About Cor
Cor is an evidence-graded atlas of what a human being needs. Modern suffering is mostly accurate signaling in a world that no longer fits us, not a defect in the person. The map shows where the fit breaks, and how strong the evidence is at each point. One idea runs through all of it. Once you see it, the rest follows.
The one idea
Your body runs old systems - for hunger, for safety, for closeness, for status, for rest. Each one works the same simple way. It sends up a signal ("you need this"), and the signal is meant to stay on until a real-world function actually delivers the thing and quiets it down.
Hunger is the clean example. The pang is the signal. Food is the function. You can quiet the pang in two very different ways: eat a meal that nourishes you, or eat something engineered to taste like a meal while feeding you almost nothing. Both can hit the "this tastes like food" cue. Only one keeps the body running. The second is a proxy: it fires the cue without supplying the function.
The function that resolves it - draining
A proxy fires the cue without supplying the function that would resolve it. That one sentence is what the whole framework turns on.
Two things matter about this picture, and Cor is careful about both. First, it is a slope, not a switch. The old way of saying it - "does it trigger the system, yes or no" - is too crude. Real life lives on the staircase: a video call with a close friend sits high on it; an endless feed of strangers' highlights sits low; a chatbot that says it loves you sits lower still. Second, the steps are an order, not a score. Cor will tell you one rung sits below another. It will never pretend to hand you a precise number for how lonely you are. The map shows direction, not a score.
The claim
Loneliness, anxiety, restlessness, burnout, the hollow feeling after hours of scrolling - the usual story is that something is wrong inside you and needs to be patched. Cor's reading is different and more hopeful: in a great many cases the signal is working exactly as designed. It is firing because a real function is missing. You are getting the cue of connection without the connection, the cue of status without a stable place among people, the cue of rest without rest.
Instead of only asking "how do I cope with this feeling," Cor asks "what real thing is this signal asking for, and why is the environment handing over a proxy instead?" The fix it points to is usually changing the inputs - the conditions around the person - not silencing the person's alarm.
This is also why Cor keeps circling back to AI and the systems that shape our days. Those systems are extraordinarily good at producing proxies: the cue of intimacy, the cue of approval, the cue of progress, delivered on tap. A system that learns only from what we click and ask for, with no model of what we actually need, will keep getting better at firing cues that resolve nothing. Naming the gap between cue and function is what Cor is built to do.
How Cor reads a situation
The method is more practical than it sounds. Faced with any recognizable struggle, Cor works through the same four questions.
Is this mainly about closeness, safety, status, grief, rest, movement, play, or something else? Name the system before judging the feeling.
What real-world function would let it settle? Not the cue that imitates it - the thing itself.
Where on the staircase is this person living? Are they getting the function, or just the cue that mimics it?
Instead of only asking the person to cope harder, ask how the environment, the tool, or the institution could stop manufacturing the mismatch in the first place.
You can watch this run on a real life on the cases, and you can see all 15 of the systems Cor maps - each with its own working version of the staircase you saw above - on the mechanisms page.
How it was built
Cor is built from the bottom up, from evidence to conclusions, and never reverse-engineered to fit a theory decided in advance. The sources are books from a personal library and papers pulled through PubMed and citation chains, with full-text access to the harder journals through Groningen's university library. Primary sources only. No textbooks, no summaries standing in for the research they describe.
The full-text rule is strict because an internal audit found a complete failure rate on claims pulled from snippets: fabricated quotes, drifted paraphrases, invented findings.
Works are tiered by how load-bearing each one is. Foundations carry epistemic grades that separate convergent evidence from single-source claims. Mechanisms carry tier ratings for evidential strength.
One rule governs all three: when there is doubt about how strong a piece of evidence is, it is graded down, not up. "Replicated" is reserved for genuine agreement across the field. Mismatch interpretations stay marked theoretical until directly tested.
The thinking came first, by hand. From there, a multi-agent AI pipeline did the corpus work at scale - finding the evidence, grading it, checking sources, and linking it to the systems it bears on. That is how the atlas reached this coverage in a year.
It is also the source of the data-quality problems the next phase of work fixes. Extractions found to be snippet-only are quarantined and held out of the evidence base. Don't read "AI-assembled" as a reason to trust it - read it as a reason to check it. Every claim opens onto its source.
The structure the evidence is fitted into is a pyramid, derived once and now locked: two framing premises, three core premises, the derived properties of the architecture, the convergences between lines of evidence, and at the top the mechanisms themselves. The full chain, and exactly how well each layer is grounded, is laid out on the derivation page.
What to trust, and how much
A map is only honest if it marks its own edges. Cor is built from three different grades of certainty, and it never quietly mixes them. When you read any page, it helps to know which grade you are standing on.
Established consensus
That the old emotional systems are shared across mammals; that attachment is a real, evolved system; that chronic mismatch wears the body down; that wanting and liking can come apart; that there are developmental windows. Each rests on several independent bodies of research. Cor did not invent these - it builds on them.
Cor's synthesis
Established findings, fitted into one connected structure, and the proxy-gradient logic carried into areas like meaning, status, grief, and parenting. The pieces are well-supported; stacking them this way, and stretching the logic that far, is Cor's call. Where fields disagree and a position is needed to stay coherent, Cor flags it as an interpretive call.
Not yet proven
The questions the map cannot yet answer well - where the natural baseline is poorly measured, or a claim has not been directly tested. Cor lists these out loud as open questions rather than papering over them.
The right way to hold Cor is the way a clinician holds an anatomy spec: a careful reference to think with, not a verdict to obey. It takes positions where the evidence forces one, marks where it is reaching, and ships in pieces. Its own biggest known weaknesses are on the record in the FAQ and the open gap register, stated before anyone else raises them.
What the counts mean
The atlas reports its counts live on the home page, and they move as the work continues. Any figure quoted anywhere else is superseded by what the dashboard shows. What each number actually counts, and where the caveats sit:
What Cor is not saying
Where to go next
Cor is the reference work at the center of Demismatch. For the wider project - the two conclusions it drives, the film, and the essay - see About Demismatch. Corrections, questions, grants: write to maartenrischen@protonmail.com. I read every message.
In one sentence Cor is an attempt to put a better model of the human into the systems that increasingly shape human life - so they stop optimizing for the cues we reach for and start serving the functions we actually need.