Figure Legend Auto-Generation
A figure legend should explain what is plotted, how many observations it represents, which test was used, and how statistical information is encoded. Licklider therefore assembles legend drafts from the same figure state used during export review.
STEP 1 - The Pitfall: Incomplete and Inconsistent Legends
Even when the analysis is correct, legends often omit critical details such as the error bar type, per-group n, correction method, or the disclosure context required to interpret the figure responsibly.
- Error bars are often unlabeled. Readers cannot infer whether a figure shows SD, SEM, or CI.
- n is often incomplete. Total n alone does not tell the reader how observations were distributed across groups.
- Statistical context is easy to underreport. Test name, correction method, sidedness, effect size, and CI are often split across multiple locations or omitted.
- Disclosure language drifts. If preprocessing or caveats are not carried forward, figure text can overstate certainty.
STEP 2 - Journal Requirement
Journal requirements differ in style, but they consistently expect legends to disclose the figure contents, statistical method, variability metric, and relevant sample-size context.
STEP 3 - Licklider's Solution
Input
- Figure type and display configuration, including error bar selection and significance annotation style.
- Statistical result metadata such as test name, correction choice, effect size, confidence interval, and sidedness.
- N disclosure and export-time reporting requirements for the current figure.
- Relevant preprocessing or caveat disclosures that should remain visible in draft output.
Output
- Reviewable legend draft. Licklider assembles editable draft text covering the figure, display choices, test information, and disclosure-sensitive context.
- Formatting guidance. Legend wording can adapt to the selected journal profile without pretending the text is final.
- Explicit unresolved markers. Missing information stays visible as draft labels rather than being silently dropped.
Guard
Guard condition - Error bar type not selected: If a figure shows error bars but the error bar type has not been declared, export cannot proceed until the researcher resolves that choice.
Guard condition - Legend completeness check: If required elements such as n, effect size, or disclosure text are still unresolved for claim-bearing output, the draft is marked incomplete and export remains gated.
STEP 4 - Draft Output (Draft / Needs review)
All generated text is marked as Draft and requires researcher review before submission because statistical language carries responsibility.
Example draft: Two-group comparison
[Draft / Needs review] Figure 2B. Tumor volume at day 21. Individual data
points are shown for control and treatment groups. Error bars represent the
selected variability metric for this figure. Review the final test name,
per-group n, effect size, confidence interval, and any preprocessing
disclosure before manuscript use.Example draft: One-way comparison
[Draft / Needs review] Figure 3A. Cell viability across four treatment
conditions. Statistical comparisons were generated from the current export
state and should be reviewed for the final correction method, significance
annotation wording, and reporting style required by the target journal.