How long before Tessary's frustration classifier starts judging a call site?

Tessary’s frustration classifier starts judging a call site once its learned reference holds 100 scored conversations, down from the 200 it used to require, and the reference now keeps learning from every conversation after that until it reaches 1,000, instead of freezing the moment judging begins.

Before the reference holds those first 100 conversations there’s nothing to compare a call site’s current rate of frustrated conversations against, so nothing is judged yet. A call site too quiet to reach 100 conversations inside the 28-day replay window is never judged at all, and its Tuning row reads learning n/100. The lower start exists because a call site that had to wait for 200 conversations went a long time before a new project saw its first finding. The same tradeoff shows up in how long a call site’s cost and duration baseline take to become trustworthy: a classifier is only useful once it has watched enough of a call site’s own traffic to know what normal looks like there, and a lower starting count trades a slightly noisier first few hundred conversations for a usable answer sooner.