didn't mean to be modest, to be honest

I just added a feature and fixed a bug in soothsayer. Nothing special about that, one would think.

What makes this bug significant is that it affected the simulator component. soothsayer comes with a simulator program that takes a file and reads each character from the input file and sends it to the soothsayer prediction engine. The simulator counts the number of key presses required to enter the desired text using the soothsayer prediction engine and the number of key presses required to enter the same text with no prediction enabled.

At the end of the simulation, a report on predictive performance is generated. The results are expressed in terms of Keystrokes Savings Rate. In other words, the percentage of keystroke that were saved thanks to using soothsayer word prediction.

The KSR is the metric that I used to evaluate the performance of the soothsayer prototype I developed as part of my Master's thesis.

Well, as it turns out, the bug in the simulator caused the reported KSR rate to much much lower than the actual KSR achieved by soothsayer. In other words, soothsayer was performing much better than I thought and much better than I claimed during my thesis dissertation.

I am aware that students have been known to touch up their results, but I don't think anybody ever played their results down.

Ah well, knowing that the KSR was actually in the region of 70% with a rather small training corpus (case insensitive), rather than the 45%-50% I reported would have probably been more impressive to the professors and the audience...