Chart prompts: inspect axes, units, and missing series first

Operational case

A dashboard screenshot plots weekly late-delivery rates for two depots. Its vertical axis starts at 3 percent rather than zero, so the difference between 4.0 and 7.5 percent looks much larger than it would on a full scale. A pale third line is visible, but its legend is clipped in the screenshot. The assistant asks for the source table and full legend before naming that series or stating exact values. Once the table arrives, it reports the rates with denominators and notes that 23 unresolved records are outside the plotted late-rate cohorts. It does not claim that a staffing change caused the difference.

Visual intake: x=week; y=late rate (%); y-min=3%, not 0%.
Series: north, south, third series label clipped.
Table check: north=4.00%; south≈7.48%; unknown status=23.
Caption: observed rates and denominators; no causal claim.
If source table absent: estimated visual values, not exact numbers.

Performance and operating cost

Reading a source table of P plotted points is O(P) for a simple validation pass, while image interpretation adds model cost and can fail on low-resolution labels. Avoid repeated image prompts when a structured export is available. Keep the source table, chart configuration, and generated caption together so a chart refresh cannot leave an obsolete narrative on screen. Measure unit errors, missing-series omissions, and false exactness separately. A larger screenshot is not a substitute for the underlying data when the decision depends on precise values.

Common Mistakes

• Do not treat a truncated axis as evidence of an extreme change.
• Do not name an obscured series from its color alone.
• Do not report image-derived estimates as exact table values.

Connected lessons

• Prompt engineering applications
• Prompt Engineering
• Image prompts: verify claims against named regions and crops
• Multimodal prompts: separate what an image shows from what it suggests
• Report summaries: preserve contrary results and missing data
• Dataset intake prompts: define a column before analyzing it
• Classification prompts: write the label boundary first
• Aggregation prompts: pin the denominator and recompute the rate
• Project: verify a shipment performance report
• Prompt engineering for data workflows

Continue with: Spreadsheet release checks: recalculate, reconcile, then inspect.

Continue with: Slide prompts: one claim, supporting evidence, usable order.

Continue with: Chart prompts: choose marks and scales by data type.