Home › Page Design › How It Works

The Engine of Reason: How the Idea Stock Exchange Works

How structured argument, evidence-weighted ranking, and one permanent page per idea turn scattered opinion into cumulative knowledge.

1. The Core Problem: The Chaos of Threads

The flaw in the modern internet is not that people are stupid. It is structural. Bad arguments are never actually defeated, they just move to a new thread. Because every discussion lives on its own disconnected page, humanity has no long-term memory for debate. We refute the same nonsense every morning, forever, like a civilization with amnesia.

The fix is almost boringly simple to state. There should be one page per topic. When an argument is debunked there, it stays debunked, and the next person who shows up starts from the current state of the evidence instead of from scratch.

2. The Solution: Structured Argumentation

The Idea Stock Exchange replaces the linear thread with a recursive argument tree. Instead of a comment box where remarks pile up and scroll away, every belief has a structured page, and you contribute by acting on its logic directly rather than talking near it.

In practice that means two kinds of moves. You add a row to a structured table, a new reason to agree or a new reason to disagree. Or you challenge a number. Every score is a doorway. The app ships a live explainer page for each score, a provenance view that derives a single argument's impact factor by factor, and an add-a-reason form that will not let you post on a high-stakes belief until you have acknowledged the strongest point on the other side. The wiki templates are the readable blueprint; the app is the working version.

3. The Database: A Family Tree of Ideas

Under the hood, the structure is a genealogy chart, except it tracks logic instead of people. Arguments are parents that combine to support a conclusion. Those conclusions become parents in turn, premises for larger ideas built on top of them. The link between any two of them is not a bare hyperlink, it carries a linkage score that says whether this is a reason to agree or a reason to disagree, and how strongly.

Call it the turtle stack. A child has parents, and those parents have parents, and every argument on the exchange is itself a full belief page with its own pros and cons. You can drill down as far as you want until you hit raw evidence at the bottom. There is no hardcoded floor, on purpose.

4. The Numerator: ReasonRank

How do we know which side is winning? Not by counting upvotes, because a vote measures how many people showed up, not whether they are right. ReasonRank weighs the evidence instead. It is the same insight that made PageRank beat keyword-counting for web search, pointed at arguments: a claim is strong when strong claims support it, recursively, all the way down to evidence.

The conceptual core is a balance of supporting and opposing weight, where each argument counts only as much as it is relevant and only as much as it is not a restatement of another argument:

Conclusion Score = ∑(Agree Arguments × Linkage × Uniqueness) − ∑(Disagree Arguments × Linkage × Uniqueness)

That is the readable summary, not the exact math. The full recursive definition, with normalization and the twelve score dimensions, lives on the ReasonRank page and in the engine docs, which are canonical. Two of the terms above are worth a sentence here, with the full definitions one click away:

A worked illustration: should we have joined World War II?

Consider two arguments for entering the war, both of which happen to be true.

• "Nazi leaders were rude and unpleasant." True, and almost irrelevant to a decision to go to war. Low linkage. Illustrative score: [+2].
• "Germany was carrying out systematic genocide." True, highly relevant, backed by overwhelming evidence. High linkage. Illustrative score: [+95].

The point of the design is that the second argument dominates the first by structure, not by whoever repeats their point most often. The genocide claim carries the decision because it is both true and tightly linked to the conclusion, and the ranking is built to surface exactly that. Those two numbers are hand-picked to show the linkage contrast; live scores belong to the app, where they move as arguments land.

5. The Denominator: Compared to What?

A strong numerator is only half of an honest score, and it is the half that fools people. An idea can have a tall stack of real benefits that beat its real costs and still be a bad call, because you were never choosing between the idea and nothing. You were choosing between the idea and the next best thing you could have done instead. The engine scores that too. It is the layer that keeps the machine asking "compared to what."

