b/models/runner/README.md
@@ -14,20 +14,26 @@ conda install --yes --channel pytorch pytorch==1.5.0 torchvision==0.6.0 cpuonly
 conda install --yes --channel fastai fastai==1.0.61
 conda install --yes --channel conda-forge pyinstaller==3.6 exifread==2.1.2
 conda install --yes --channel conda-forge opencv==4.4.0
+conda install --yes matplotlib==3.1.3 scipy==1.1.0
 ```
 
-4. build the bundle - go to `models/runner` and run `release_model_win10_x86_64.bat`
-   *Note:* If you run into issues with loading dependecies, set a shorter `$env:PATH` with only system paths and conda (by running something like `$env:PATH="C:\Users\your_username_here\miniconda3\condabin;C:\Windows\system32;C:\Windows"`), and re-activate the environment (Windows has trouble loading packages if $PATH after activating the environment is too long ¯\\\*(ツ)\_/¯)
+4. Build the bundle - go to `models/runner` and run `release_model_win10_x86_64.bat`.
+    * If you run into issues with loading dependecies, set a shorter `$env:PATH` with only system paths and conda
+      (by running something like `$env:PATH="C:\Users\your_username_here\miniconda3\condabin;C:\Windows\system32;C:\Windows"`),
+      and re-activate the environment (Windows has trouble loading packages if $PATH after activating the environment is too long ¯\\\*(ツ)\_/¯).
 
 5. Now your model runner is in `runner_win\main\main.exe`.
 
-6. Run it for example with: `main.exe infer_to_csv --model=..\..\serengeti\trained_model.pkl --input_folder=..\..\serengeti\images_fun_examples --output=output.csv --overwrite --pytorch_num_workers=0`
-   (Warning: this needs to be run from `main.exe`'s directory for Pytorch to locate the necessary source files.)
+6. Deactivate the environment (`conda deactivate`) and test the binary, for example with:
+   `main.exe infer_to_csv --model=..\..\serengeti\trained_model.pkl --input_folder=..\..\serengeti\images_fun_examples --output=output.csv --overwrite --pytorch_num_workers=0`
+    * Warning: this needs to be run from `main.exe`'s directory for Pytorch to locate the necessary source files.
+    * If you get a `PyInstallerImportError`, try to locate the directory with the missing DLL on your system and add it using `--paths` option for pyinstaller,
+      e.g. `--paths C:\Users\your_username_here\miniconda3\envs\mbaza\Lib\site-packages\torch\lib`. Rebuild (step 4) and try again.
 
 7. Add `biomonitoring_stations.csv` to the main `models` directory. Distribute the entire `models` folder as a zip archive containing `biomonitoring_stations.csv`, `runner_win` and directories with models (e.g. `gabon` and `serengeti`). Do not include the `runner` directory as it contains source code and temporary build files. You can package this with:
 
 ```
-Compress-Archive -Path .\models\runner_win\, .\models\gabon\, .\models\serengeti\, .\models\biomonitoring_stations.csv models_v002.zip
+Compress-Archive -Path .\models\runner_win\, .\models\gabon\, .\models\serengeti\, .\models\biomonitoring_stations.csv models.zip
 ```
 
 ## Building on Linux