Corrections

Click any cell to edit its value. Corrections are logged with the original value, timestamp, and user. Choose a propagation scope: this_document_only or all_similar (same field + method + source field across all documents). Corrections feed back as training signals for future runs, helping the system learn from your edits and improve accuracy over time. When you correct a value, the system records both the original AI-extracted value and your correction, creating a complete audit trail that is preserved even after subsequent jobs run.

When correcting a value, consider using all_similar propagation if the same mistake appears across multiple documents — for example, a reference table code that was consistently mapped to the wrong label. This applies your fix to every document in the run that matched the same way, saving you from correcting each cell individually. The system learns from these corrections, so the same error is less likely to recur in future jobs.

Corrections create a full audit trail: the original extracted value, the corrected value, who made the change, and when. This audit log is preserved even after subsequent jobs are run, giving you a complete history of manual interventions. When you export results with the full metadata option, correction history is included so downstream systems can distinguish between AI-extracted and human-corrected values.

For best results, correct the root cause rather than individual symptoms. If a field consistently produces wrong values, update the schema field's manual instruction or reference table rather than correcting cells one by one. If a reference table code is missing, add it to the table — future runs will pick it up automatically at Step 1 confidence (0.95). Corrections are most valuable as a feedback mechanism when they inform schema improvements.

1. Click the cell in the results grid and enter the corrected value.
2. Choose the propagation scope: this_document_only fixes the single cell, all_similar fixes every cell in the run that was resolved with the same field, method, and source.
3. For all_similar, check the preview count of affected cells before you confirm the change.
4. Confirm. The cell is marked as a manual correction and the original value, your user, and a timestamp are written to the audit log.

Corrections are most powerful when they inform structural improvements to your schema. If you find yourself correcting the same field repeatedly across multiple jobs, this is a strong signal that the underlying schema configuration needs attention. Common fixes include: adding missing synonyms to a reference table (so Step 1 lookup resolves the value automatically), refining the manual extraction instruction (so the AI extracts the correct value in the first place), or adjusting a format constraint pattern (so valid values are not incorrectly flagged). Each of these root-cause fixes eliminates the need for future corrections on that field, compounding time savings across every subsequent job.

Corrections and review decisions also differ in who may make them. In the Spec path, Field Review applies role rules on top of the queue: viewers can read but not decide, and fields tagged as complex can only be decided by a Senior Member or an Admin, while other reviewers leave a recommendation for them. A Senior Member who confirms a recommended action credits the colleague who proposed it, and the decision log records both people by name and email, so the audit trail shows who proposed and who decided.