People have bad days. The --grace-period flag issues a single warning before collection. Use it. Or don't. We don't judge.
The core assumption behind the grace period is that most people, when clearly and honestly told that their behavior is causing harm, have the capacity to change it. Not all of them will. But the offer should be made.
It's also just fair. You've been tracking someone for days or weeks. You have a detailed picture of their behavior. They have no idea this has been happening. Before you act on information they don't know exists, giving them a chance to respond to it is the decent thing to do.
It is not another chance to explain themselves. It is not an invitation to negotiate about whether the behavior happened. It is not an opening for them to reframe your experience or make you feel responsible for their patterns.
The grace period is a notification: this behavior has been observed, it has reached a threshold, and if it continues, there will be a consequence. It's honest. It's direct. It gives them real information to act on.
The default grace period is 7 days. During that time:
Seven days is long enough to be a real opportunity and short enough to not become indefinite delay. Adjust with --grace-days N if your situation calls for it.
The algorithm distinguishes between actually getting better and performing improvement for the duration of the grace period. The reset threshold is 0.30 — well below the 0.55 that defines a bad day. You can't slip through the grace period by merely being not-terrible. You have to be genuinely decent.
Additionally, the grace eligibility system tracks whether improvement during a grace period has preceded relapse before. A subject who improves during grace periods and then regresses has their grace eligibility suspended for 90 days on the next flag. The tool has a memory.
The grace period is optional for a reason. There are situations where issuing a warning first is the wrong call:
The system won't write the notification for you — that's your voice, not ours. But what it should contain is simple: what you've observed, how long you've observed it, what you need to change, and what happens if it doesn't. No hedging. No softening that obscures the meaning. Clear and direct is kinder than vague and deniable.
Avoid starting it with "I feel like…" That's an invitation to debate your feelings instead of address your observations.