Manage batches
MLabel distributes work through files. Use a batch register alongside the app to track ownership and returns; there is no central assignment queue.
For the actual buttons and screens, see split and join files.
The workflow
Section titled “The workflow”Source file → split into parts → assign parts → label and export → join finished outputs. Handle remaining records separately, then assign them for another pass if needed.
Splitting
Section titled “Splitting”Use Prepare mode with the tested project configuration. Choose a batch size that a labeler can finish and return regularly. Record the input row count and keep the original dataset. Parts contain consecutive records, with sizes differing by at most one.
How many parts
Section titled “How many parts”Start with one part per person for a small project. More, smaller parts make it easier to reassign work and see progress. Keep each part’s assignment unique unless the project deliberately calls for repeated judgments of the same records.
While labeling is happening
Section titled “While labeling is happening”Track at least:
| Batch | Assigned to | Config version | Input rows | Output rows | Remaining rows | Status |
|---|---|---|---|---|---|---|
| part1-of-3 | Labeler A | review-v1 | 100 | 80 | 20 | Returned |
| part2-of-3 | Labeler B | review-v1 | 100 | — | — | In progress |
Ask labelers to return both output and remaining files. Do not treat a remaining file as missing work: it is the source data for the next pass. It does not preserve partial answers.
Joining
Section titled “Joining”Group outputs by configuration version and join them together in Prepare mode. Review validation results and duplicate warnings against the batch register. Join remaining files in a separate operation.
Use a new destination filename for each accepted delivery, and retain the individual submissions until the combined result has been checked.
Reassigning leftovers
Section titled “Reassigning leftovers”Join remaining files, then split the result into new assignments. Keep a record connecting the new batches to the earlier return. Send the matching configuration again.
What round-trips exactly
Section titled “What round-trips exactly”Remaining files preserve source values, column order, and CSV dialect. Incidental file formatting can change. See export behavior.
Accept a batch only once
Section titled “Accept a batch only once”Mark a delivery as accepted in the register before adding it to the combined dataset. If someone sends a correction, replace the earlier delivery in your collection rather than joining both.
Check that output rows plus remaining rows account for each assignment. Use copied record IDs and configured session fields to identify the source record and labeler in the combined output. MLabel only exports these identifiers when the administrator includes them.