---
name: encoding-integrity
description: "Audit D3.js visualizations for honest data encoding. Use when the user says 'encoding integrity', 'check honesty', 'lie factor', or wants to verify data representation is truthful. Checks zero baselines, area scaling, dual-axis risk, binning bias, and metamorphic robustness (do visuals scale correctly when data changes?)."
---

# Encoding Integrity

A visualization has encoding integrity when it faithfully represents data now AND would continue to do so with different data. This skill audits both dimensions in a single pass.

## Part 1: Static Honesty

Check the current encoding for mathematical deception.

### Zero Baseline (Bars & Areas)

**Check:** If bar or area chart, does the Y-scale domain start at 0?
**Rule:** Bar/area charts encode data as length from baseline — baseline must be zero. Exception: line charts and dot plots focused on variance.

### Lie Factor

**Check:** If size encodes data, is `d3.scaleSqrt` (area-correct) used instead of `d3.scaleLinear` (radius-based)?
**Rule:** Radius-based scaling for data is a major encoding violation. Doubling radius quadruples area — lie factor of 2.

### Quantile Trap (Choropleths)

**Check:** Are color bin boundaries mathematically honest for the data distribution?
**Rule:** Flag `scaleQuantile` on skewed data if it hides extreme outliers. Suggest `scaleThreshold` or `scaleLog`.

### Dual-Axis Correlation Risk

**Check:** Does the chart use dual-Y axes to compare different units?
**Rule:** Flag as risk. If demonstrating dual-axis technique, score on execution (clear labels, distinct channels). Suggest index charts or small multiples.

### Smoothing Deception (KDE)

**Check:** Does the KDE bandwidth hide important features (bimodality, gaps)?
**Rule:** Flag smoothing that hides structure. Suggest histogram alongside KDE.

### Silent Gaps

**Check:** Does `line.defined()` handle missing values? Missing data drawn as a straight line implies false continuity.

### Honesty Checklist

| Pattern | Trigger | Fix |
|:---|:---|:---|
| Truncated baseline | Bar/area starts non-zero | `yScale.domain([0, ...])` |
| Radius deception | `scaleLinear` for circle size | `scaleSqrt` |
| Dual-axis fabrication | Two Y-scales, one chart | Index chart or small multiples |
| Binning bias | Arbitrary histogram bins | Sturges' or Scott's rule |
| Hidden outliers | `scaleQuantile` on skewed data | `scaleLog` or `scaleThreshold` |
| Silent gap | Missing data = straight line | `line.defined()` |

## Part 2: Metamorphic Robustness

Test whether visual properties hold correct relationships when data changes. Violated relation = bug in scales, generators, or data joins.

### Scaling Relation

**Transform:** Multiply all values by constant $k$.
**Invariant:** Visual dimensions scale by $k$ (linear) or follow the scale's transform.
**Catches:** Hardcoded domains, fixed max values, magic numbers. With synthetic data: code quality issue, not correctness bug.

### Permutation Relation

**Transform:** Shuffle the data array.
**Invariant:** Set of visual element properties remains identical (DOM order may change).
**Catches:** Missing key functions in `.data(data, d => d.id)` causing visual corruption after sort.

### Subset Relation

**Transform:** Remove one data point.
**Invariant:** Remaining elements stay visually identical to their source state.
**Catches:** Scales that recalculate domains from current subset, causing jumping axes.

### Shift Relation

**Transform:** Add constant to all values.
**Invariant:** Relative visual differences stay identical.
**Catches:** Accidental dependencies on absolute values instead of relative differences.

## Scoring Guide

Score 1-10 on combined encoding integrity:

| Score | Meaning |
|:-----:|:--------|
| 1-3 | Active deception (lie factor > 2, fabricated correlation) or fundamental encoding breakdown |
| 4-5 | Multiple honesty violations or severe robustness failures |
| 6-7 | Structurally honest but brittle — hardcoded domains, minor baseline issues. Typical for synthetic-data demos |
| 8-9 | Honest encoding with data-driven domains, correct area scaling, proper joins where needed |
| 10 | Publication-grade — handles edge cases, documents encoding choices, correct under all transformations |

### Synthetic Data Caveat

Most generated blocks use synthetic data. Score on structural correctness (scale choices, baseline rules, encoding math, domain derivation). Hardcoded domains matching synthetic data are acceptable at 6-7. Reserve highest confidence for real datasets.

## Invariants Checklist

| Check | Pass Signal |
|:---|:---|
| Zero baseline | Bars/areas start at 0 |
| Area encoding | `scaleSqrt` for size, not `scaleLinear` |
| Scale domain | `d3.max(data)` or `d3.extent` used, not magic numbers |
| Key functions | `.data(data, d => d.id)` present where updates happen |
| Gap handling | `line.defined()` for missing values |
| Binning | Data-driven bin thresholds |
| Dual-axis | Labeled, distinct channels, or avoided |

## References

- Tufte, Edward. *The Visual Display of Quantitative Information* (Lie Factor)
- Cairo, Alberto. *How Charts Lie* (Deception taxonomy)
- Segura, Sergio et al. "Metamorphic Testing: A Review of Challenges and Opportunities" (MT framework)
- Bostock, Mike. "Object Constancy" (Key functions and identity)