Home > Algorithms & Scores

The ISE Algorithms: A Complete Map

This is the index to every algorithm that turns a pile of pro and con arguments into a single, traceable score. If you want the plain-English story of why argument scoring matters, start with ReasonRank. This page is the engineering counterpart: what each score is, how it is computed, and where the full derivation lives. Nothing here is a black box. Every number on the platform decomposes back into the scores below.

The Core Recursion

ReasonRank borrows PageRank's central trick. A web page earns authority from the pages that link to it; a claim earns its score from the arguments that support and oppose it. The difference is that PageRank only counts positive links, while ReasonRank runs two channels, pro and con, and subtracts one from the other. The score of any node is built from the scores of its children, so when a foundational premise weakens, every conclusion resting on it weakens automatically.

That final ratio is how a score stays bounded between 0 and 1 without any artificial squashing: it is simply the share of total weighted support that is positive. The edge weight linkage × importance × uniqueness is the gate. A true but irrelevant point has near-zero linkage and contributes nothing. A restated point has near-zero uniqueness and contributes nothing. Either gate alone can zero out a contribution. Full derivation: Argument Scores from Sub-Argument Scores.

The Eleven Scores

The number at the top of the tree synthesizes a stack of interdependent scores, each measuring a different dimension of argument quality. They fall into three groups.

Fundamental Scores (argument and evidence quality)

• Truth Scores combine two independent checks: whether the logic holds (Logical Validity) and whether the facts check out (Verification). An argument can fail on either, and the system catches both.
• Linkage Scores test whether the evidence actually proves this conclusion or just points in the same general direction. Most bad reasoning fails here, not on the facts.
• Importance Scores separate truth from relevance, so a mountain of minor true points cannot bury one decisive counterargument by sheer numbers.
• Evidence Scores rate the source material on a tiered quality framework (T1 through T4). A peer-reviewed meta-analysis and a confident tweet do not weigh the same.
• Objective Criteria Scores measure performance against benchmarks that do not depend on values or ideology, the kind both sides can agree to in advance. Each criterion is rated on four sub-dimensions: validity, reliability, linkage, and independence (is the data source free of conflicts of interest).
• Confidence Stability Scores track how settled a score is under sustained scrutiny. A score that holds steady as new arguments arrive means more than one that lurches with every new entry.

Administrative Scores (anti-gaming, not quality)

These prevent repetition and source confusion from inflating the fundamental scores. The de-duplication rule is one line: an argument's contribution is scaled by its uniqueness, where uniqueness = 1 − (similarity to the most similar prior argument). A point 90% identical to one already on the board contributes 10%. The first to make a point is never penalized for being first. These anti-gaming scores sit under the Duplication Scores umbrella.

• Media Truth Scores and Media Genre and Style Scores flag editorializing, sensationalizing, or misleading sources even when the underlying facts are accurate.
• Topic Overlap Scores stop the same point from counting ten times because ten people phrased it ten ways. Repetition is not confirmation. This also powers the Related Pages feature.
• Belief Equivalency Scores detect when two differently-worded beliefs make the same underlying claim (mechanical match at or above 85% counts as identical; otherwise a blend of token and semantic similarity), so the platform links them and surfaces the better-argued version instead of running parallel debates.

Subsystem Scores

• Cost or Benefit Likelihood Scores power the Automated Cost-Benefit Analysis. Each projected cost or benefit gets its own argument tree, and its probability of actually occurring is read off the strength of that tree rather than guessed by a single analyst.
• Legal Support Scores measure how strongly existing laws and precedents agree with a belief.

Domain-Specific Scores (media and books)

The same framework extends to creative and published work. The Book Logical Validity Score and its siblings score books and films, and Media Linkage Scores carry two readings: how central a theme is to a work, and how much that work has moved a real-world belief (the way Silent Spring moved environmental policy).

The Three Continuums

Before a belief can be scored, it has to be matched with the other beliefs that are saying the same thing. The platform places every belief on three independent axes so that variants cluster onto one page instead of fragmenting into a hundred near-duplicate debates.

The axes are independent on purpose. A claim can be strongly negative and weakly stated at the same time, so direction and volume have to be measured separately. Plotting a percentage on each lets the system group beliefs that mean the same thing at different strengths. See the full positivity framework and the topic continuum layout used on every topic page.

How a Claim Gains or Loses Weight

Because the gates multiply, weaknesses compound rather than average out. A claim built on a debunked study, an irrelevant fact, and a logical fallacy takes three separate hits at once: one against its Evidence Score, one against its Linkage Score, and one against its Logical Validity. Multiplied together, the claim sinks fast, and every step of the sinking is visible. Misinformation does not get banned here. It loses on the merits, in public, where anyone can challenge the scoring and see exactly which argument failed and why.