Question class and the 30-resolved gate

Every question carries class (default "general"; letters, digits, space, _ or -). The record is kept per person, per class, per read type (gut, heart on forecasts, head, final).

Below 30 resolved forecasts in the class, the human answer is the final read alone, and the record is not used to route asks, weight team answers, or let agents act on standing reads.

At 30 or more, the human answer is the weighted mean of that person's own reads, each weighted

w = 1 / (0.05 + mean Brier of that read type in that class)

Each answer returns question_class, resolved_behind (resolved reads behind the number), gate (30) and method (final or weighted_reads).

Teams (12 or fewer): each member's final is weighted the same way; members under 30 get the median weight. A read more than 2.5 from the weighted median, or more than two weighted standard deviations from the weighted mean, gets a gentle check ("Your read is far from the group. Take a breath and look again?") with the slower reads before the combined answer returns. The member may keep their number. Names and reads stay inside the consenting group.