OCR for Expert Witnesses: Working From Scanned Case Files

OCR for expert witnesses is a way to turn the scanned case files counsel sends into text you can search and tables you can analyze. Before you upload anything, get counsel's consent under your engagement letter and any protective order. Afterwards, cite the scanned page itself in your report, never the converted copy.

An expert stands outside the law firm but handles its case documents, and everything the expert does with them may later be asked about. That shapes each choice in this post: what to log when files arrive, who says yes to an outside tool, how to get from a scan to figures you can defend, and what happens to the copies when the engagement is over.

When you prepare for testimony, keep the expert witness file tied to the matter the trial court assigned. Search the deposition transcript and the medical records counsel produced, then check each figure you will discuss with the judge against the page image. Your duty of confidentiality still covers those copies after you export them, so store the working files with the engagement and delete extras when counsel says the work is done.

What scanned material does counsel send an expert witness?

Counsel decides what to send an expert witness. It can include produced documents, deposition transcripts with exhibits, records obtained from third parties, pleadings and discovery responses, and the other side's expert reports. Any of it may arrive as scanned PDFs with no selectable text, and in more than one delivery. Log each delivery the day it arrives.

The log matters for a reason beyond tidiness. In federal court, a retained expert's written report must contain, among other things, "the facts or data considered by the witness in forming them", meaning the opinions, under Federal Rule of Civil Procedure 26(a)(2)(B)(ii). State courts and arbitral bodies have their own disclosure rules, and counsel will tell you which apply. Under any of them, a complete record of what you received is easier to keep from the first day than to rebuild later.

A receipt log

Fill in the last two columns yourself. Test for a text layer by trying to select a word on a few pages. For the order column, look for a confidentiality legend on the pages and ask counsel if you are unsure.

What the log tells you about OCR

Files that already have text need no conversion for searching. Files without it are the candidates, and the order column tells you which of those need a conversation with counsel first. In this fictional engagement, the transcript can be searched as it is, and both sets of confidential production pages are on hold until counsel answers.

Keep every file exactly as received, under its original name, in a folder you do not work in. Work on copies.

May an expert witness upload case documents to an outside OCR tool?

An expert may upload case documents to an outside OCR tool only if the engagement letter, any protective order in the case and counsel all allow it. Read the first two, then ask counsel in writing and wait for the answer. The decision belongs to counsel, and it can differ for each case and each category of document.

Three documents control the answer. Check them in this order.

1. Your engagement letter

Read the confidentiality terms for anything about where case materials may be stored, whether you may use subcontractors or outside services, and what happens to materials at the end. If the letter says nothing on the point, that silence is a reason to ask and not a permission.

2. The protective order, if there is one

A protective order may limit who can see designated material, and may require people who receive it to sign an acknowledgment. You may have signed one yourself. Whether an outside software service counts as a permitted recipient is a question about the wording of that order, and counsel reads the order for you. The post on OCR for documents under a protective order explains what counsel will be looking for.

3. Counsel's consent

Even where the letter and the order are silent or permissive, ask. Counsel may know of a client instruction, a regulatory constraint or a plan for the case that you cannot see.

Make the request specific enough to answer in one reply:

I would like to convert the scanned production (VAN-000001 to VAN-001602) to searchable text and spreadsheets using [tool name]. The vendor's written commitments are quoted in this email with a link. I would upload the files from my own account, export the results to my encrypted drive, and delete the uploads when my report is served. Is this acceptable under the protective order and my engagement? If not, could your office provide searchable copies?

Quote the vendor's commitments exactly. For Schema Law they are four of the lines listed under every plan: "Your documents are never used to train an AI model", "Documents stay inside your own account", "Encrypted in transit and at rest" and "Delete any document, at any time", printed on the pricing page. The security section of the homepage sets out the same four in its own, longer wording. Add nothing to either. If counsel needs to know something they do not cover, counsel can put the question to the vendor.

