The production checklist, in the order you work

1. Before anything is numbered

Every item here is a decision, not a check, and each one is cheaper to make now than to discover later. The protocol facts are worth writing down in the transmittal even when nobody asked, because they are what the other side will measure the set against.

• Confirm the protocol: prefix, digit width, page size, confidentiality legend, filename pattern, whether natives are allowed and whether slip sheets are required.
• Confirm the range: what was served last time, and what number this production starts at. Re-using a range is the one error that cannot be fixed after service.
• Decide the file order before the stamper sees the folder. A tool numbers in the order it is given, and the file dialog's order is alphabetical, not the order of the documents.
• Remove the working copies. A redacted export and its unredacted original in the same folder is a production that undoes its own redaction.

2. While numbering

The stamp placement check exists because the setting is applied to a whole batch and a batch is never uniform. One landscape exhibit with a dark scan edge is enough to put a stamp where nobody will see it until the set is printed.

• Number from the order you decided, not from the order the folder lists.
• Use one prefix and one digit width for the whole production. ABC-9 and ABC-000010 in the same set is a set that answers to two citation formats.
• Check the stamp position on the first file before running the batch: over body text and inside the trim margin are the two placements that fail on paper.
• If a page has to be withdrawn, decide now whether the number is withdrawn with it or the set is re-stamped from that point.

3. After numbering, before the load file

This is the stage where the manual method quietly stops working: reading the stamp off 4,000 pages is the step everyone skips, and it is the only step that finds a single missing page in a volume that otherwise looks continuous.

1. Read the first and last number of every file

   The span a file claims is the arithmetic every other check depends on: last minus first plus one must equal the page count. A mismatch is a dropped page or a duplicate inside the file.

2. Check the joins between files

   Sort by first number and confirm each file begins one number after the previous one ended, with the same prefix and the same padding.

3. Look for holes and count them

   A hole inside a volume is a dropped page. A hole between volumes is a missing volume or an unused range. They are different problems and they get different answers in the transmittal.

4. The load file, generated from the folder you are actually serving

A production with several documents and no load file at all is a warning rather than a defect - some protocols allow it - but if the protocol expects one, a set without one is a production the other side has to rebuild by hand.

• Generate the DAT and OPT from the final folder, after the last document was added - not from the working folder an hour earlier.
• Resolve every row against the images: a row with no file imports nowhere, and a file with no row is invisible to review.
• Check case: a path that differs only by case works on the machine that made it and fails on the Linux server that imports it.
• Confirm document breaks in the OPT against the page counts of the PDFs they claim, not against the page counts you expected.

5. Redactions, in the file rather than on the screen

The check can prove that no extractable text remains under a redaction, and it can prove that a marked redaction was never applied. It cannot tell you whether the right words were removed, which is why this is the one stage where the tool is a second pair of eyes rather than the decision.

• Apply the redactions, don't just draw the boxes. In most editors these are two separate operations and the second one removes the characters.
• Search the exported file for the redacted term. If it is returned, the redaction is a picture.
• Flatten or re-render the page so no text layer survives underneath.
• Check that the visible remainder does not identify the person the redaction was meant to protect - no tool can decide that for you.
• Confirm no unredacted version of the same document is in the set.

6. Before it goes out

• Run the checks against the set you are about to send, not the one you assembled an hour ago.
• Read the blockers, not the count. A set with two blockers and forty warnings has two problems.
• If a gap is intentional, declare it at intake so the report says 'explained' rather than leaving the other side to ask.
• Keep the certificate with the matter file. It names the checks that ran, the checks that did not, and the hash of the set it was issued against.

What this checklist cannot do

It cannot tell you whether the right documents were collected, whether a document is privileged, or whether OCR recognised the right words. Those are judgements about the matter, and every one of them belongs to a person. What the list does is make the mechanical part of the production checkable, so the judgement is spent where it is needed.