OCR Trial Exhibits: Preparing a Searchable Set

To OCR trial exhibits into a searchable set, sort the exhibit list into files that already carry text and files that are only pictures, run OCR on the picture-only ones, and export each as a searchable PDF that keeps the page image. Then test every exhibit with a real search before the set goes anywhere near a courtroom.

The aim is narrow. You are not editing exhibits or changing what the jury sees. You are giving yourself a way to find a sentence in four thousand pages while a witness is still answering. The sections below cover which exhibits need the work, how to build the set, what stickers and cover pages do to it, and how to use it under time pressure.

Why should trial exhibits be searchable?

Trial exhibits should be searchable because trial does not wait. A witness says something you did not expect, and the document that contradicts it is somewhere in the binders. With a searchable set you type three words and have the page. Without one, you are relying on memory and tabs.

Search earns its keep at several points before and during trial:

• Preparing witness outlines. Pull every exhibit that mentions a name, a product or a date, and you see the full paper trail for one witness in minutes.
• Checking your own exhibit list. In federal court, pretrial disclosures must include "an identification of each document or other exhibit", and unless the court orders otherwise they are due at least 30 days before trial, under Rule 26(a)(3) of the Federal Rules of Civil Procedure. Searching the set for duplicates and near-duplicates before that list goes out saves corrections later. Your court's pretrial order may set different dates and requirements, so read it.
• Answering objections. When the other side says a document does not say what you claim, you can go to the passage.
• Cross-examination. Impeachment depends on finding the exact prior statement while the answer is still fresh.
• Closing. Quoting an exhibit accurately means finding the line again, sometimes at eleven at night.

Searchability also changes who can help. A paralegal at counsel table with a laptop can find a passage for the examining lawyer without knowing the case as well as the lawyer does. That only works if every exhibit responds to a search, which is why a set that is 90 percent searchable is more dangerous than it sounds. People stop looking once a search returns nothing.

Try the exhibit set the way you will search it at trial. Type a date, a name, and an amount you can see on the page. For an evidentiary foundation, start from a line on that page. An authentication question sends you to a signature or a stamp. Note demonstrative evidence that contains words, and any exhibit you may raise in a motion in limine. Have the set searchable before voir dire.

Which exhibits need OCR and which already have text?

An exhibit needs OCR when its pages are images with no text behind them. That covers scanned paper, faxes, photographs of documents and most stamped copies. Exhibits saved straight from a word processor or email program already have text. Many exhibits are a mix of both, so test each file.

A fast way to triage the whole list is to open each exhibit, press Ctrl+F, and search for a word you can see on the first page and another on the last. Record the result in a column of the exhibit list.

Three cases cause most of the surprises.

Text that is present but wrong. Some exhibits were OCRed years ago by a copier, and the hidden text is poor. A search for a word on the page fails even though you can select text. Paste a line into a blank document to see what is really behind the page. If it is garbage, convert the exhibit again from the image.

Compilations. A single exhibit number can cover a typed summary followed by forty scanned backup pages. A search on page one succeeds and tells you nothing about the rest. Always test the last page too.

Handwriting. Recognition of handwriting is far less reliable than print. Run it, but do not count on a search finding a handwritten word. Note on the exhibit list that the exhibit has to be read by eye. The post on OCR for handwritten notes sets out what to expect.

Exhibits made up mostly of figures, such as ledgers and schedules, raise separate questions about rows and columns. Those are covered in the guide to OCR for exhibits with tables.

Building a searchable exhibit set

Build the set from the final exhibit list, one file per exhibit, named so the files sort in exhibit order. Convert only the exhibits that failed the search test, keep the page images exactly as they are, and finish with a page-count check and a search test on every file.

1. Freeze the list first. Converting while numbers are still changing produces files with the wrong names. If the list is not final, work from the latest version and note its date.
2. Gather one clean source file per exhibit. Use the copy that will be offered, with its sticker or stamp if that has been applied. If marking happens later, decide now whether you will convert again afterwards.
3. Name files to sort. Pad the number: PX-007, not PX-7. Add a few words of description after it, such as PX-007_Lease_2021-03-03.pdf. Sub-exhibits take a suffix: PX-007A.
4. Convert the image-only exhibits. In Schema Law, upload the PDF and wait for the pages to be read. A long exhibit may need splitting to fit a page or size limit. The guide to splitting a batch before upload explains how to cut at sensible points and put the parts back in order.
5. Export a searchable PDF for each. This is the working copy for the courtroom, because it shows the original page.
6. Export plain text as well. A folder of text files, one per exhibit, can be searched all at once with tools already on your computer, and it shows you what the software read.
7. Verify. Page count of the export equals page count of the source. Search for one word from the first page and one from the last.
8. Record it on the exhibit list. Add columns for "searchable: yes, no, partial" and "notes". Partial means a person must read it.

What the searchable PDF contains

A searchable PDF that Schema Law exports from a PDF upload is your original file with an invisible text layer added to each page, so stickers, stamps, signatures and handwriting look as they did. A search brings you to the correct page, and the highlight may not fall on the matching line. The post on searchable PDFs for court filing explains why the text layer behaves that way. Try one exhibit in the PDF reader you will use in court before you convert the rest. The export formats section of the homepage lists the other file types available.

