Step 1: inspect job state before interpreting metrics
Do not judge yield while a job is queued or processing. Wait for completed or downloadable status. For failed or partial jobs, inspect parsing, batch-limit and unrecognized-row errors before resubmission. On success, distinguish source-file rows from records actually admitted to screening, and remember that registered does not guarantee reachability or permission to contact.
Step 2: investigate format and duplicates when counts shrink
Empty rows, hidden characters, several numbers per row and mixed note columns often reduce import count. Use UTF-8 plain text or a single-column CSV, one normalized number per row. Remove spaces, hyphens and parentheses. Deduplication can legitimately reduce local and international variants of the same number, so retain the original file and reconcile exactly what was removed.
Step 3: separate setup errors from list characteristics
When registration yield falls sharply, verify that selected market matches every country code and that the job measures registration rather than activity or search visibility. If setup is correct, stale, disconnected or rebound numbers may explain the result. Retest a representative sample with known control numbers; stable controls and low main-list yield indicate a data issue rather than a broken process.
Step 4: reconcile export detail with summary
Compare export rows with admitted job count, confirm the newest file, and map registered, unregistered and indeterminate columns according to documentation. Retest isolated unexpected records in a small batch. A few boundary differences are normal; a broad reversal calls for another format and market review. Retain parameters, admitted count and result-file version for every batch.
If the cause remains unclear
Stop resubmitting the same uncleaned full list. Run a few dozen well-structured records, including known controls, then expand in country-specific batches. Record every change—deduplication, country code, file format and batch split—so the next result can prove which correction mattered.
