Home > Page Design > Algorithms > One Page Per Belief
The Atomic Unit of Reason: One Page Per Belief
Every distinct belief deserves a permanent home. Not a comment thread that disappears into the archive. Not a tweet that gets buried in six hours. A single canonical page where every argument for and against that belief accumulates, gets scored, and stays. By treating each belief as a unique object with its own page, we replace the repetitive noise of scattered debate with a single evolving signal. The same rebuttal no longer needs to be typed a thousand times. It gets typed once, scored once, and linked everywhere it applies. This is the difference between a room full of shouting people and a collective intelligence that actually accumulates knowledge.
The Problem: Fragmentation and Amnesia
Online debate suffers from four overlapping failures, all of which trace back to the same root cause: the platforms hosting the conversation have no incentive to remember what was already said. The same argument runs in a thousand different comment threads, none of which connect to each other. Users waste millions of hours typing the exact same rebuttal that someone typed yesterday. The best counter-argument gets buried on page ten of a forum thread, never to be seen again. And because the system has no shared memory, every debate resets to zero each day.
We are not short on intelligence. We are short on organization. The intellectual work has largely already been done in studies, investigations, expert analyses, and years of debate. The problem is that none of it accumulates. (For the broader diagnosis see the Problems hub.)
The Solution: One Belief, One Page
We use Natural Language Processing and community moderation to identify when people are saying the same thing in different ways. Statements that make the same underlying claim get consolidated into a single Belief Page that becomes the canonical home for every argument and piece of evidence relevant to that claim.
Example: Ford Truck Reliability
Instead of five separate arguments scattering across five threads, the system groups them into one:
| User Input | System Action | Result |
|---|
| "Ford trucks aren't reliable." |
→ MERGE → |
Single Belief Page: "Ford trucks have below-average mechanical reliability." (All arguments and evidence from the inputs are combined here.) |
| "Ford pickups break down too much." |
| "You can't count on a Ford to last." |
Before a user posts, the system checks the vector database for semantically similar existing beliefs and prompts them: "It looks like you're trying to say [Existing Belief]. Would you like to add an argument to that page instead?" Wordsmithing does not produce a new argument node. It produces a restatement of an existing one, which the system can recognize and consolidate.
What You See on a Belief Page
When you view a Belief Page, you see a living map of reasoning. The page is dynamic, sorted by the ReasonRank algorithm, with reasons to agree and reasons to disagree displayed in separate columns and sorted by quality.
This is the central design move of the entire platform. Two people who disagree about Ford truck reliability are not asked "are you pro-Ford or anti-Ford." They are asked which of these specific arguments they find compelling, and what evidence each one rests on. The conversation moves from tribal alignment to argument-level analysis, which is the only frame in which evidence can actually do its job.
Each argument carries a contribution score derived from two inputs:
Contribution Score = Linkage Score × Truth Score
The Linkage Score asks how relevant this reason actually is to the main belief. The Truth Score asks whether this reason is itself supported by evidence. An argument that is relevant but factually weak gets a moderate score. An argument that is factually strong but irrelevant to the specific belief also gets a moderate score. Only arguments that are both relevant and well-supported rise to the top. Weak arguments fade automatically. The best counter-argument is never buried on page ten again.
The Web of Beliefs: Upstream and Downstream
Beliefs do not exist in a vacuum. Every page displays the logical lineage of the idea, both where it came from and where it leads. This is what allows ReasonRank to work at all: scores propagate up and down the network because every belief knows which other beliefs depend on it and which it depends on.
Upstream (Assumptions)
More general beliefs that support this one.
These are the foundational principles or facts you must accept to hold this belief.
Example: To believe "We should cut taxes," you likely hold the upstream belief "Government spending is inefficient."
|
Downstream (Conclusions)
More specific beliefs that this one supports.
These are the implications. If this belief is true, what else must be true?
Example: If "Government spending is inefficient" is true, it supports the downstream belief "We should privatize the postal service."
|
When new evidence weakens an upstream belief, every downstream belief that depended on it weakens automatically. When a sub-argument gets debunked, its parent argument loses score. The network keeps the bookkeeping current. Nobody has to remember to update the conclusion when its premise changes; the system does it.
How Beliefs Are Organized: Three Spectrums
A belief is not just a flat text file. It is a data point located on three spectrums simultaneously, which lets users find the exact nuance they're looking for rather than wading through everything ever said on a subject.
Spectrum A: Positivity / Negativity (Valence)
Does this belief portray the subject as good or bad? Beliefs about Ford trucks span the full range, from extremely positive ("Ford makes the best trucks in history") through mildly positive ("Ford trucks are generally reliable"), neutral ("Ford is a truck manufacturer"), mildly negative ("Ford trucks have some maintenance issues"), to extremely negative ("Ford trucks are dangerous junk"). A reader looking for the strongest case in either direction can navigate straight to that end of the spectrum without scrolling through the moderate middle.
Spectrum B: Specificity (The Abstraction Ladder)
Is this a broad principle or a specific case? Beliefs range from the highly general ("Corruption undermines government") through the moderately general ("U.S. Presidents are often corrupt") and a baseline level ("The Clinton Administration had ethical issues") down to the highly specific ("Bill Clinton accepted illegal donations in 1996"). The same evidence may support a specific belief strongly while supporting the general version only weakly, and the spectrum lets the system represent that distinction explicitly.
Spectrum C: Strength / Intensity (The Confidence Claim)
How forceful is the claim? Stronger claims require stronger evidence. A weak claim ("This product might have some glitches"), a moderate claim ("This product is not very smart"), and an extreme claim ("This product is completely stupid and useless") all require different amounts of evidence to substantiate. The system tracks intensity so that the evidentiary bar scales with the claim being made.
Ready to help build it? Contact me to contribute.
Related Pages
Comments (0)
You don't have permission to comment on this page.