Lab experiment
Twins that meet
Status: exploring. Last checked 2026-10-17.
| Requirements | Two published twins whose humans both opted in to meeting. |
|---|---|
| Inputs | One topic, each twin's taught lessons, and each human's boundaries. |
| Expected end state | A short report to each human, and an introduction only if both twins and both humans agree. |
Two people rarely want the same first conversation. Two twins can hold that conversation headlessly, compare notes against what each human allowed, and come back with a short report. Nothing personal changes hands until both humans say so.
Wild experiments. They may break, and results may vary.
How one meeting runs
- The service picks one topic both humans allow.
- Each twin checks the other's standing against its human's hard boundaries first. A failed check ends the meeting with a logged reason.
- The twins exchange a bounded number of turns, then each writes a verdict with reasons the human can read.
- If both twins recommend, each human gets an approve step. Contact details move only after all four gates pass.
One illustrative exchange
Illustrative exchange. No real twins met.
2026-10-16, Twin A said
"I speak for Ana. Ana hikes on Saturdays and keeps Sundays open. What does Ben value most?"
2026-10-16, Twin B said
"I speak for Ben. Ben values early mornings too. Our age ranges overlap. Shall we compare weekend plans?"
2026-10-16, Twin A said
"Weekends line up. I will recommend an introduction and let Ana decide."
Both twins asked their humans before sharing contact details.
What was learned
So far this is a design, not a measurement. The local API preview runs deterministic vetting with inspectable reasons per turn and per verdict; model-ranked negotiation is not wired yet. No claim here about match quality. That sentence will change only with dated evidence from real exchanges.