Feature Suggestions:

When I'm writing a character that I'm not completely sure I remember, I often find that I am helped to remember by the way the evaluator prompts me as I go along. Since I'm not reproducing the character entirely from memory, I feel that it impairs my learning. Also, for some characters, it is hard to remember whether there are additional dot-strokes at the end. Could we have an option where we can write the entire character before the evaluator starts processing?

Also, along those same lines: Often when the server is a bit slow, I input several strokes before the evaluator starts processing. When one stroke (usually a tricky hook-stroke) fails to register correctly, all the subsequent strokes are flagged as incorrect. Could the evaluator just cancel any pending strokes after a stroke is marked as incorrect?

We're going to be able to address most of those issues when we move toward radical/component-based recognition. We'll have all characters cut up into pieces, and it won't snap-to until you finish a component. That'll reduce the excess hinting. That'll also allow it to be smarter about recognizing multiple strokes that go together. What do you think?

It'll take a while to build that. If we made the option you describe as a temporary thing, where it waits until you're done writing to recognize, then it would slow things down a lot, because the server-side recognition has to be done in parallel, so you'd be waiting a while to get all your strokes back.

When we finalize the recognition code and translate it into ActionScript, it'll happen instantly, which will be really sweet. We aren't doing that yet, though, because we can't tune it as easily in ActionScript as we can in Python. Hopefully before too long we'll work around that and get things running much faster.

I think that in a couple weeks when I have time to fix most of the tricky-to-recognize strokes, you'll see less problems related to missing a stroke in the middle.