Take "We should fund Program A." In isolation, A might have genuine benefits that outweigh its genuine costs, so it scores positive and everyone feels good. But you have one budget. If Program B would do more with the same dollars, then funding A quietly costs you B, and A's real score has to carry that subtraction. The denominator does this in two layers, matching the spec in the repo:

• Layer one, the justification score (internal denominator). How decisively do a belief's own arguments favor it over simply doing nothing? Not "is the support net positive" but "how lopsided is it." A belief that barely clears zero is a very different animal from one that buries its opposition, and a raw sum hides the difference.
• Layer two, opportunity cost (external denominator). How does the belief rank against the mutually exclusive rivals competing for the same budget, the same afternoon, the same policy slot? A belief's real value is docked by the best alternative you gave up to hold it.

This is the old cost-benefit discipline made mechanical: compared to what, measured in what, and how likely. A belief that looks great alone and mediocre against its alternatives is exactly the comfortable mistake this layer is built to catch. Belief pages render the contrast class directly, the rival options and the opportunity-cost values sitting right next to the argument tree, so the denominator is never out of sight while you admire the numerator.

6. From Argument to Agreement: Conflict Resolution

Once beliefs are decomposed into scored trees, something useful falls out almost for free. You can read the same trees sideways and turn a shouting match into a map. This runs through the cost-benefit analysis workflow and as a section rendered on belief pages, and it borrows its logic from Fisher and Ury's "Getting to Yes," at internet scale.

Most debates get stuck on positions, the "I want X" surface. The exchange pushes past that to interests, the "I need X because of Y" underneath, because interests are where the overlap hides. Once both sides' costs and benefits are scored, the conflict resolution pipeline pulls out four things: the interests both sides actually share, the single primary conflict pair really driving the disagreement, the genuine value conflicts where one side prices freedom and the other prices safety, and the compromise candidates where a small achievable shift would flip a category's net.

The primary conflict pair is worth one careful sentence, because it is easy to get wrong. It separates how much an interest supports this particular position (its issue-relative claim strength) from how legitimate that interest is in general (its standalone validity). Those are two different numbers, and treating them as one is a scoring error the pipeline is built to avoid. The compromise candidates are the payoff. They point you at the winnable disagreements instead of the symbolic ones.

The design also adds friction where it helps. On high-stakes beliefs the add-a-reason form imposes speed bumps before you can post: acknowledge the strongest point on the other side first, and check whether your proposed solution is consistent with the moral principle you claimed a paragraph ago.

7. The Vision, Scoped Honestly

It is tempting to say we are building the collective consciousness of the internet. We are not, and promising that would burn exactly the people we need. Here is the honest version of the ambition, which is still large enough to be worth twenty years.

Wikipedia gave the world one shared, cumulative page for facts. The Idea Stock Exchange aims to do the same thing for reasoning: one shared, cumulative, evidence-ranked page for the arguments on each idea. That is a narrower claim than "fix democracy," and it is one the mechanism on this page can actually deliver. If it works, three things follow. A refuted argument stops circulating instead of respawning in the next thread. Each round of debate builds on the last instead of restarting from zero. And anyone can see the whole map of reasons to agree and disagree with a claim in one place, ranked by evidence rather than by volume. Whether that adds up to better collective decisions is the horizon we are aiming at, not a result we are claiming.

Get Involved

This is an open-source project with a real repository, a running scoring engine, and, honestly, not enough hands. If you are a developer, the vision is not the entry point, a concrete first issue is. The stack is Next.js and TypeScript, the engine and the denominator and the market layer all have written specs, and there are functions in them waiting to be implemented against tests. Clone it, run it, and pick something small. If you are a thinker rather than a coder, the wiki needs well-formed pro and con arguments and clean belief pages far more than it needs another manifesto. The fastest way to matter: the graph is young, and filling it with real arguments is what turns scores into verdicts worth trusting.

• View, run, and contribute on GitHub
• Contact the founder
• Start a new belief page from the template

Related scoring pages

The scores referenced above each have their own methodology page, and the app hosts a live explainer for each: ReasonRank, Linkage, Logical Validity, Evidence, Truth, Importance, Likelihood, and the broader algorithm to promote good ideas. The denominator layer described in section five has its canonical spec in the repo (THE_DENOMINATOR.md).