eDiscovery production checklist: 24 items before you hit send

A production fails for boring reasons. The load file will not parse, the Bates numbers skip, an attachment arrived without its parent, a redaction was drawn rather than burned in, or the privilege log does not reconcile with what was withheld. None of these is a judgment call, and all of them are cheaper to find before the volume leaves than after opposing counsel does.

This is the pre-send checklist, then a shorter one for collection. It assumes the mechanics were agreed in an ESI protocol. If they were not, start there, because half of these checks have no defined right answer without one.

What do you check before a production goes out?

Specification and scope (1–5)

1. Re-read the protocol against this volume. Format, field list, delimiters, endorsement text and delivery method all come from the ESI protocol, not from the platform’s defaults. Confirm the export profile matches it.
2. Confirm what defines the set. Which review decisions, which date range, which custodians. Write it down, so the volume can be reproduced later if challenged.
3. Confirm the Bates prefix and starting number against the last number used in the previous volume. Numbers are never reused and never restarted mid-matter.
4. Confirm confidentiality designations and that the legend text matches the protective order, not an approximation of it.
5. Resolve the exception report. Every password-protected, corrupt, unsupported or zero-byte file from processing is a document nobody has looked at. Work it down, and be able to say what remains and why.

Documents and families (6–10)

6. Family completeness. No parent produced without its attachments, no attachment without its parent. Check the family relationships by comparing BEGATTACH and ENDATTACH ranges against the documents actually in the volume.
7. Placeholders where natives replace images. Every natively produced file has a slip sheet bearing its Bates number and any confidentiality legend.
8. Withheld documents are actually absent. Documents pulled for privilege are not in the volume, and their Bates numbers are not either. A gap in the range is a signal, which is why most protocols do not number withheld documents at all.
9. Redactions are burned in, not drawn. Open a redacted page in a viewer and try to select the text underneath. If it selects, the redaction is a drawing object over live text and the production leaks exactly what it was meant to hide.
10. Redacted documents are handled completely. Original extracted text withheld, OCR regenerated from the redacted image, native withheld, and the REDACTED field set in the load file.

Images, text and natives (11–15)

11. Page counts reconcile. Image count, Bates range span and OPT row count are the same number. A mismatch means a document was dropped or numbered twice.
12. Every document has text. Electronic documents carry extracted text; scanned and redacted ones carry OCR text. An empty text file is a document that is invisible to the other side’s search.
13. Endorsements are on every page and are not printed over content. Bates number and legend both, on every image, including natives’ placeholder pages.
14. Every NATIVELINK resolves. For each populated native path in the DAT file, the file exists at that path, opens, and is the right document.
15. File names match Bates numbers. Images, text files and natives are named for the Bates number they belong to, in the folder structure the protocol specifies.

Load files (16–20)

16. DAT field order matches the protocol exhibit, field for field, in order, with the agreed field names rather than the platform’s internal ones.
17. Delimiters and encoding are correct. The Concordance defaults are þ (ASCII 254) as text qualifier and ¶ (ASCII 020) as field separator. Confirm the encoding, and confirm no field value contains a stray delimiter or an unescaped line break. That single defect shifts every following field on the row.
18. The OPT file is well formed. One row per page, correct volume name, correct relative paths, and a Y flag on the first page of each document so boundaries are right.
19. Test-load the volume into a clean workspace. Not the workspace it was built in. This is the check that catches what the others miss, because a load file only reveals its problems when a different system reads it.
20. Spot-check loaded documents against the source. Pick a handful across custodians and file types and confirm the images, text, natives and metadata all describe the same document.

Before it leaves (21–24)

21. Cross-check the privilege log. Every document withheld or redacted appears on the log; nothing on the log was produced in full; the Bates references on the log point at documents that exist. Rule 26(b)(5)(A) is the standard the log has to meet.
22. Write the volume manifest. Bates range, document count, page count, native count, and a hash value of the deliverable so the receiving party can verify what arrived.
23. Confirm delivery and encryption, and that the transmittal letter states the Bates range, the counts, the designations applied, and any deviation from the protocol.
24. Retain exactly what was sent, with a chain of custody entry recording the export, the hash and the date. A production you cannot reproduce is one you cannot defend.

What do you check before a collection is processed?

Ten checks before data is handed to processing. Defects here are more expensive than production defects, because they propagate through everything downstream.

1. The litigation hold is issued and acknowledged by every custodian in scope, with the acknowledgements retained.
2. Custodian interviews are documented: what systems they use, what devices, what personal accounts touch the matter, what they have already deleted.
3. The source list is complete: mailboxes, file shares, laptops, phones, chat platforms, cloud storage, databases, third-party systems.
4. Auto-deletion is suspended where a source has retention rules that would destroy data in scope.
5. The collection method preserves metadata. Drag-and-drop copying rewrites system dates. Use a defensible tool or a forensic examiner where the collection may be challenged.
6. Hash values are computed at acquisition and recorded, not computed later from whatever arrived.
7. The chain of custody is open and contemporaneous, with a row for every transfer, not reconstructed afterwards.
8. Scope is recorded per custodian: date ranges, folders included and excluded, and what was deliberately not collected, with the reason.
9. Volumes are verified against expectation. A mailbox that comes back at a tenth of its expected size is a failed collection, not a small custodian.
10. Nothing is culled before it is preserved. De-NISTing, deduplication and culling belong after collection, on a preserved copy, never as a way of collecting less.

Why do these particular checks matter?

The list is specific because generic lists do not catch anything. Each item above corresponds to a failure that recurs: a shifted load-file row, a Bates gap across a rolling production, an orphaned attachment, selectable text under a black box, an unreconciled privilege log, an exception report nobody worked.

Two of them earn their place several times over. Test-loading into a clean workspace (item 19) exercises the load file the way the receiving party will, and it is the only check that finds delimiter, encoding and path errors reliably. Trying to select text under a redaction (item 9) takes seconds and is the difference between a redaction and a decoration.

The format decisions behind several of these are made much earlier, in the protocol. The format risk planner shows which file types lose content when imaged, and the protocol is where format, fields and load-file structure get fixed, before there is anything to check.

How this works on a flat rate

Nothing on this checklist changes with the pricing model: a production is defective or it is not. What changes is the cost of the fix. Where productions and exports are billed per gigabyte, a re-production issued to correct a load-file defect is charged again, which is a quiet argument for shipping a flawed volume and hoping. Kavarn, the platform Konvergence Inc operates, includes productions and exports in the instance rate, so re-running a corrected volume costs the time it takes and nothing else. The pricing page publishes the rate in full. See what the platform includes, or book a walkthrough.

Sources

• Fed. R. Civ. P. 34(b)(2)(A), (B) and (E) on the form and timing of production.
• Fed. R. Civ. P. 26(b)(5)(A) on describing withheld privileged material.
• Fed. R. Evid. 502(d) on non-waiver of inadvertently produced privileged material.
• EDRM, Production Guide and load-file guidance.
• The Sedona Conference, Commentary on Achieving Quality in the E-Discovery Process; Glossary: eDiscovery & Digital Information Management.
• Concordance delimiter conventions (þ ASCII 254, ¶ ASCII 020) as used by the common DAT/OPT exchange format.