Methods Text Auto-Generation
Writing the Methods section from memory is one of the easiest ways for a figure to drift away from what was actually analyzed. Licklider therefore generates Methods-facing drafts from the same recorded state used for export review.
STEP 1 - The Pitfall: The Memory Gap Between Analysis and Writing
Biomedical manuscripts are often written long after the figure is finalized. In that gap, details such as preprocessing disclosures, sidedness choices, effect-size reporting, and sample attrition can easily be remembered incorrectly or omitted entirely.
- Correction method errors. The manuscript says one correction method, but the final figure used another.
- Effect size omission. Effect sizes and confidence intervals may be required for claim-bearing output, yet they are easy to forget when writing by hand.
- Preprocessing disclosure drift. The final figure reflects imputation, winsorization, or log use, but the written Methods text fails to disclose it clearly.
- Guard decisions disappear. A warning acknowledged during export review is material to reproducibility, but it may vanish from a hand-written summary.
STEP 2 - Journal Requirement
Journal reporting standards do not simply ask for a polished paragraph. They ask for the material analytical choices to be disclosed consistently, including sample size, preprocessing, statistical test choice, and limitations that affect interpretation.
STEP 3 - Licklider's Solution
Input
- The Data Contract, including observation unit, variable mapping, outcome type, and analysis intent.
- Captured preprocessing disclosures and summaries, such as the current preprocess snapshot and the repro bundle's
preprocess_report.json. - The selected statistical test and its configuration, including sidedness, pairing, and correction choices.
- Reporting completeness state, including effect size, confidence interval, sample size, and unresolved disclosure requirements.
- Generation provenance metadata, including app version, ruleset version, stats backend version, model version, and execution time.
Output
- Structured Methods draft blocks. Licklider builds reviewable draft sections for Data, Preprocessing, Statistical test, Effect size, and Caveats.
- Companion legend and provenance drafts. The same export flow also produces a legend draft and a provenance note so figure text and version metadata stay aligned.
- Explicit unresolved markers. Missing information is surfaced directly in the draft with labels such as
[Needs review]and[Not resolved yet].
Guard
Guard condition - Required elements missing: If a claim-bearing figure is still missing required disclosures or reporting elements, the draft is marked incomplete and claim-bearing export remains blocked until those items are resolved.
Guard condition - Caveats stay visible: If Licklider detects unresolved confounding, attrition, or other important caveats, those notes remain surfaced in the generated draft instead of being silently dropped.
STEP 4 - Reviewable Draft Output
The examples below illustrate the kind of draft text Licklider can assemble from the captured export state. Researchers should review and refine the wording before manuscript use.
Example: Two-group comparison draft
[Draft / Needs review] Prior to analysis, tumor volume measurements (mm3)
were log-transformed and winsorized as disclosed in the preprocessing record.
The treatment comparison was evaluated with a two-sample Welch t-test.
Review the final effect size, confidence interval, and sample-size wording
before using this text in a manuscript.Example: One-way ANOVA draft
[Draft / Needs review] Group differences in cell viability were assessed
using one-way ANOVA with a multiple-comparison correction selected during the
analysis workflow. Confirm the final post-hoc method, reporting terms, and
sample-size language before manuscript use.