Missing values
Treat missing-value cleanup as a documented data decision
Blank cells are only one form of missingness. Exports often encode the same absence as N/A, NA, null, none, a dash, or a business-specific placeholder. A careful workflow first detects the pattern, then applies an explicit reviewed rule.
Normalize known tokens before measuring missingness
The supported normalize-missing-tokens operation converts an explicit list of tokens into the engine’s standard missing representation. The target columns and token list are visible in the plan, can be edited through controlled parameters, and remain part of the frozen run snapshot.
This is different from guessing a replacement value. Normalization makes missing data consistent so downstream tools can count and handle it predictably without inventing observations.
Why automatic filling remains preview-only
Mean, median, mode, constant, interpolation, or model-based imputation can materially change analysis. The choice depends on column meaning, distribution, missingness mechanism, and downstream use—not simply on which function is available.
Because the current Worker does not support those operations, the real interface does not serialize or send them. Development previews may show how safeguards and explanations should look, but an advanced option cannot enter the real Server Action or Worker payload.
- Real: normalize explicit missing tokens
- Real: drop fully empty rows or columns with high-risk review
- Preview: fill or impute missing values
- Deferred: Worker and audit contracts for advanced methods
Verify the result in the run report
Impact preview shows the likely effect of confirmed rules before execution. The completed report uses actual Worker results and warnings, so a rule that encountered failed values or unexpected input remains visible.
The cleaned file is a new artifact. If a reviewed rule removes empty rows or columns, that change applies only to the output and is represented in the report rather than overwriting the source.