Preprocessing Audit Log

What the Preprocessing Audit Log records, where to find it, and how preprocessing disclosures affect claim-bearing output.

Every action applied to a dataset before analysis — removing outliers, filling missing values, reshaping the table, or aggregating replicates — is recorded in the Preprocessing Audit Log. This log exists so that the state of the data at the time of analysis is transparent and reproducible.

This page is about the preprocessing record itself. It does not introduce a statistical result such as a p-value, effect size, or model estimate. Instead, it shows what data handling stands behind those outputs so you can judge whether the analysis is appropriate for your study.

What gets recorded

The log captures every preprocessing action applied to the dataset, including:

  • Missing value handling — drop rows, fill with fixed value, fill with most common value, fill with median, fill with mean
  • Outlier handling — keep flagged outliers, cap flagged outliers (winsorize), exclude rows with flagged outliers, manual outlier override
  • Transformations — log transform, bin numeric values, recode category values
  • Structural changes — reshape wide columns to long format, aggregate replicate measurements

For each action, the log records what was done, which columns were affected, and how many rows or cells were changed.

If no preprocessing was applied, the log reflects that state as well.

Where to find it

Dataset view The full preprocessing record is visible in the Prep panel of the dataset view. This shows the complete history of applied actions including before-and-after counts, imputation details, outlier thresholds, and flagged row samples.

Figure Inspector — Audit Trail tab When a figure is selected, the Audit Trail tab in the Inspector shows the preprocessing record for the source dataset alongside the figure's outlier and attrition information.

Important: The preprocessing record shown in the Audit Trail tab reflects the current state of the dataset's preprocessing record. It is not a frozen snapshot from the time the figure was generated. If preprocessing has been updated after the figure was created, the Audit Trail tab will show the updated record.

This design keeps the Audit Trail tab focused on the dataset as it stands now, so the figure review surface and the dataset review surface do not drift apart. If you need generation-time evidence for a specific figure, use the figure-level records linked from this page, especially the Outlier Exclusion Log and Versioning and Provenance.

Preprocessing and disclosure

When preprocessing has been applied, Licklider requires that it is acknowledged before claim-bearing output can be exported. The specific requirements depend on what was done:

Outlier removal When rows have been excluded using an IQR or manual criterion, the exclusion must be acknowledged. An unresolved outlier removal disclosure blocks claim-bearing export.

Winsorization When extreme values have been capped rather than removed, the winsorization must be acknowledged. An unresolved winsorization disclosure also blocks claim-bearing export.

Other preprocessing Imputation, log transforms, reshaping, and aggregation are recorded and disclosed but do not independently block claim-bearing export.

Licklider records these actions automatically when they are applied, and the disclosure status is then used to decide whether claim-bearing export can proceed.

The stricter export rule for outlier removal and winsorization exists because these actions directly change which observations contribute to the result or cap the values that drive the estimate. They therefore require an explicit acknowledgment before claim-bearing output is exported.

Other preprocessing actions are still shown because they matter for transparency and reproducibility, but on this page they do not trigger an export block by themselves. That does not mean they are scientifically neutral. It means the log records and surfaces them, while the researcher remains responsible for deciding whether the preprocessing choice is appropriate for the study design, outcome, and analysis plan.

The disclosure status for each preprocessing type is visible in the Inspector. Resolving a disclosure involves confirming that the preprocessing is appropriate for the analysis and that it has been reported.

What this page does not guarantee

The Preprocessing Audit Log helps you see what was done. It does not prove that the preprocessing was the right choice.

In particular, Licklider does not use this log to determine whether:

  • an imputation strategy is appropriate for the missingness mechanism
  • a log transform matches the scientific meaning of the variable
  • an aggregation step collapses structure that should have been modeled explicitly
  • an outlier rule is justified for the study rather than merely documented

These are important risks because preprocessing can change effect sizes, uncertainty, group balance, and the interpretation of the final figure or result. The log reduces hidden preprocessing. It does not replace researcher judgment, protocol review, or method-specific sensitivity checks.

What this page does not cover