--- type: skill lifecycle: stable inheritance: inheritable name: chart-interpretation description: Read any chart (image, HTML, screenshot) and extract insights, patterns, anomalies, bias, and narrative -- the reverse of visualization tier: standard applyTo: '**/*chart*,**/*interpretation*' currency: 2026-04-22 lastReviewed: 2026-04-30 --- # Chart Interpretation | Property | Value | | ------------ | --------------------------------------------------------------------- | | **Domain** | Data Analytics | | **Category** | Visual Analysis & Insight Extraction | | **Components** | SKILL.md + chart-interpretation.instructions.md + interpret.prompt.md | | **Depends** | data-visualization (chart type knowledge), data-analysis (validation) | ## Overview The reverse of data visualization. Instead of data → chart, this skill reads chart → insights → narrative. It extracts meaning from existing charts (screenshots, images, HTML, Power BI reports) and produces structured analysis adapted to the target audience. The cardinal rule: **read what the chart says, then read what it doesn't say**. The visible data points tell one story; the missing context, truncated axes, and suppressed categories tell another. ## Module 1: Chart Type Recognition Identify the chart type to determine the correct reading strategy. | Chart Type | Key Visual Features | Reading Strategy | | ---------------- | -------------------------------------- | ----------------------------------------- | | Bar / Column | Rectangular bars, one axis categorical | Compare bar lengths, check sort order | | Horizontal Bar | Bars extend left-to-right | Rank comparison, read labels first | | Line | Connected points over axis | Follow trend direction, find inflections | | Area | Filled region under line | Volume over time, stacking if multiple | | Pie / Donut | Circular segments | Part-to-whole, count segments, check % | | Scatter | Points in x-y space | Look for clusters, outliers, trend line | | Bubble | Scatter with size encoding | Three dimensions: x, y, size | | Histogram | Bars touching, x is continuous | Distribution shape, skew, outliers | | Heatmap | Color grid | Pattern density, row-column relationships | | Treemap | Nested rectangles | Hierarchical proportions | | Sankey | Flow ribbons between stages | Volume flow, biggest paths | | Box Plot | Box + whiskers | Median, IQR, outlier dots | | Network | Nodes + edges | Clusters, hubs, isolates | | Violin | Mirrored density curves | Distribution shape + density | ## Module 2: Visual Decoding Extract data from visual encodings: | Encoding | What to Read | Precision Level | | ---------------- | ------------------------------------- | ------------------- | | Position (axis) | Exact values from gridlines/labels | High (if labeled) | | Length (bar) | Relative magnitude between items | High | | Color hue | Category membership | Categorical only | | Color intensity | Value magnitude in sequential scheme | Medium | | Size (area) | Third variable (bubble, treemap) | Low (area perception is poor) | | Angle (pie) | Proportion (poor human accuracy) | Low | | Slope (line) | Rate of change | Medium | ## Module 3: Pattern Detection | Pattern | What to Look For | Significance | | ---------------- | --------------------------------------- | -------------------------------------- | | **Trend** | Consistent upward/downward direction | Growth, decline, momentum | | **Inflection** | Direction change point | Market shift, intervention effect | | **Plateau** | Flat region after growth/decline | Saturation, stabilization | | **Cluster** | Groups of points in scatter/network | Natural segments, sub-populations | | **Outlier** | Points far from the main group | Anomaly, error, or special case | | **Periodicity** | Repeating pattern at intervals | Seasonality, weekly cycle | | **Gap** | Missing data or discontinuity | Data quality issue or deliberate omission | | **Skew** | Asymmetric distribution shape | Non-normal population, concentration | | **Dominance** | One item >> all others | Power law, market leader, outlier | ## Module 4: Misleading Visual Detection Check every chart for these deceptive patterns: | Deception | How to Detect | Actual Impact | | -------------------------- | ---------------------------------------------------------- | -------------------------------------- | | **Non-zero baseline** | Y-axis starts above 0 | Exaggerates differences (sometimes 2-5x)| | **Truncated axis** | Axis range excludes data or starts mid-range | Hides context, magnifies small changes | | **Dual axes** | Two Y-axes with different scales | Implies correlation where none may exist| | **3D