@dashstofsk Tavie's reading was correct -- the question was about qokeey at the end of a line and what appears at the start of the following line.
@tavie First a request, since I can't edit the original post anymore: could you delete it with a note like "I'm withdrawing this post due to a methodological error, see following posts for explanation"? Then people won't have to read it only to realize afterward that it was a waste of time.
Now the correction. I checked by hand in Notepad++ - you were right. It's a real blunder and I have to retract the post.
7 of 11 qokeey end-occurrences sit at paragraph ends. On 6 of those 7, the next paragraph starts with p. So my "qokeey -> p" effect was almost entirely the well-known top-line p effect that you and others have described. I just hadn't noticed - clearly on me.
Two things contributed. First, I'd been running my calculations on a cleaned EVA transcription that had stripped the <$> markers, so paragraph ends weren't visible in my data at all. But that's only the surface cause. The deeper one is methodological - It was my mistake; I should have thought of that, of course, and that means the other calculations are wrong too.
On the AI question as i write it before: Claude does the calculations - I couldn't run them at this volume otherwise. I cross-check against ChatGPT (deliberately given less context, to catch errors), and only post when both agree. That's worked so far... well, until now.
The flaw here was that both ran on the same cleaned file without <$> markers, so neither could see the issue.
I also let Claude do a final language pass on my drafts because writing statistical reasoning in English isn't my strong suit
. Some AI phrasing inevitably ends up in the text - you spotted that, and it's a fair call.
@tavie First a request, since I can't edit the original post anymore: could you delete it with a note like "I'm withdrawing this post due to a methodological error, see following posts for explanation"? Then people won't have to read it only to realize afterward that it was a waste of time.
Now the correction. I checked by hand in Notepad++ - you were right. It's a real blunder and I have to retract the post.
7 of 11 qokeey end-occurrences sit at paragraph ends. On 6 of those 7, the next paragraph starts with p. So my "qokeey -> p" effect was almost entirely the well-known top-line p effect that you and others have described. I just hadn't noticed - clearly on me.
Two things contributed. First, I'd been running my calculations on a cleaned EVA transcription that had stripped the <$> markers, so paragraph ends weren't visible in my data at all. But that's only the surface cause. The deeper one is methodological - It was my mistake; I should have thought of that, of course, and that means the other calculations are wrong too.
On the AI question as i write it before: Claude does the calculations - I couldn't run them at this volume otherwise. I cross-check against ChatGPT (deliberately given less context, to catch errors), and only post when both agree. That's worked so far... well, until now.
The flaw here was that both ran on the same cleaned file without <$> markers, so neither could see the issue.
I also let Claude do a final language pass on my drafts because writing statistical reasoning in English isn't my strong suit
. Some AI phrasing inevitably ends up in the text - you spotted that, and it's a fair call.