I stumbled on Refresh Cannon today. The entire “game” (it’s extremely simple) is controlled by refreshing your browser. The goal is to use a cannon to hit a house on the other side of the screen. The first refresh starts increasing the cannon’s angle, the second refresh locks-in the angle and begins increasing the power. The final refresh locks-in the power and fires.

The angle goes from 0-90º over 10 seconds, so half a second is the difference between 75º and 79.5º. Since the novelty of refreshing your browser gets old extremely quickly, I wrote a Ruby script to give myself the high score. The script just tries a bunch of angles and powers that are likely to result in a hit. I have no clue how to do image processing in Ruby, so I just checked the HTTP response size. The “Win!” image is usually about 1k larger than the “Miss!” image.

Unfortunately, it looks like somebody else wrote their own script and started running it several hours before I discovered the blog post. I’ll have to settle with second place. Update: I am now in 1st place.

I didn’t put much effort into the angle/power guessing, since latency screws up any attempt to methodically carpet bomb the map. I account for constant latency, but jitter is bad enough that trying the same times between HTTP requests will cause the cannonball to land only in the general area intended.

Here’s my script, in case you’re curious. Yes yes, I know it’s hideous.

If I felt like wasting more than an hour on this, it could be improved by:

• Storing success rates of angle/power pairs, then trying the most successful pairs first.

• Registering multiple IPs to the same username, allowing multiple copies of the bot to increase a single user’s score.

• Using image processing and OCR to find the difference between the attempted angle/power and the actual angle/power.