effects** | Perspective distortion on bars/pies | Area comparison becomes inaccurate | | **Cherry-picked range** | Time window starts/ends at convenient point | Hides contrary trend outside window | | **Suppressed categories** | "Other" aggregates significant items | Hides important segments | | **Area distortion** | Variable-width bars, non-proportional icons | Size doesn't match value | | **Missing denominator** | Percentages without base size | 50% of 10 ≠ 50% of 10,000 | | **Reversed axis** | Values increase downward or rightward | Readers misread direction | ## Module 5: Structural Element Reading Always read these elements before interpreting the data: | Element | What to Extract | If Missing | | ---------------- | -------------------------------------- | --------------------------------------- | | **Title** | Author's intended takeaway | Chart lacks stated purpose | | **Subtitle** | Time range, filter condition, context | Context must be inferred | | **Axes labels** | What variables are plotted | Interpretation becomes guesswork | | **Legend** | Category-to-color mapping | Color meaning unclear | | **Annotations** | Author-highlighted insights | No guided reading | | **Data source** | Where the data came from | Credibility unknown | | **Date/time** | When data was collected/reported | Freshness unknown | ## Module 6: Narrative Extraction Convert visual observations into prose at three audience levels: ### Executive Summary (30 seconds) ``` 3 bullets maximum: • [Primary insight — the main takeaway] • [Supporting evidence — the strongest proof point] • [Recommendation or implication — what to do about it] ``` ### Detailed Analysis (2-3 minutes) ``` The chart shows [chart type] plotting [X variable] against [Y variable] for [time range / scope]. Primary finding: [Main pattern or insight with specific numbers] Supporting observations: - [Pattern 1 with evidence] - [Pattern 2 with evidence] - [Anomaly or exception worth noting] Context and caveats: - [What the chart doesn't show] - [Potential biases or limitations] - [Comparison to benchmarks if available] ``` ### Talking Points (presenter-ready) ``` "What you're seeing here is [explain the main pattern in plain language]." "The key number to focus on is [highlight], which tells us [implication]." "What's interesting is [surprise or anomaly] — this suggests [hypothesis]." "The action item here is [recommendation]." ``` ## Module 7: Confidence Rating Rate interpretation confidence honestly: | Level | When | Signal to User | | -------- | ----------------------------------------------- | ---------------------------------------- | | **High** | Clear labels, clean data, familiar chart type | "The chart clearly shows..." | | **Medium** | Some inference needed (unlabeled, partial data) | "Based on visual estimation..." | | **Low** | Ambiguous visual, missing context, blurry image | "This appears to show, but verify..." | ## Module 8: Follow-Up Recommendations After interpreting, suggest what would strengthen the analysis: | Suggestion Type | Example | | ----------------------- | ---------------------------------------------------------- | | Missing variable | "Add cost data to see if revenue growth is profitable" | | Time extension | "Extend to 24 months to confirm the seasonal pattern" | | Segmentation | "Break this down by region to check for Simpson's Paradox" | | Alternative chart | "A scatter plot would better show the correlation" | | Baseline addition | "Add a target line to show performance vs. plan" | ## Module 9: CSAR Loop Integration Use the Dialog Engineering CSAR Loop for structured chart reading: | Phase | Action | | ------------ | ----------------------------------------------------- | | **Clarify** | What chart type? What variables? What time range? | | **Summarize**| State the main finding in one sentence | | **Act** | Extract specific data points, patterns, anomalies | | **Reflect** | What's missing? What would I want to see next? | ## Anti-Patterns | Anti-Pattern | Problem | Fix | | ------------------------- | ------------------------------------------ | ---------------------------------------- | | Describing, not interpreting | "This is a bar chart" (no insight) | Say what the bars MEAN, not what they ARE| | Ignoring the title | Missing the author's intended message | Read title first -- it's the thesis | | Over-precision from visual | "Revenue is exactly $4,237,892" | Estimate from visual: "roughly $4.2M" | | Missing bias check | Accepting the chart at face value | Always scan for misleading elements | | Single-lens reading | Only one interpretation offered | Provide primary + alternative reading |