Possible answers and what each means

A "no" is not a verdict on the tool. It is counsel managing a risk that is counsel's to manage. Experts whose material is medical charts will find the nurse consultant's version of this question in the post on OCR for legal nurse consultants.

Turning scanned financial or technical records into data an expert can analyze

To turn scanned records into data, convert the pages, export the tables to a spreadsheet, and then verify every figure you will rely on against the page image before you calculate anything. Conversion saves the typing. It does not certify the numbers, and an expert's analysis is only as sound as the data entry beneath it.

A procedure you can describe under oath

1. Define the dataset. Decide which pages hold the data and write the page range down.
2. Convert and export. Take the tables as Excel or CSV. Schema Law's format list describes its Excel export as "Each table on its own sheet, with the rest of the text on another" and its CSV as "The document's tables, or its text line by line with page numbers".
3. Add a source column. For each row, record the Bates number or page it came from. Do this before you sort or merge anything.
4. Verify. Choose a method from the table that follows and apply it to the whole dataset, or to a sample whose size you can justify.
5. Record what you corrected. Keep a list of every cell you changed and why.
6. Freeze the verified data. Save a read-only copy. Run the analysis on a duplicate.

Ways to verify converted figures

Digits are where recognition errors do the most damage, because a wrong digit is still a plausible number. The post on why OCR gets numbers wrong lists the confusions to look for. How tables are found and exported is covered in the guide to converting a scanned PDF to Excel.

A worked example

A fictional forensic accountant, Dr. Imelda Castellanos, is retained in Vantrel Logistics v. Northgate Cold Storage to quantify spoilage losses. She receives 362 scanned pages of maintenance and temperature logs. With counsel's consent she converts them and exports the tables to Excel.

She adds a Bates column to each sheet. The logs have no printed totals, so she uses row counts and range checks: every page should have 24 hourly rows, and no cold-room reading should fall outside minus 30 to plus 20 degrees. The range check flags eleven cells. Nine are misread digits that she corrects from the image. Two are real excursions on the page, which turn out to matter to her opinion. She lists all eleven in a corrections memo, saves the verified workbook as read-only and begins her analysis on a copy.

If asked how the data got from paper to spreadsheet, she can answer in four sentences and produce the memo.

Handwritten and poor pages

Technical records include handwritten field notes, marked-up drawings and faded printouts. Recognition is weaker on handwriting and faint print than on clean print, and a converted file does not have to mark which passages were hard to read. Schema Law's homepage says of its own output that "Nothing marks a passage as uncertain". Treat such pages as material you read and key yourself, with the conversion as a starting draft at best.

Citing a scanned source page in an expert report after conversion

Cite the scanned page as produced, by its Bates number or exhibit and page number, and never the Word, Excel or text file you made from it. The scan is the evidence. The converted file is your working tool, and it has no page numbers anyone else can check.

Getting the reference right

A converted Word file reflows text, so its page 14 is not the source's page 14. A spreadsheet has no pages at all. Carry the source reference with the data from the start, which is why the source column is added before any sorting.

For each fact in your report, the citation should let a reader open one page image and see the fact:

• A document in a production: the Bates number of the page, for example "VAN-001377".
• A deposition exhibit: the exhibit number and the page within it, and the Bates number if it has one.
• A transcript: page and line as printed in the transcript, not as numbered by the PDF viewer.
• A figure derived from many pages: the Bates range, plus a schedule that ties each input row to its page.

Quoting

When you quote a document, copy the passage from the converted text if you like, then read it against the image character by character and fix what differs. Reproduce the original's spelling and punctuation, errors included. The method is set out in the post on how to quote text from a scanned PDF.

Saying what you did

Where your analysis rests on converted data, say so in the report or its appendix in plain terms. One workable form of words:

The temperature data in Schedule 3 was transcribed from VAN-001241 to VAN-001602 using optical character recognition and verified against the page images by row count and range check. Corrections are listed in Appendix C.

