Home > Page Design > Algorithms > ReasonRank > Argument Scores from Sub-Argument Scores
Argument Scores from Sub-Argument Scores
"A conclusion is the place where you got tired of thinking." — Steven Wright
It doesn't have to be. With the right infrastructure, analysis can continue while you sleep. The problem isn't that people refuse to think. It's that we keep starting over — relitigating the same arguments every election cycle because there's no shared structure connecting arguments to sub-arguments to evidence to conclusions. This page explains the core mechanic that makes a permanent structure possible.
In one sentence: Arguments receive their scores from the strength of the arguments beneath them. Truth flows upward through a network of linked claims, allowing evidence to automatically update every dependent conclusion.
Arguments in the ISE are organized as a tree, not a list. At the root sits the conclusion being evaluated. Branching down from it are the arguments for and against. Each of those arguments has its own branches — the sub-arguments that support or undermine it — and those sub-arguments have branches too, all the way down to raw evidence.
Every node in this tree is itself a belief with its own page. "Roundabouts reduce fatal accidents" isn't just a bullet point under "We should build roundabouts." It's a full claim with its own supporting evidence, its own opposing arguments, and its own score. That score feeds upward into its parent. The rebuttal to any bad argument is permanently attached to the claim it refutes — always one link away, never buried in a comment thread nobody reads anymore.
Technically the structure is a Directed Acyclic Graph (DAG): directed because scores flow upward from evidence to conclusions, and acyclic because the system prevents circular reasoning. An argument cannot be its own ancestor. Navigation between pages is bidirectional, but scoring always flows one way: up.
The Three Factors That Determine an Argument's Influence
Each sub-argument influences its parent through three independent scores, and all three must be strong for a sub-argument to significantly move a conclusion. A claim can be true but irrelevant, relevant but weakly supported, or important but poorly linked to the conclusion it's supposed to prove. The scoring system captures these distinctions explicitly.
- The Truth Score asks: is this sub-argument actually true? It's derived recursively from the sub-argument's own foundations. For leaf nodes at the bottom of the tree, it comes from Evidence Scores based on study design, sample size, replication, and peer review.
- The Linkage Score asks: if this sub-argument is true, does it actually prove the parent? "Construction causes temporary delays" might be 100% true, but its linkage to "we should never build roundabouts" is weak. High truth, low linkage, small contribution.
- The Importance Weight asks: even if true and relevant, does this factor actually matter to the decision? A minor convenience factor shouldn't outweigh a major safety factor even if both are equally well-supported.
- A fourth factor — Independence — prevents correlated arguments from artificially inflating scores. Four claims built on the same dataset shouldn't count four times. Arguments that share underlying evidence have their contributions scaled down proportionally.
The Contribution Formula
Each sub-argument's contribution to its parent is:
Contribution = Truth Score × Linkage Score × Importance Weight × Independence Score
Algorithm summary:
- Evidence determines Truth Scores.
- Truth Scores combine with Linkage, Importance, and Independence to produce Argument Scores.
- Argument Scores combine to produce Conclusion Scores.
- Any change propagates upward through the network automatically.
Add up all the positive contributions, subtract all the negative ones, and you have the parent argument's net score. That score becomes the Truth Score input for the argument above it, repeating all the way up to the conclusion. Weak arguments automatically fade. Strong arguments dominate. And changing evidence anywhere in the tree automatically updates every conclusion above it.
Some arguments only activate when a prerequisite belief is sufficiently established. For example, "we must regulate AI autonomous weapons" only matters if AI can actually deploy autonomous weapons autonomously. These conditional dependencies add a fifth multiplier — the Truth Score of the prerequisite belief — ensuring that argument chains built on uncertain foundations are automatically discounted rather than treated as fully active.
A Worked Example
The conclusion: "We should replace stop signs with roundabouts."
| Sub-Argument (Support) | Truth | Linkage | Importance | Contribution |
| "Reduces fatal accidents by 37%" |
80% |
+90% |
0.9 |
+64.8% |
| "Improves traffic flow" |
70% |
+85% |
0.5 |
+29.75% |
| Sub-Argument (Oppose) | Truth | Linkage | Importance | Contribution |
| "Confusing for elderly drivers" |
50% |
-70% |
0.7 |
-24.5% |
| "Construction causes delays" |
100% |
-10% |
0.1 |
-1% |
| Net Score: |
+69.05% (Favor roundabouts) |
The construction delays row is 100% true — but because its linkage and importance are both low, it contributes almost nothing. Being technically correct about an irrelevant detail is not the same as making a good argument.
Now notice that "Reduces fatal accidents by 37%" has a Truth Score of 80%, not 100%. That 80% is the score of that claim's own belief page, calculated from the traffic studies supporting it and any methodological criticisms opposing it. If new research weakens the fatality reduction estimate and drops that Truth Score to 40%, the net score recalculates automatically to +37.75%, and that change ripples upward through every higher-level conclusion that depends on roundabout safety. The analysis keeps running. The scores keep updating. The strongest arguments keep rising.
- Semantic clustering groups similar arguments so the same claim can't be counted multiple times under different wordings. "Roundabouts reduce crashes" and "Roundabouts improve safety" get merged into one node, not two. You can't inflate a score by restating the same point fifty different ways.
- Independence Scores catch a subtler version of the same problem. Four arguments built on the same underlying dataset aren't four independent reasons — they're one reason wearing four hats. Independence weighting scales their combined contribution down to reflect the actual evidentiary base, not the number of phrasings an organized group could generate.
- Linkage Scores are evaluated independently from Truth Scores. A true claim still has to prove it's relevant. Someone can't boost a conclusion by piling on true-but-irrelevant sub-arguments — the low linkage score neutralizes the contribution regardless of how well-evidenced the claim is.
- Community and expert review can challenge any score at any level of the tree. Flag a weak linkage, identify a missing counterargument, submit new evidence — the whole network recalculates from that node upward. Every revision is traceable.
Most importantly: no score in the Idea Stock Exchange is authoritative. Every score is the output of an argument that can itself be challenged. Clicking any number shows the sub-arguments and evidence behind it. There are no black boxes. If you think a score is wrong, the correct response is to add the counterargument and the evidence. The system doesn't suppress disagreement — it structures it. Every challenge strengthens the reasoning network instead of derailing it.
Related Pages
Comments (0)
You don't have permission to comment on this page.