Compare both tables field by field, log every disagreement, and resolve each one against the cited source passage before merging into a single evidence table. The reconciliation log is what makes the merged table auditable: each resolved field names the passage that settled it.
- Align the rows first. Match records by study identifier and outcome, then confirm both reviewers extracted the same result. A row that exists in one table only is a disagreement in scope, so record it as such before comparing fields.
- Compare field by field. Walk each shared row through the same field order (population, intervention, comparator, outcome, effect measure, result, risk-of-bias judgement). Mark each field as agree or disagree, and write the two values side by side.
- Route each disagreement to a source check. Return to the full text and locate the passage that supports one value, the other, or neither. For risk-of-bias fields, RoB 2 judgements rest on signalling questions and written justifications, so compare the reviewers' justifications as well as their low, some concerns, or high risk labels.
- Write the resolution. In the log, record the field, both original values, the source passage used, the final value, and who resolved it. If the passage supports neither value, mark the field unresolved and name the next check, such as contacting the study authors or re-reading a supplement.
- Merge and re-check. Build the merged table from resolved values only, then confirm every changed field has a matching log entry and every unresolved field is flagged.
Where reconciliation pays off later. Effect measures and heterogeneity inputs feed directly into meta-analysis, so a corrected effect measure or sample size at this stage prevents a downstream recalculation. GRADE certainty ratings are made at the outcome level, so disagreements about outcome definitions or risk-of-bias judgements should be settled here before certainty is rated.
A useful boundary. A field with no supporting passage stays unresolved in the log; recording it as such keeps the merged table honest and gives the next reviewer a clear starting point.