guide

The 10-step production, and the 4 steps we check

Producing a document set runs from collection to service in about ten steps. Four of them decide whether the set survives review: the order documents are assembled in, the Bates numbering, whether every page has a text layer, and whether the load file agrees with the images. This page walks the ten steps, marks the four we check, and is explicit about the six we do not.

The ten steps

  1. Collect

    Gather from custodians and systems. We do not collect.

  2. De-duplicate

    Remove exact duplicates. We report duplicates; we do not remove them.

  3. Filter and cull

    Date, custodian, keyword. We do not cull.

  4. Convert to images

    Native files printed to TIFF or PDF. We do not convert at launch.

  5. OCR

    Add a text layer to scans. We report which pages lack one; we do not OCR.

  6. Assemble in order

    Order the documents as they will be served. We check it.

  7. Bates number

    Stamp a unique identifier on every page. We check it; we do not stamp.

  8. Build the load file

    DAT/OPT or equivalent, one row per document. We reconcile it.

  9. Index and slip sheets

    Human-readable index, placeholders for natives. We check families and gaps.

  10. QC and serve

    Verify the set before it goes out. This is the step we exist for.

The four steps we check, and the one we add

Order, Bates numbering, text layer and assembly completeness are the four that decide whether the set loads and reads correctly. We add a fifth: reconciliation between the load file and the images. A set can pass the first four and still fail on the fifth, because the DAT and the folder were built at different times by different tools.

  • Order: is the numbering monotonic across the set as assembled?
  • Bates: gaps, duplicates, prefix and padding changes.
  • Text layer: which pages are images with no searchable text.
  • Completeness: page-range arithmetic against the stated range.
  • Reconciliation: DAT rows without images, images without rows, break flags that disagree with page counts.

What we do not do, stated plainly

At launch this is a read-only check. We do not collect, cull, convert, OCR, stamp, renumber, generate an index or generate slip sheets - those are the fixes, and they ship with Prepare. A page that offered them today would be selling work the code cannot do, which is why the claim test fails the build if one does.

Frequently asked questions

Can I run the check before the load file exists?

Yes. Upload the PDFs alone and the set is checked for order, Bates, text layer and page ranges. Load-file reconciliation runs when a DAT or OPT is present.

How long does a run take?

A 200-page single file is checked in under a second. A set is dominated by page-count and text extraction; a 10,000-page set is minutes, not hours.