Got a base going for the #heliboard gesture testing harness.

According to my _not_ tested or verified outputs, the google library is getting gestures right pretty often.

I was ranking it this way:
correct word -> 1
on the suggestion bar -> 1/3
in the long press area -> 1/21

Then average...

batch-1-en-US, it got 0.88286 overall. The words we collected don't follow a typical word distribution; it isn't leaning on word frequency much. Gonna be work to beat that...

Now that we have so many submissions, and with a several months left in the gathering, I am considering changing the distribution of words we pick up in the active gathering in #heliboard

I am thinking of making it more closely resemble the real distribution of the words, and introducing a small bias towards short words.

We have a lot of really odd words, which is good, but we do need to make sure that we get more common words.

Let me know if you would find this more boring to contribute to!

Got 100k gestures for the dataset!

https://pawb.social/post/43577519

Well, that's surprising. I was typing something, and decided I wanted to replace a sentence with _something_ different, but in brackets. So, I *highlighted* the text to be removed, and started typing, «expecting» the highlighted sentence to be (deleted) by my ‹1st› key press.

Instead, it "placed" opening and closing brackets. „Worked“ with a couple markdown ‘options’ like bold and italic as well.

I thought it might be #Fedilab behind this, but it seems to be #Heliboard

Hmm, finally figured out how to add a Canadian dictionary to #Heliboard and stop having my words auto-correct to Australian. I have it set as the dictionary for the US English layout, although I'm not sure what the difference is between a US and Australian on-screen KB layout.

Prioritize! Colourize! uh...what else?