The Review step is where you interpret the match. You see aggregate stats, the Genie's narrative summary, and a queue of borderline cases that need your judgment.
What you see
Stat cards
Four top-level numbers:
- Source rows processed
- Matched (count + % of source)
- Needs review (count + % of source)
- Unmatched (count + % of source)
These three percentages sum to 100% of source rows.
The Genie's Take
A paragraph-long narrative summary at the top, describing the result in human terms. Example:
"The Genie matched 3,147 of your 4,812 source rows (65%) against your CRM master. Another 842 (17%) fell into the review band — mostly close-but-not-exact name matches where the address or phone differed. The remaining 823 (17%) had no plausible match in the master. Of the matches, 2,890 were driven by the email column; the rest came through fuzzy name+address scoring."
Always read this first. It tells you the shape of the result before you dive into numbers.
Pass breakdown
A table or chart showing how many matches came from each pass:
- exact_id — identifier match (email, NPI, account number)
- deterministic — all comparable fields near-exact
- fuzzy — scored via the fuzzy pipeline
- phonetic — matched primarily on phonetic code
A healthy run is dominated by exact_id and deterministic when your data supports them. A run dominated by fuzzy is fine but suggests your data could benefit from enrichment.
Score distribution chart
A histogram of _lmg_match_score values, so you can see match quality at a glance. Good runs have a clear two-peak shape — one peak around 95+ (confident matches) and one around 0 (unmatched) — with thin middle. A smeared middle means your data or profile has ambiguity.
Review queue
A table of every row classified as review, sortable by score. Click any row to open the side-by-side comparison:
- Source row on the left, top candidate on the right
- Matching fields highlighted green
- Differing fields highlighted amber
- Missing fields shown as
— - Per-field scores displayed
- Three action buttons: Approve, Reject, Skip
Work through the queue high-score-first. Skip is useful when you want to come back to a case later; skipped cases remain in the queue.
Notes
Every review case accepts a freeform note. Notes are captured in _lmg_notes and follow the row through the export. Useful for recording why you accepted a borderline case so the downstream team has context.
Bulk actions
On large review queues, bulk actions save time:
- Approve all above score X — accept all cases at or above a threshold
- Reject all below score Y — reject all cases at or below a threshold
- Filter by field match — e.g. only show cases where the email matches but the phone doesn't
Bulk actions are logged individually so the audit trail still shows a decision per case.
Advancing
The Next button advances to Export. You don't have to clear the review queue to export — unresolved review cases simply export with _lmg_match_status = review and empty _lmg_review_decision. You can resolve them later via the job detail page.
Related reading
- Confidence scores — how classification works
- lmg columns — the fields that record your review decisions
- Jobs — the persistent job detail page