Keep the offered exhibit separate

The searchable set is a finding aid. Unless your court's rules or the judge's instructions say otherwise, think of it as your copy, and keep the version you will offer or file in its own folder. If exhibits are to be lodged or filed electronically, format requirements apply, and they differ from court to court. The post on searchable PDFs for court filing gives verified examples.

Exhibit stickers, numbers and cover pages

Stickers and cover pages are part of the page image, so OCR may read them as text. That helps when you search for an exhibit number and hurts when a sticker covers words. Decide whether to convert before or after marking, and make the file name the reliable source of the exhibit number.

Convert before or after marking? Each order has a cost.

• Before marking: the text is cleaner, because nothing covers the page. The searchable copy then lacks the sticker, so it does not match what the witness is holding.
• After marking: the searchable copy matches the offered exhibit exactly, and the sticker text may come through as a searchable phrase. A sticker placed over a corner of text can hide words from the recognition.

For courtroom use, matching what is in the witness's hands usually matters more. Convert the marked version, and put stickers in a clear margin where you can. How stamps and stickers interfere with the words beneath them is the subject of the post on whether OCR can read stamps over text.

Do not rely on the sticker for the number. A sticker is small, sometimes handwritten, sometimes sideways. It may be read as "Exhibit l4" or not at all. The file name is under your control, so that is where the number lives.

Cover pages and slip sheets shift page numbers. If each exhibit starts with a sheet reading "PLAINTIFF'S EXHIBIT 7", then page 1 of the PDF is the cover and page 2 is the first page of the document. A cite to "page 5 of Exhibit 7" becomes ambiguous. Pick one convention and write it at the top of the exhibit list. Either cite by Bates number or internal page number printed on the document, which is the safer choice, or say that PDF page numbers include the cover.

Renumbering late. Exhibit numbers change when exhibits are withdrawn or combined. Renaming files is quick. Any cover page or sticker inside the image still shows the old number. Keep an "old number, new number" column on the list so that searching for either finds the right file.

Bates numbers still matter. Many exhibits are produced documents, and the Bates stamp is the link back to the production. If the stamp was recognized, it is searchable along with the rest of the page. See the guide on how to OCR Bates numbered documents for when stamps are read reliably.

Deposition exhibits attached to transcripts

Deposition exhibits are worth converting separately from their transcripts, with a cross-reference between deposition numbers and trial numbers. The same document can be Smith Deposition Exhibit 12, Jones Deposition Exhibit 4 and Trial Exhibit 31, and testimony about it refers to the deposition number.

The connection between testimony and exhibit begins at the deposition. Under Rule 30(f)(2)(A) of the Federal Rules of Civil Procedure, documents produced for inspection during a deposition "must, on a party's request, be marked for identification and attached to the deposition." What comes back from the court reporter is typically a transcript with a set of exhibit scans, and those scans are image-only more often than the transcript is.

A workable routine:

1. Split the exhibit bundle. Reporters often deliver all exhibits from one deposition as a single PDF. Break it into one file per exhibit at the sticker pages.
2. Name by witness and number. Smith_Dep_Ex_012.pdf sorts with the other Smith exhibits.
3. Convert the image-only ones and export searchable PDFs, as for trial exhibits.
4. Build the cross-reference. A three-column table like the one below is enough.
5. Check for different versions. Two depositions may have used different copies of one document: one with handwriting, one without, or one missing a page. Compare page counts and look at the first and last pages before you treat them as the same trial exhibit.

With the cross-reference in place, a search of the transcripts for "Exhibit 12" leads to the testimony, and the table leads to the trial exhibit. Making the transcripts themselves searchable, including page and line numbers and condensed formats, is covered in the guide to making deposition transcripts searchable.

Finding a passage in the courtroom

In court, search the whole set at once, confirm the hit against the page image, and only then hand the page to whoever is on their feet. Set this up and rehearse it before trial. The method has to work offline, on the laptop that will be at counsel table, in a few seconds.

Set up for searching across files. A search inside one open PDF is no use when you do not know which exhibit holds the passage. Two approaches work with ordinary software:

• Keep the plain-text exports in a single folder and use your computer's file search or a text editor's "find in files". Results list the exhibit file names, which carry the exhibit numbers.
• Use your PDF reader's search-a-folder feature, if it has one, on the folder of searchable PDFs.

Keep everything local. Courtroom internet is unreliable and sometimes not permitted. Copy the full set to the laptop, and to a second device, and test with networking switched off.

Search the way OCR fails. If a search for "indemnification" finds nothing, try a shorter piece such as "indemnif", a distinctive neighboring word, or a number from the same paragraph. Recognition errors often damage one word and leave the next intact.

Confirm on the image, every time. The text is a finding aid. Before anyone reads a passage aloud or puts it on a screen, look at the page image and check the words, the date and the page reference. A misread digit in the text layer does not matter if you read from the image, and it matters a great deal if you read from the text. The post on quoting text from a scanned PDF describes that check in detail.

Know the dead spots. Keep the exhibit list's "searchable" column in view. When a search of the set finds nothing, the next question is which exhibits were marked partial or no, because the passage may be in one of them.

Rehearse. A week before trial, have someone call out ten phrases from ten exhibits and time how long each takes to find. The misses show you which exhibits need converting again and which need a human index.

One last practical matter is cleanup. Exhibits often contain confidential material covered by a protective order. When trial is over and the exports are filed with the matter, remove what you no longer need from any conversion tool. Schema Law lets you delete a document at any time, and its other security commitments are listed on the homepage.