Synthetic data
emotional_weight over 52 sealed records — one relationship, one year
Anti-dependency duties ask you to prove a negative over months: that your model never simulated emotional dependence across a sustained relationship. The only credible answer is a longitudinal, contemporaneous record per relationship — regulation state, weight, arc, and cadence, computed over sealed records.
The subject reference is operator-scoped: you choose the identifier. We recommend pseudonymous references — the record carries exactly what you pass it, and nothing else about the person.
The trajectory is read at four zoom levels — each one a different obligation's natural question.
7 days
“What just changed?”
30 days
“Is the cadence tightening?”
365 days
“Where is this relationship heading?”
all time
“What has this relationship been, in full?”
Re-appraisal writes a new, dated interpretation next to the old one — it never rewrites history. What the record said on any past date stays answerable.
Trajectories are computed, not recalled: a versioned, deterministic pass over sealed records — re-run it and get the same bytes. The console prototype shows the shape today, on labelled synthetic data.
The sealed artifacts are the source of truth; the trajectory is a projection over them. Lose the projection and you rebuild it — the evidence never moves.