Recompute the pooled estimate from the recorded study-level data and analysis settings, then compare it with the published forest plot and log any divergence. The audit succeeds when the recomputed value matches the reported estimate under the same effect measure and model.
- Assemble the inputs: the study-level effect estimates and their variances or standard errors, the chosen effect measure (for example a ratio or mean difference), and the model (fixed or random effects). Cochrane guidance treats these choices as the core analysis settings, so record each one alongside the dataset.
- Rerun the analysis from those inputs and settings, and record the pooled estimate, its confidence interval, and the heterogeneity statistics. Save the code or software settings so the run can be repeated.
- Compare the recomputed pooled estimate and interval with the published forest plot. A match confirms the record reproduces the analysis; a divergence points to a difference in inputs, effect measure, model, or weighting, which you then trace to its source.
- Write a reconciliation note for each outcome stating the recomputed value, the published value, and the reason for any difference. Cochrane interpretation guidance supports reporting exact P values and confidence intervals alongside the pooled result.
- Finalize the synthesis record with the dataset, the code or settings, and the reconciliation notes, so each outcome carries a traceable path from study data to pooled estimate.
GRADE rates certainty at the outcome level, so keep the reconciliation note attached to each outcome rather than to the review as a whole.