can you engineer serendipity?
Perspectives. Offline conversation circles in Delhi and Bangalore, co-founded and run from March 2025 to April 2026. The participatory leg of the thesis.
- (role)
- co-founder · participatory systems design
- (period)
- mar 2025 to apr 2026
- (context)
- 250+ circles · delhi · bangalore
- (status)
- wound down
Serendipity cannot be put on a calendar, but its preconditions can be manufactured. The framework turned out to be the cheap part. The binding constraint was the person opening the room.
A chain, and the link that broke
The project started as a bet on format. Write the protocol well enough and anyone can run it. Hand it to enough people and depth becomes a property of the system rather than of whoever is hosting.
What came back moved the weight off the protocol and onto the person running it. The protocol held from room to room. The rooms did not.
TKTK · confirm the starting bet really was on the format; this paragraph and the first box of the diagram assume it
- cohort
- 6 to 12 people, curated. TKTK · how a cohort was curated: who decided, from what list, on what basis
- duration
- 2 hours, in person.
- phase 1 · icebreaker
- TKTK · what it is for, how long it runs, what people actually do
- phase 2 · bubbles
- TKTK · what a bubble is, what it is for, how long it runs
- phase 3 · reflection
- TKTK · what it is for, how long it runs, how it closes the room
- initiators
- TKTK · how initiators were selected and what they were given before their first circle
- topics
- TKTK · who set the topic, and when people learned it
- composition
- TKTK · the rule for mixing a group: strangers only, ratio of repeats to first-timers, anything else
TKTK · the prompt sequence from one real circle, verbatim, phase by phase. Set it as `prompts:` (a list) in the frontmatter, with `promptsSource:` naming the circle and date.
-
01
Initiator quality dominates framework quality.
TKTK · the comparison that shows it: e.g. the same protocol under different initiators, with whatever you measured, and the n
-
02
The constraint on depth is permission, not time.
TKTK · what showed it was not time: a circle that went deep early, or a long one that never did, and how you knew
-
03
Depth is trainable, because the format carries the structure.
TKTK · evidence that an initiator got better: first circle against a later one, and what changed
-
04
Repeat attendance concentrates in a small anchor set doing unpaid social labour.
TKTK · what the anchors did in the room that nobody asked them to. The share is computed from the data in (05) below.
Ties between people with no path to each other
Every dot is a person. A line joins two people who sat in the same circle. A dark line is a tie that formed between two people who had no path to each other through any earlier circle. The data cannot see who already knew each other outside, so "no prior path" means none through Perspectives.
(sample data) drawn from 37 placeholder rows in perspectives.json, not from records. Replace them before this page is shown to anyone.
- tie with no earlier path between the two
- tie between people already connected
- size: circles attended
- anchor: 3 or more circles
fig.0225 people and 85 ties. 98% of ties formed between two people with no earlier path between them. 2 anchors hold 42% of all repeat attendances (5 of 12).
(sample data) drawn from 37 placeholder rows in perspectives.json, not from records. Replace them before this page is shown to anyone.
60%
of the people who initiated a circle had first sat in someone else’s.
3 of 5 initiators, read from the attendance records: an initiator counts if they attended a circle they did not run before the first one they did.
The conversations, as a graph
Topics, linked when the same people turned up to both. Topics on the same theme sit together.
(sample data) drawn from 37 placeholder rows in perspectives.json, not from records. Replace them before this page is shown to anyone.
(every topic, as a table)
| topic | theme | circles | people |
|---|---|---|---|
| sample topic 2 | sample theme A | 2 | 13 |
| sample topic 1 | sample theme A | 1 | 6 |
| sample topic 3 | sample theme B | 1 | 6 |
| sample topic 4 | sample theme B | 1 | 6 |
| sample topic 5 | sample theme C | 1 | 6 |
TKTK · paragraph one: the failure that cost the most. Name it plainly, say what you tried, say what happened.
TKTK · paragraph two: the failure you would design around first if you ran it again, and why the project wound down in April 2026.
This page handles data about private conversations between more than a thousand people. No names appear on it. Every person in the figures is a salted hash, and the salt is not in this repository. Topics are recorded at the level a circle was announced, never what anyone said inside it. TKTK · what was actually recorded during circles (notes, audio, nothing), how consent was asked for at sign-up and again for photographs, and who holds the raw records now
Can initiator quality be selected for before someone opens their first room, or does it only show afterwards?
Does a tie formed in a circle survive without the next circle to hold it?
When the anchors stop showing up, does the network keep its shape, or was it their shape all along?
TKTK · confirm these three are the questions you would put your name to, or replace them
The pol houses of Ahmedabad meet the street over an otla, a plinth wide enough to sit on. The square in front of them has a chabutra. The agora did the same work for Athens. None of these were built for conversation. They were built so that people would stop near each other with no one in charge, and conversation followed. Perspectives took that arrangement off the street and ran it as a protocol, in cities where the street had mostly stopped doing it. That makes it urban design by other means.
Perspectives was co-founded by TKTK · co-founders' names, as they want to be credited. I designed the session frameworks and the facilitation method, and every touchpoint a participant met on the way in. TKTK · confirm this is the contribution you want stated, and add anything that was yours alone