The exercises we looked at were everything from the index-card/finger and stop-watch to flashing words all-around the screen. Based on the small bits of data available and the effectiveness of the various exercises on our own reading speeds, we found that flashing "chunks" of text (similar to spreeder.com) and the "word-snake" method (that we decided on) were the only effective methods of increasing reading speed. Because the word-snake most closely approximated off-line reading, we settled on that as the basis of our curriculum. For sources, we looked at available literature like Evelyn Wood and software like 7SpeedReading.

As far as a data blog - that is one of our middle-term goals, but until we have more users, the best we can offer are anecdotal evidence. This type of shoddy analysis is a big part of the problem with existing reading research, so the plan is to wait until we have a more robust data-set before we start publishing results. I'll post on HN when we start publishing results.

The reason for the difference between the final treadmill reading speed and the suggested speed is that users regularly demonstrated a significant loss of comprehension at the final speed. We think that it was because of the delay between realizing that the treadmill was too fast and stopping the treadmill. At first, we reduced the speed by 10%, but after a few more users, it was clear that was insufficient, so we're trying it at 20%. The quiz speed is adjustable (with the faster/slower buttons or by using f and s as hot-keys), so it's meant more as a suggested starting point than as a definitive reading speed.

Thanks for your feedback!