How long before Tessary's groundedness classifier starts flagging answers?

Tessary’s groundedness classifier starts flagging a call site once its learned reference holds 100 traces, down from the 200 it used to require; the reference still keeps learning after that until it reaches 1,000, the same as before this change.

Before the reference holds those first 100 traces there’s nothing to compare a call site’s current share of flagged answers against, so nothing is judged yet, and a call site too quiet to reach 100 traces inside the 28-day replay window is never judged at all. Unlike Tessary’s frustration classifier, which used to freeze its reference the moment judging started, groundedness already kept learning to 1,000 before this change; only the starting point moved, from 200 traces down to 100, so a new project sees its first finding sooner without giving up the accuracy a fuller reference provides.