Cleaning a big list, without paying twice to learn the same thing.

Run 1,000 addresses before you run 200,000. Not the first thousand rows, which are usually your oldest and worst records, and not a thousand you hand-picked. Take a random slice across the whole file.

That sample costs a rounding error and answers the only question that matters: on your data, with your domains, how many addresses does this tool come back uncertain about? Multiply and you know what the full job will really tell you before you pay for it.

Every credit spent on a duplicate is a credit spent twice for one answer, and large exported lists are often ten to fifteen percent duplicates once case and whitespace are normalised.

Lowercase the domain, trim the whitespace, drop the exact duplicates, and pull out the addresses you already know about: the ones that hard bounced last time, the ones that unsubscribed, the ones you should not be mailing regardless of what a verifier says. Verification answers whether an address exists. It has no opinion on whether you are allowed to write to it.

Four numbers describe the outcome, and people usually look at the wrong one first. The percentage of deliverable addresses feels like the score. It is not: it mostly describes how good your list was already.

Read the unknown and catch-all counts instead. Those are the addresses the job did not settle, and together they tell you what you actually bought.

• Deliverable, meaning the server said it would accept mail there.
• Undeliverable, meaning it said it would not. These come off the list.
• Risky, usually catch-all domains, role accounts or reputation signals. A decision, not a verdict.
• Unknown, meaning no clear answer was available. A good service tells you which of those reasons applied to each row.

Bulk pages advertise capacity and rarely advertise constraints. Check these before the file is the wrong side of a limit.

• Maximum file size and maximum row count, which are two separate ceilings.
• Whether your own columns survive the round trip, in order. A verifier that returns two columns and loses your CRM ids has created work rather than removed it.
• How results are delivered when the file is large: a download, a callback, or an email link that expires.
• Whether a job can be paused or cancelled once it has started, and what happens to the unspent credits.
• How long the uploaded file is kept afterwards, and whether you can delete it yourself.

Remove the undeliverable addresses. That part is simple and it is most of the value: those are the bounces you were going to take.

The risky and unknown rows need a decision rather than a rule. A pragmatic approach is to keep them out of your first send after cleaning, when your sending reputation is most exposed, and reintroduce them later in a smaller segment where a few bounces will not cost you the campaign. Role addresses are worth their own segment: they are real, they are often the right recipient in a small company, and they complain at a higher rate than personal mailboxes.

Upload a CSV, get back the same file with the verdict, score, reason codes and recommended action appended to each row, plus a PDF summary you can hand to someone who is not going to open a spreadsheet.

The engine is the same one behind the single-check dashboard and the API, so a spot check on ten addresses predicts what the job of half a million will say. Credits on signup are enough to run the sample step above before deciding anything.