Week 1 of 12
Tuesday
20:00
Chosen this week by —
Venue not set
The circle
Eight people. The venue rotates in this order, so nobody is the permanent host.
Join code
------
The agreement
The run
Twelve weeks. Attendance is the sum of check-ins, so nobody has to police it and nobody has to ask.
Tokens
The coin buys the room. Nothing inside the room is metered.
0
tokens left
No subscription. Nothing is charged for a month in which nothing happened.
Open question, unresolved by design
Metering messages would put a turnstile inside the room, on the exact thing the room exists to produce. This build ships threshold-only: the coin buys check-in, and talking is free. The pilot decides whether that changes, not the whiteboard.
The Science
Why eight, what is actually proved, and what the evidence says, including the parts that argue against this design.
What is proved
Place eight people on a circle so that distance around it stands for how different they are. The 28 relationships between them can then be traversed in exactly seven rounds, four pairs at a time, with every round holding the same chord length. The far pairs come last.
There are exactly two such schedules, so once you accept the design the timetable is essentially forced. The group sizes that admit one at all are the powers of two: 4, 8, 16, 32. Six, ten and twelve do not work.
All of this is proved and checked by machine. It is also not new: it follows from a lemma published in 1980. The paper claims novelty for none of the three results, only for the idea of ordering the rounds by similarity.
Why eight, honestly
Eight is the smallest size above four where the schedule exists. That is a fact about a schedule, not evidence about people. Nothing here proves that eight-person groups work better than seven or nine, and the paper does not claim it.
No study we could find has ever randomised group size, a meeting schedule, or a rotation. So the central operational claim of this design is untested.
What the evidence says
Six randomised field experiments bear on structured peer contact among business owners. Five of the six cut against this design in some respect.
- No control group, ever. No commercial peer advisory group has been evaluated against a control. What exists is vendor-commissioned comparisons that cannot separate treatment from selection.
- Rotation may be wrong. The nearest comparison found stable groups outproduced one-off cross-group meetings.
- Assignment is a two-sided lottery. Founders paired with less formal managers were more likely to fail. A schedule that assigns all 28 pairings maximises both tails.
- Random targeting does not help. A 2,000-run simulation holding tie count fixed found that forming more ties at random produces no more useful connections.
- The population is wrong. Every causal study was run among small firms in developing economies, not established owners in a high-income one.
What this paper is not
It presents no empirical validation. No cohort, no data, no measured outcomes. The mathematics is proved; everything else is a design proposal with an argument attached and eight named ways it could be wrong.
The full paper states all of this at greater length, including the seating problem, its failure case, and a sensitivity sweep showing how much its headline number depends on a constant chosen by hand.