This lesson opens the remediation module by reducing an audit backlog to six repeatable fix patterns, each tied to the scorecard dimension it moves and to the evidence that it works. It is the map the rest of Module 5 follows.
What this lesson covers
- The six fix patterns
- The research behind each one, with the numbers
- Mapping patterns to Clarity, Coverage, Consistency, Connection and Corroboration
- Effort against impact, and local against external work
- Where AI helps and where it must not decide
- Choosing the right pattern from the symptom
The six patterns are disambiguation, canonical naming, attribute completion, entity-first extractable structure, internal entity linking and off-site corroboration. The lesson defines each one, shows the demo site’s own example and names the dimension it moves. It then sets out why they work: the 2014 study that defined entity salience found raw keyword frequency scored below a simple first-sentence baseline, while position and coreference-aware counting did far better; a 2026 study found that how widely an entity is spread through a text predicts salience better than how many times it appears; and recent retrieval research shows that self-contained, sourced statements are retrieved and quoted more often.
The second half is about choosing. An effort-and-impact matrix shows which patterns to do first, and the lesson points out that five of the six are local edits you control, while corroboration depends on other people and takes longest. It then separates the work AI can take on — drafting at volume from audit fields, proposing links, generating markup — from the decisions it cannot make, and closes with a simple rule: read the symptom, name the pattern, and never campaign for an entity you could simply rename.
Each pattern is presented with the same three things: the symptom in the audit that points to it, the evidence that it works and the dimension of the score it should move when it does. That structure is what lets the later lessons in the module measure their own results rather than assert them, and it gives you a vocabulary for explaining a remediation plan to people who will never read the audit.
This is the first lesson of Module 5, Remediation: Fixing Entity Representation, in The AI-powered Entity Audit. It suits anyone with an audit backlog who needs to explain to a client or a team why each fix is on the list and which one comes first.