The five classes, the spam rule, and falsifiability
Every intervention arises from an observed gap. archon score check rejects an intervention whose addresses field names a question that has no entry in the Evidence Gap Register. Do not recommend something because it is fashionable in answer-engine optimization.
Interventions come in five classes
| Class | Examples | Addresses |
|---|---|---|
Entity |
One canonical identity page; one set of names and founding dates; Person-Organization relationships; author pages; product ownership; |
Entity Clarity gaps, conflicting facts |
Evidence |
Case studies the client agreed to; documented outcomes; references; public talks; research; technical documentation; publications; authoritative external citations |
Evidence strength and corroboration gaps |
Content |
Capability, problem, and use-case pages; a comparison page only where it is genuinely useful; FAQs that answer real questions; glossaries; book pages; methodology pages |
Broken |
Technical |
Repair 404s; remove redirect chains; canonical URLs; structured data; sitemap; metadata; publication metadata; accessible HTML; crawlability |
Technical Legibility gaps |
Relationship |
Connect people to organizations, organizations to products, authors to books, concepts to publications, capabilities to case evidence, services to outcomes, projects to owners |
Broken |
Every intervention records the change, the gap, and what would show it failed
change, the question ids it addresses, a hypothesis, the result that would show it did not work (no_effect_if), why it publishes real evidence (legitimate_because), a window, an owner, a status, and, after the retest, an outcome.
Pages written to manufacture evidence are not interventions
Publish only evidence that exists. Never propose, and never accept as an intervention:
-
thousands of thin, generated pages;
-
FAQ farms and mechanical restatement of questions;
-
city or region pages with no unique value;
-
synthetic comparison pages, or pages about competitors with no use to a finder;
-
fake or solicited-for-effect reviews, invented expertise, unsupported "best" pages;
-
keyword stuffing or text written for an answer engine and not for the finder.
Every hypothesis gets the observation that would disprove it
For each intervention ask what result would convince us it did not work, write it down before shipping, and look for it at the retest. For each theory of why a subject is absent, look for the observation that would disprove it.
Hypothesis: the subject is missing because it lacks enough content. Disconfirming observation: a subject with fewer pages is included consistently, or a subject with many pages is not. Then the hypothesis needs to be replaced by a better one.
Report interventions with no measurable effect, and report regressions.
The 30/60/90 order is a default the evidence can override
The 30/60/90-day plan puts factual contradictions, broken relationships, and obvious technical failures first, then high-value evidence and capability-to-outcome connections, then deeper content gaps and external corroboration. Let the evidence set the order: a gap that blocks the highest-weight question comes first, whatever its class.