---
title: "Memory & Cognition"
source: "/reference/memory-cognition"
description: "Rules for reducing unnecessary mental work, supporting recognition, preserving context, and helping users build reliable expectations about the system."
---
# Memory & Cognition

Rules for reducing unnecessary mental work, supporting recognition, preserving context, and helping users build reliable expectations about the system.

> Rules for reducing unnecessary mental work, supporting recognition, preserving context, and helping users build reliable expectations about the system.

Every interface asks the user to perceive information, hold some of it temporarily, connect it to prior knowledge, and decide what to do next. Problems appear when the product makes the user remember information the system already knows, infer hidden state, translate inconsistent terminology, or reconstruct relationships that could have remained visible.

Layr should treat cognitive effort as a budget. Some of that budget belongs to the user's real task: comparing investments, configuring infrastructure, reconciling records, diagnosing a production incident. The product adds avoidable cost when its own structure, language, sequencing, or state management becomes another problem to solve.

This category therefore emphasizes externalized context, coherent concepts, recognition over recall, and stable mental models. It also includes explicit safeguards against common misuse of psychological research, especially rigid interpretations of Miller's Law.

## Principles in this category

- [Chunking](/__layr/pages/reference/chunking.md) — Information becomes easier to process when related pieces are organized into meaningful units that can be understood as one structure.
- [Cognitive Bias](/__layr/pages/reference/cognitive-bias.md) — Judgment is shaped by predictable mental shortcuts, prior beliefs, framing, defaults, salience, and context.
- [Cognitive Load](/__layr/pages/reference/cognitive-load.md) — Every task consumes limited mental resources; interface complexity adds cost when users must remember, infer, translate, compare, or recover information unnecessarily.
- [Mental Model](/__layr/pages/reference/mental-model.md) — People operate a product through an internal explanation of what the system is, what its objects mean, and what actions will cause.
- [Miller's Law](/__layr/pages/reference/millers-law.md) — Human short-term information processing is limited; the famous 'seven plus or minus two' finding should be treated as historical context, not as a universal UI item limit.
- [Working Memory](/__layr/pages/reference/working-memory.md) — Working memory temporarily holds and manipulates information needed for the task currently in progress.
- [Serial Position Effect](/__layr/pages/reference/serial-position-effect.md) — Items near the beginning and end of a sequence often receive stronger recall than items in the middle.

## Category model

Cognitive design is the management of information across time. A user may see the correct data and still struggle because it appears too early, disappears too soon, uses unstable names, or requires several pieces to be combined mentally. Layr should therefore inspect not only what is present on one screen but also what must survive navigation, interruption, comparison, and state changes.

A useful distinction for agents is between **domain thinking** and **interface thinking**. Domain thinking is the reasoning the user came to perform: evaluating a tradeoff, diagnosing a problem, configuring a policy. Interface thinking is work created by the product: remembering hidden values, translating terminology, reconstructing state, or figuring out which component owns which action. Layr should preserve necessary domain thinking while reducing avoidable interface thinking.

### Agent evidence hierarchy

Prefer direct evidence such as hidden dependencies, repeated input, inconsistent terminology, lost state, cross-screen recall, and missing summaries. Treat claims about memory limits or mental effort as lower-confidence unless the workflow clearly forces recall or reconstruction.

## Category review questions

- What information must the user remember instead of simply seeing it?
- Does the product use stable names and behaviors for the same concepts?
- Can users understand how objects relate and predict what an action will do?
- Are long or complex information sets organized into meaningful chunks?
- Does the design reduce avoidable reasoning without hiding information required for a correct decision?

## How agents should use this file

Treat these rules as review criteria, not as automatic violations. A finding is valid only when the interface provides observable evidence that the principle applies and that the current implementation creates a meaningful usability, comprehension, accessibility, or interaction cost.

For every finding:

1. Identify the exact component, state, route, or interaction.
2. State the observable evidence.
3. Name the principle that applies.
4. Explain the user consequence in the current context.
5. Estimate severity from the consequence, frequency, and reversibility.
6. Recommend the smallest change that addresses the issue while preserving the product's existing design system.
7. Lower confidence when intent, frequency, audience, device, or product constraints must be inferred.
8. Do not produce generic advice such as "improve hierarchy" or "make this clearer" without specifying what is wrong and why.

### Finding contract

```yaml
finding:
  rule_id: string
  location: string
  evidence:
    - observable fact
  impact: string
  severity: low | medium | high | critical
  confidence: low | medium | high
  recommendation:
    change: string
    rationale: string
    preserve:
      - existing design-system behavior
      - valid product constraints
```

A clean interface can legitimately produce zero findings.

## Agent routing hints

Route a review here when the user must remember, infer, translate, reconcile, or reconstruct information. These rules are especially relevant when the system already possesses context that the interface fails to expose.