Ask counsel whether and where a statement like this belongs. Its job is to make sure nobody can suggest you presented software output as the record.

Before the report goes out

• Every citation points to a produced page or marked exhibit.
• No citation points to a file you created.
• Every quoted passage has been read against the image.
• Every schedule row has a source reference.
• The corrections list matches the verified data.

What should an expert keep, return or delete when an engagement ends?

Follow the engagement letter, the protective order and counsel's written instruction, in that order of specificity, and do nothing until you have the instruction. Engagements end in different ways, and a case that settles today may still have a preservation duty tomorrow. Ask counsel what to keep, what to return and what to destroy, and by when.

Take an inventory first

List everything the engagement produced, wherever it sits:

• Files received from counsel, in the original folder.
• Uploads still in any OCR tool.
• Converted files: Word, Excel, CSV, searchable PDF.
• Working spreadsheets, schedules and the corrections memo.
• Notes, drafts and the final report.
• Email and attachments.
• Backups and synced cloud folders.

Then ask

Send counsel the inventory with one question: "Please tell me which of these to return, which to destroy and which I may or must keep, and whether you need written confirmation." A protective order may set its own deadline for return or destruction of designated material after the case ends, and may ask for a certification. Counsel will tell you whether one applies and what form it takes.

Do not decide for yourself that working files or drafts can go. What an expert must preserve or may discard is a legal question that depends on the court and the stage of the case. In federal court, for instance, Rule 26(b)(4)(B) addresses protection for "drafts of any report or disclosure required under Rule 26(a)(2), regardless of the form in which the draft is recorded", which is a separate matter from whether you keep them. Leave both questions with counsel.

Carrying out the instruction

1. Return what is to be returned, by the route counsel names.
2. Delete uploads from the OCR tool and confirm the document list is empty for that engagement. Schema Law's homepage says of deletion: "The original file and everything parsed from it are removed."
3. Delete local and cloud copies that are to be destroyed.
4. Keep what you are told to keep, in one clearly named archive.
5. Write to counsel confirming what you did and the date.

Upload retention is worth settling before this point. If your tool lets you delete at will, delete uploads once the verified export is safely stored, so the end-of-engagement step has little left to do. The broader subject of retention rules and proof of deletion is in the post on OCR with data retention control.

Who pays for an expert's OCR: the expert, the firm or the client?

Who pays is set by your engagement letter. An OCR subscription you hold for all your work is your own overhead unless the letter says otherwise, a cost incurred for one case is an expense only if the letter allows expenses of that kind, and a firm that converts the files itself bears the cost directly.

Three arrangements

Raise it when the volume is known. If counsel sends 1,600 scanned pages and your plan covers 1,000 a month, say so before you convert: "This production exceeds my current allowance. I can convert it over two months at no charge, or move to a larger plan for one month at a cost of $X, which I would bill as an expense. Which do you prefer?"

Count your verification time

The subscription is one cost and your time is another. If your engagement bills the hours you spend verifying converted data, those hours belong in your estimate next to the hours that keying the same pages by hand would have taken. Put the verification time in your budget in plain view.

One point for testifying experts

In federal court, a retained expert's report must include "a statement of the compensation to be paid for the study and testimony in the case" under Rule 26(a)(2)(B)(vi). How expenses figure in that statement is counsel's call. Keep your invoices and receipts orderly enough that counsel can answer it.

Sizing your own plan

If you buy for yourself, check two limits against your receipt log: pages a month and pages per document. Schema Law's Solo plan lists 1,000 pages a month and up to 250 pages per document for $49 a month, and its Team plan lists 10,000 pages a month and up to 1,000 pages per document for $199 a month, with uploads of up to 20 MB per file. A 1,240-page production volume is over both per-document limits, so it would go up in parts, with the Bates range of each part recorded. The pricing page lists each plan in full.