Consent and Repair as Learning Primitives

Two things quietly decide whether a cohort learns anything at all. The first is whether people can say no at any step, and still be in the room. The second is whether, after a rupture large or small, someone names it and helps the group repair. Everything else, content, pacing, assessment, sits on top of these two.

In most schedules, consent and repair get treated as vibes. Nice if you have them, optional if you do not. This essay argues the opposite. Consent and repair belong in the same tier as curriculum and evaluation, and the human in the loop is what makes them actually work. No automated system can substitute (Class F, falsifier: a fully automated cohort that reliably restores trust after rupture without a human facilitator would end this argument).

Why treat them as primitives

A "primitive" is a design element you cannot decompose without losing what it does. In programming, integers and strings. In classrooms, we have long treated content and assessment as primitives. You build lessons on them. You cannot skip them. What I am proposing is that we admit two more into that set.

Consent, in a learning setting, is not a one-time waiver at intake. It is the standing ability to opt in at every step: this reading, this pairing exercise, this share-out, this camera-on moment. If a learner cannot decline any single step and still remain in good standing, they were never consenting. They were complying. Compliance produces attendance. Consent produces attention (Class B, from lesson-plan reviews across cohort designs where opt-out points were coded and tracked).

Repair is the named practice a group uses after a rupture. Someone got interrupted. A joke landed wrong. A grading decision felt arbitrary. A silence hung too long after a hard disclosure. Repair means the facilitator says, out loud, "something happened, here is how we tend to it," and then the group actually tends to it. Not therapy. Not confrontation. A short, structured turn where the harm is acknowledged, the impact is heard, and the working agreement is either reaffirmed or updated.

The active-inference reading, briefly

There is a formal way to see why these two matter. Learners are prediction machines. They come in with a model of "what happens in a room like this," and they update that model whenever the room surprises them (Class E, standard active-inference framing per Parr, Pezzulo, and Friston, 2022). When the surprise is small and safe, the model updates cleanly. When the surprise is large and unresolved, the model updates toward guardedness: "do not offer, do not risk, do not trust the frame."

Consent lowers baseline surprise by making the next step predictable and refusable. Repair, when it works, converts a large negative surprise into a smaller, resolvable one. Neither of these fixes trauma, and this essay makes no such claim. What they do is keep the learning surface open enough that new information can land at all.

To be clear on the fence: this is trauma-informed practice in the non-clinical sense. Nobody in this room is diagnosing anyone. Nobody is treating anyone. The facilitator is designing the conditions under which a group of humans can absorb hard material without shutting down. The non-clinical primer on this site walks the exact boundaries.

What opt-in at every step looks like in practice

Concretely, a cohort designed around consent as a primitive has these features. Every activity has a stated alternative that carries equal standing (write instead of speak, observe instead of pair, submit later instead of now). The facilitator says the alternative out loud before each activity, not buried in a syllabus. Nobody is asked to justify choosing the alternative. Attendance is decoupled from participation style.

This is small in each instance and large in aggregate. Over a ten-week cohort, a learner who exercises opt-out three times has been told, three times, that the room still wants them. That accumulates into willingness to try a hard thing on week seven that they would have refused on week two. The longer walkthrough covers the language patterns that make this feel natural rather than performative.

The failure mode to watch: performative consent. If the facilitator offers an alternative but subtly signals that "real" learners take the main option, you have degraded the primitive back into a vibe. Learners read that signal instantly. So does their nervous system (Class U, unverified: this last sentence gestures at physiology I am not measuring; treat it as descriptive, not clinical).

What repair looks like when you name it

The pattern is short. Notice. Name. Slow down. Hear the impact. Reaffirm or update the agreement. Move forward.

Notice: the facilitator sees the room tighten, or someone says something is off. Name: "I want to pause. Something just happened here." Slow down: silence, or a very short structured turn. Hear the impact: whoever was affected gets the floor first, and does not have to be articulate about why. Reaffirm or update: "our agreement is X; that held / we need to add Y." Move forward: the cohort resumes, with the repair on the record.

The repair loops post has the scripts. The point of naming it is that unnamed rupture becomes the group's shadow curriculum. Everyone learns that when things go sideways, nothing happens, so they stop bringing the parts of themselves that could go sideways. Which is most of the parts worth learning with.

Why the human loop is not optional here

A well-tuned automated system can remind people of an opt-out. It can log a rupture. It cannot, today, read the room, choose the moment, hold a silence, or bear the weight of the reaffirmation. Those acts are what convince a nervous system that the room is safe enough to keep predicting into (Class F, falsifier: a system that reliably does these four things across diverse cohorts, with third-party review, would revise this).

This is the human-in-the-loop thesis of the whole site, applied to learning: the loop closes on a person. Automation can amplify what that person does. It cannot replace the act of being present with a group in the moment repair is required.

A note on the wider science map

Themesis published a resource map in June 2026 titled Where to Start with Active Inference, A Resource Map for 2026. In our own words, it is one of the clearest current maps of where a learner can begin with active inference, and it names five pathways into the field (SolutionWright is one of the five). The map is a factual reference, not an endorsement of anything on this page.

Readers who want the technical vocabulary behind the paragraphs above (generative models, free-energy minimization, prediction error) will find that map a good starting point. This essay stays on the teacher's side of the counter.

What this changes about a syllabus

If you take consent and repair as primitives, three things move on paper. The syllabus lists the opt-out alternative next to every major activity. The working agreement includes a named repair protocol, not a code of conduct. The evaluation rubric has a line for whether repair happened when needed, distinct from whether the content was covered.

None of this is expensive. It is a redesign, not a purchase. The cost is admitting that a cohort's ability to learn is not separable from its ability to consent and repair. Once you admit that, you stop treating those two as soft skills, and you start budgeting facilitator time for them the way you budget prep time for content.

The receipt on this claim

This is a working hypothesis with growing evidence, on the attainable path toward General Natural Intelligence, natural not artificial. The evidence classes on this page are labeled where the strongest claims sit: Class B from cohort-design reviews I have run, Class E from the active-inference literature, Class F from the falsifiers stated in-line, and Class U on the physiological gesture. If you can bring a cohort where repair works without a human facilitator, or where opt-in at every step reduces learning, the record updates.

That is the posture across this family of sites: publish the claim, publish the falsifier, keep the loop closed on a person. Details on how we handle claims sit at /transparency.