Blog
Notes on texting, and how this app reads it
Deep dives into how the parser and statistics engine actually work, and what real texting patterns look like once they are measured instead of guessed at.
- How software tells 03/04 apart: DD/MM vs MM/DD without askingAmbiguous dates are not solved by guessing a locale. Here is the actual resolution strategy: sample a window, commit to a configuration, then scan the whole file for a day above 12 to confirm or flip it.
- Who is more into whom: the four numbers behind the asymmetry scoreThe asymmetry score weighs day-openers at 30%, double-texts at 25%, reply speed at 25% and volume at 20%. Here is why that order, why ±0.15 is called dead even, and what the score ignores on purpose.
- Why "signature word" does not just mean your most-used wordYour signature word is a normalised gap between two people, not a raw frequency count. Here is the arithmetic, worked through on a small example, plus why +1 smoothing and an 8-use floor exist.
- Your reply time is two different numbers, and they disagree on purposeWhy the same WhatsApp chat can honestly report a reply time of seconds and a reply time of hours at once, what the 30-minute session rule actually protects, and the questions to ask before comparing two tools' numbers.