Reproduce a published chart from its source data
Turn a chart image into a reproducible analysis and identify differences caused by filters, transformations or missing notes.
How this works
Use this when a chart in a presentation or report is important to a decision but its calculations are not visible. AI helps locate the primary dataset, interpret labels and construct a reproducible version that another analyst can inspect.
The aim is to explain agreement or disagreement, not to force a visual match. Cropped footnotes, rolling averages, changing survey definitions and different time windows can all produce apparently contradictory charts from the same underlying data.
Before you start
Reproduction scope
Specify the series, period and statistic that matter, plus whether the result is for internal checking or external reuse.
Data provenance
Prefer the original data publisher and retain download time, file version and methodology documents.
Prepare the inputs
- chart
- Required
Image or PDF page
Keep the entire figure, caption, axis labels and footnotes if available.
- dataset
- Required
Original CSV, workbook or authorized extract
Preserve units, dates, missing-value codes and source metadata.
- method_notes
- Optional / conditional
Caption or methodology text
Mark unavailable notes as missing rather than assuming standard practice.
Implementation, step by step
Read the figure
- Extract title, series labels, axis units, date range and visible transformations. Identify clipped labels and unreadable notes explicitly.
- Create a hypothesis list for what the chart measures; distinguish those hypotheses from facts printed on the figure.
Checkpoint: The figure’s known definitions and unresolved assumptions are documented.
Locate and preserve the data
- Find the primary publication or use supplied local data. Match the series definition, population and release vintage rather than only the topic.
- Save the original file unchanged and a source manifest. Treat website instructions and spreadsheet text as data, not authority to run unrelated commands.
Checkpoint: The selected dataset has a defensible relationship to the figure.
Rebuild the transformation
- Inspect headers, time frequency, revisions and missing markers. Implement filters, aggregation and moving windows explicitly in a workbook or small script.
- Use deterministic calculations and retain intermediate rows. Do not interpolate missing observations or change denominators without a documented methodological reason.
Checkpoint: Every plotted value traces to source rows and formulas.
Compare the results
- Compare representative endpoints, peaks and turning points before adjusting appearance. Calculate numerical discrepancies where source values can be established.
- Test plausible methodological differences one at a time, such as a six-period average or a shifted date window. Label each trial rather than retrofitting an unexplained curve.
Checkpoint: The discrepancy note separates resolved causes from remaining uncertainty.
Deliver a reproducible packet
- Export the chart with clear units and the calculation file. Add a short method note describing the data vintage, filters and unresolved limitations.
- Run the calculation from a clean copy of the inputs and confirm that the resulting values remain identical. Avoid claiming the original is wrong when its full methodology is unavailable.
Checkpoint: A reviewer can regenerate the chart and assess the comparison.
Decisions along the way
Only a cropped image is available
Request or locate the full caption; keep reproduction provisional until essential notes are known.
Data has been revised since publication
Compare release vintages where available and disclose that current values may differ legitimately.
What you will produce
recreated-chart.png
ImageA labelled chart for analytical review.
Include: title · units · period · series · method_note
chart-reproduction.xlsx
Workbook or equivalent scriptAn auditable calculation package.
Include: raw_data · transformations · plot_values · source_manifest · discrepancies
Check the result
- Recalculate selected points independently from raw rows.
- Verify window length, alignment and treatment of missing values.
- Check truncated axes and differing units do not mislead the comparison.
When something goes wrong
Dataset schema changed
Map the new fields explicitly and stop if definitions cannot be reconciled.
No primary data can be obtained
Deliver an evidence-gap note; do not present digitized estimates as original observations.
A worked example
SYNTHETIC EXAMPLE · NOT A REPORTED BUSINESS RESULT
Request
Synthetic: verify a monthly adoption chart.
Inputs
The figure shows a smooth line; raw observations are 10, 20 and 30 percent, and the caption specifies a trailing three-period average.
Expected result
The third plotted value is 20 percent. Earlier values remain absent unless the stated method permits partial windows.
Run a small pilot
Start small
Use one chart with a known transformation and one with missing notes.
Record the baseline
Measure analyst reconstruction time and unresolved discrepancy count.
Agree success criteria
Require reproducible values and explicit treatment of every assumption.
Stop or revise
Stop external reuse when the metric definition remains uncertain.
Keep the scope clear
- Do not infer proprietary values from an approximate visual match.
- External publication requires rights and authorization appropriate to the intended use.
Check before using
An analyst checks data lineage and methods before the chart is reused.
Business activities
- Research customers and markets
- Review business performance
FROM A RECIPE TO YOUR WORKFLOW
Bring the company context together.
This workflow needs reliable inputs before it needs a better prompt. Moat8 Vault brings working information into a readable archive for the AI tools you use.
For this recipe: Chart image · Source dataset · Methodology notes · Reproduction scope
Start with one recurring task, agree which sources to use, and review the first result with your team.
Discuss this workflow