Conflict inside a cohort is not a failure of the room. It is a signal the shared model just diverged, and the next few minutes decide whether the loop reopens or closes.
This is teacher notes, not a therapy manual. The frame is the same working active-inference pedagogy the rest of this site uses: participants, facilitator, and a running shared model of what the room is for right now. When two participants disagree in a way that pulls that model apart, the job is neither to moralize nor to smooth it over. The job is to walk the repair path. Nothing here treats a condition. Nothing here rewires anything. This is craft, and it is testable.
What "conflict" means in a working cohort
A cohort is a small group holding a shared task for a fixed stretch of time. Disagreement about the task is normal, welcome, often the whole point of the room. What we mean by conflict here is narrower: two participants, or a participant and the facilitator, arrive at incompatible models of what the last thing that just happened actually meant. Whose turn it was. Whether a comment landed as a joke or as a shove. Whether a question was answered or dismissed. The disagreement is about the shared model itself, not about content (Class B). That is the thing repair aims at.
The pedagogy literature has covered this ground for decades. Cohort-based learning benefits from explicit rupture-and-repair protocols, and adult learners in particular report better outcomes when the facilitator names the drift rather than papering over it (Class E). What we contribute is not the underlying insight. It is the specific sequence we run, published so it can be checked.
The repair path, four moves in order
The path is small on purpose. If it takes ten minutes to walk, we are already off it and into a different session.
Move one, name the divergence without a verdict
"Two different reads just landed in the room. That is worth pausing on." Not "someone here is wrong." Not "this is fine, keep going." The naming is the smallest honest statement of what happened. It hands the room a shared observation to work from, before anyone has to defend a position.
Move two, ask each side for the observation, not the interpretation
"What did you see or hear that led you there?" The question keeps both participants in the concrete, and it slows the reflex to argue over intent. Interpretations diverge fast. Observations usually do not, and when they do, that is important information too. If a participant declines, that is a full answer, and the path continues without them.
Move three, offer the smallest next step that keeps both agencies intact
Sometimes that is a two-sentence acknowledgement from one side. Sometimes it is a decision to hold the disagreement open and return to it in the next session, on the record so no one has to carry it alone. The facilitator's job here is configuration, not verdict. The step has to be one either participant can decline without losing standing in the room.
Move four, log what actually happened
A one-line entry in the session log, visible to the cohort at the next opening, describing the divergence, the step taken, and what remains open. No characterization. This is the receipt. Participants can read it, correct it, and see that the room does not pretend a rupture did not occur (Class C). The log is the thing that makes the path checkable across sessions rather than a story the facilitator tells about themselves.
What the path is not
- It is not mediation. The facilitator does not adjudicate whose reading was correct.
- It is not conflict avoidance. The disagreement is named out loud, in the room, on the record.
- It is not therapeutic. No one is asked to disclose history, feelings, or context they did not choose to bring.
- It is not a script. If the path is being read off a card, the room already knows and the repair is not working.
What would tell us the path is wrong
The falsifier we hold ourselves to is specific (Class F): if cohorts running this path show a rising rate of participants withdrawing after a named rupture, or if the session log entries stop being corrected by participants over time (the record going quiet is the same as the record going wrong), the path is not doing what we claim, and it changes. We publish the log format partly so this signal is legible.
Two things we are not saying. We are not saying every conflict can be repaired inside a cohort. Some cannot, and the honest move there is to close the loop cleanly rather than pretend to close it well. And we are not saying the facilitator is a neutral party. They are a participant with more configuration authority than the others, which is exactly why the log and the receipts matter.
The broader framing sits inside a working hypothesis on an attainable path toward General Natural Intelligence, natural not artificial. Do not take the cohort claim on faith either. Run the path, keep the log, and bring back the sessions where it stopped working. That is how the method gets sharper (Class U).