Feature Overview: Figure & Table Analysis

Figure & Table Analysis is an editorial decision support feature in Alchemist Review that automatically analyzes the figures and tables in a manuscript, helping editors identify potential issues before peer review. When a manuscript is processed, Alchemist Review extracts each figure and table and runs a series of applicable quality, reference, and statistical checks. Results are organized by asset, making it easy to review visual and tabular evidence alongside the rest of the manuscript analysis.

Rather than replacing editorial judgment, Figure & Table Analysis helps prioritize which figures and tables deserve a closer look. Every flagged result includes a specific explanation so editors can quickly verify the finding and decide how to proceed.

🔍 What It Does

For every extracted figure and table, Figure & Table Analysis:

• Runs all applicable automated checks for that asset
• Groups results by individual figure or table
• Highlights potential issues using the same severity system as the rest of Alchemist Review
• Explains why each issue was flagged, allowing editors to verify findings directly against the manuscript

Not every check applies to every figure or table. Checks that are not relevant to a particular asset are skipped rather than flagged.

📍 Where to Find It

Figure & Table Analysis is available from its own dedicated tab on the manuscript page alongside the other Alchemist Review analysis tabs.
Results are organized by severity so the most important issues appear first.

From the hover menu on any result, editors can:

• Copy to Clipboard — copy the result text for use elsewhere
• Feedback — mark a result as a Good Response or Bad Response to help improve the feature over time

The Ask AI Chat action available elsewhere in the manuscript isn’t currently supported for figure and table results.

✅ The Checks

Every figure and table is run through the checks that apply to it. A check that doesn’t apply to a particular asset is skipped rather than flagged (see How Issues Are Flagged below).

1. Missing axes Confirms that chart-type figures—bar charts, scatter plots, line graphs, and similar—have labeled axes. A figure is flagged when the x-axis or y-axis is missing a label, leaving the reader unable to tell what variable is being plotted. Non-chart figures such as diagrams, photographs, schematics, and flowcharts are exempt.
2. Missing units Confirms that quantitative data is reported with units of measurement. On charts, axis labels should specify units (for example, “Time (s)” or “Concentration (mg/mL)”). In tables, numeric columns should declare their units in the header or caption. Dimensionless quantities such as ratios, counts, and p-values are exempt.
3. Legibility Confirms that any text embedded in a figure—axis tick labels, legends, in-panel annotations, data labels—is readable at the resolution provided. A figure is flagged when text is too small, blurry, pixelated, cut off, or otherwise unreadable. Figures with no embedded text are exempt.
4. Missing references Confirms that every figure and table in the manuscript is cited at least once in the body text. An asset is flagged when it appears in the manuscript but is never referenced in the narrative—typically a sign of an orphaned asset, or an in-text callout the author meant to add.
5. Unreferenced figures and tables The complement to Missing references. Confirms that every in-text citation (for example, “see Fig. 3” or “Table 2 shows…”) resolves to an actual figure or table in the manuscript. A citation is flagged when it points to an asset that doesn’t exist—usually a numbering error, a deleted asset, or a typo in the callout.
6. Statistical checks Runs a set of established forensic-statistics tests on numeric table data to surface values that are mathematically inconsistent with how they should have been generated:

• • GRIM (Granularity-Related Inconsistency of Means) — checks whether a reported mean is achievable given the sample size and the granularity of the underlying scale (for example, integer Likert responses). A mean that can’t arise from any valid combination of integer responses is flagged.
  • SPRITE (Sample Parameter Reconstruction via Iterative Techniques) — checks whether a reported mean and standard deviation can occur together, given the sample size and the bounds of the response scale. Impossible mean/SD combinations are flagged.
  • Benford’s Law — checks whether the leading digits of a large numeric column follow the expected Benford distribution. Significant deviations can point to fabricated or heavily rounded data, though they can also reflect legitimate properties of the underlying measurement.

🚦 How Issues Are Flagged

Figure and table results use the same severity system as the rest of Alchemist Review. Each result carries one of four statuses:

• Unusual PDF formatting
• Scanned or image-only documents
• Highly complex table layouts
• Publisher-specific formatting conventions