Every subpoenaed document,
checked against the court book.
You hold a subpoena production, bank statements, loan documents, correspondence, whatever was produced, and a draft court book. Before the book is settled, someone has to say whether every produced document is in it. We identify each document on both sides by what is printed on its own face and check the two sides against each other. You get a register of what is in the book, what is not, and what cannot be read. Software does the matching. People verify it. The examiner signs. Ten working days is standard from receipt of both sets. One re-run against the final court book is included. Price any matter yourself through our pricing system. Used in Supreme Court of NSW proceedings in 2026
What you receive
Six deliverables, and every one can be re-run
Every figure comes from a saved script reading a saved register. When the court book is renumbered, the pack is rebuilt against the new version, not corrected by hand.
Your index, with presence columns
Your own court book index, returned with a column marking every subpoenaed document present or absent. The index itself is checked against the documents actually supplied.
The register of what is not in the book
One row per produced document with no counterpart: what it is, the fields that identify it, which subpoena return produced it, and the tab range it belongs in.
The production sorted and identified
Every produced document classified and identified. Bank statements by bank, BSB, account, period and page count, with the printed page count checked against the actual. Loan documents, guarantees and correspondence by party, instrument and date.
Excerpts traced to their source
Where the court orders that only the relevant pages of an email chain or a bundle go in the book, the tab is an excerpt. We trace each excerpt to the full produced document and record which pages it covers. Any difference between the excerpt and the source is reported, so an excerpt can be shown to be a faithful part of the whole.
The court book checked against itself
Every document in the book identified from its pages and compared with the index. Where the index and the page disagree, the row says so and says which side.
One date-ordered PDF per subpoena return
Ready to drop into the next version of the book. Documents that cannot be dated by rule are listed in an exceptions register, never guessed at.
A signed report
Method, standards, limitations and findings, with a page-one statement of how the work was checked and how software and AI were used. Signed by a named examiner who can give evidence.
Why we do not compare files
The obvious method fails on this material. Send us the court book index and a listing of the production. We will show you how far file comparison gets on your book, before you engage us and at no charge.
The court book is not a copy of the production
Assembly re-renders every document and adds tab stamps and page numbers. No file in the book matches any produced file byte for byte, so file hashing reports almost everything as missing.
Filenames collide
Both sets number their files from one. Matching on filename reports documents as present when they are unrelated.
We read the face of each document
A bank statement carries its bank, account number and statement period. A loan document carries its parties, instrument and date. We read those fields off each document on both sides and match on them.
We count documents, not files
Productions carry duplicate copies, loose and again inside emails. Counting files can overstate a gap by more than double. We count distinct documents, and for statements distinct periods.
Why not a paralegal
What each option checks
An experienced paralegal can do much of this by hand. The difference is time, whether every document was covered, a stated error rate, and a signature from an examiner who can give evidence. The company that assembled the book cannot check it independently, and neither can anyone inside the firm.
| The check | A paralegal at the firm | NDF |
|---|---|---|
| Finds duplicate copies of a document | Yes, by eye. May miss copies inside emails, or copies with small differences | Yes, by content, including copies inside emails |
| Matches each document on the details printed on its pages | Yes, by reading each one. Slow, with a human error margin | Yes, by script, then hand-verified |
| Matches an excerpt in the book to the full email or bundle | Yes, but labour-intensive | Yes, page by page |
| Finds gaps in a sequence, such as bank statement periods | Yes, if the sequence is built by hand | Yes |
| Covers every document and every page | Hard to be sure of on 2,000 files in a week | Yes, and the registers show it |
| States an error rate | No | Yes. Fewer than 1 in 100 rows wrong, at 95% confidence, from a random sample read by hand |
| Re-run when the book is renumbered | Starts again | Minutes |
How it is checked
Software matches. People verify. The examiner signs.
The matching is done by scripts anyone can re-run. The checking is done by people, and the signed test plan, the completed sample and the scored result are delivered with the registers.
Every finding is read blind
Each document we report as absent is put in front of a second reader without the analysis beside it, and read off the page. All of them, not a sample.
One standard, on every job
Every row a finding rests on is read in full by a second person. Everything else is sampled so we can say fewer than 1 in 100 rows is wrong, at 95% confidence, and the report states the rate we measured. The sample size and the pass rule are signed before the draw.
AI reads, scripts match, people check, the examiner signs
Specialist recognition AI, built by us and run in an isolated environment in Australia, reads each page and identifies the document and the details printed on it. Deterministic scripts then match the two sides on those details, so every match traces to what is printed on the page and anyone can re-run the registers. The examiner reviews every case the software cannot decide and signs the report. Page one of the report says what the AI did, what the scripts did and what the examiner did.
What it does not say
The report says what is in the book and what is not. It does not say why. Where a document carries a key that identifies it on both sides, absence can be shown. Where it does not, the report shows presence and says plainly that absence cannot be established for that document. An excerpt is reported as an excerpt, never as the whole document.
What we will ask you
Send the court book index and a listing of the production, and we can price the job before you commit.
Which version of the court book?
Every tab number in the pack belongs to the version we were given. When the next version lands, we re-run and re-cite. The first re-run is included.
How many documents, of what kinds, from how many subpoena returns?
These set the price. Documents are counted after email attachments and containers are opened, so the document count is usually lower than the file count. Each document type is identified by different fields, so each type is a priced line.
Are any produced files password protected?
Locked attachments can only be opened with the password the producing party sent, usually to your client by SMS. Ask for them on day one; they bound the answer until they arrive.
When does the book go to the judge?
Ten working days is standard from receipt of both sets. Shorter timeframes are quoted on inspection.
How the material is handled
Secure, reachable, and kept
All electronic documents are stored securely on servers in Australia, from receipt to delivery.
Processed on a secured, isolated system in Australia
Your production and the court book are hashed on receipt and again at the close. They are processed on an isolated system in Australia and delivered from Australian servers. Nothing is uploaded to a public service.
Delivered by secure link, on any device
The pack comes by a private link with a separate access code sent by phone or text. It opens in a browser, including on a mobile, with nothing to install. Links stay open for 14 days and are reissued on request.
Kept, so a re-run needs nothing resent
We keep the working material for at least four years under our standard terms. When the court book is renumbered or a further batch arrives, the refresh runs from what we already hold.
A recent matter
How this worked on a recent matter
More than twenty subpoena returns, a thousand-tab court book, and a week to say what was in it.
Read the case studyRequest a pricing system login
Price this matter yourself
Tell us about the matter and we will approve and send your login within one business day. The pricing system gives an indicative fixed fee from the number of documents, accounts and subpoena returns. Any figure it produces is subject to our inspection of the production.
Request received
We will approve and send your login to the address you gave within one business day. If the matter is urgent, call 1300 123 099.