OCR Employment Records for a Discrimination Case File

To OCR employment records from an employer's production, split the scan into its document types first: personnel forms, reviews and discipline, attendance sheets, the agency charge file and printed messages. Convert each type on its own, keep every production number attached to its page, and read names, dates and ratings against the image before you rely on them.

The examples use an invented case. Renata Okafor was a shift lead at Calder Freight Systems until she was let go four months after reporting her supervisor. Calder's lawyers have produced 1,140 pages stamped CFS000001 onward, all of it scanned.

The personnel file is easier to quote once the pages are searchable. Look for the personnel policy named in the termination paperwork, any discrimination or harassment complaint, and the dates connected to retaliation or family and medical leave. Those pages are exhibits in a wrongful termination case or in the agency charge. Confirm the dates and the protected activity description on the image before you quote them.

What does an employer hand over when a personnel file is requested?

An employer hands over whatever it keeps under the employee's name, and that differs from one employer to the next. Expect hiring papers, pay and job change forms, reviews, discipline, attendance records, policy acknowledgments and separation papers. Whether a pre-suit request reaches all of that depends on your state's personnel file statute, if it has one.

"Personnel file" takes its meaning from your state's statute, if there is one, and from the employer's own filing. A federal recordkeeping regulation describes the kinds of record an employer makes. 29 CFR 1602.14 requires an employer to preserve "any personnel or employment record made or kept", and gives examples: "requests for reasonable accommodation, application forms submitted by applicants and other records having to do with hiring, promotion, demotion, transfer, lay-off or termination, rates of pay or other terms of compensation, and selection for training or apprenticeship". Once a charge is filed, the same section requires the employer to preserve "all personnel records relevant to the charge or action until final disposition". That list is a useful map of what may exist, and of what to ask for when something is missing.

An inventory of the Calder production, with what each group is for and how it converts:

Three kinds of record in a production like this belong to other posts. The agreements Ms. Okafor signed, such as an arbitration agreement or a confidentiality agreement, are handled in the post on OCR for employment contracts. Punch records and payroll registers go to a spreadsheet by the method in the post on timesheets to Excel. If the employer produced phone bills or call logs for a company phone, those are tables too, and the post on phone records covers converting a printed call log to spreadsheet rows.

Splitting before converting

Cut the production into the groups in your inventory, working on a copy and leaving the produced file as it arrived. Name each part with its production number range, for example CFS000039-000060 pay change forms.pdf. A group of one document type converts more predictably than a mixed file, and a problem in one group does not hold up the rest. The post on OCR for Bates-numbered documents explains how to keep the stamped numbers usable through conversion.

Performance reviews, write-ups and attendance records as scans

Reviews, write-ups and attendance records convert in three different ways. A typed review comes out as readable text with its rating grid as a table. A write-up is a printed form with handwriting on it, so the form converts and the handwriting may not. An attendance record is a table of codes that belongs in a spreadsheet.

Performance reviews

A review form has two parts that matter: the scores and the comments. After conversion, look at how the rating grid came out. A grid with a row per competency and a column per score may export as a table, or as lines of text with the numbers trailing. Either is workable so long as each score still sits with its competency.

Watch for the score that was marked by hand. If the form lists "1 2 3 4 5" and the reviewer circled one, the converted text holds all five digits and no circle. The same goes for a ticked box beside "Meets expectations". Read every hand-marked rating from the image and type it into your notes. The post on OCR for scanned forms covers fields and checkboxes in more depth.

Capture four things from each review: the review period, the date signed, who wrote it, and the overall rating. Those four, in a table across years, show whether the ratings changed after the event your case turns on.

Write-ups and corrective action forms

A corrective action form is a few printed labels and a lot of handwriting: the incident description, the supervisor's signature and date, the employee's comments, sometimes "refused to sign". Recognition is weaker on handwriting than on print, and how it does on a particular supervisor's hand is something only a test on those pages will show. Plan to transcribe write-ups by eye. There are 19 pages of them in the Calder production, few enough to type in full, and they are central enough to deserve it. The post on OCR for handwritten notes explains what handwriting recognition can and cannot do.

Dates on write-ups deserve a second look. A form may carry an incident date, a date written, a date delivered and a date signed, in different hands. Record all of them in separate columns.

Attendance records

Attendance comes as a calendar grid with a code in each day, or as a list of dated occurrences with points. Export it to Excel and check three things: that the code key printed on the form made it into your workbook, that the number of dated entries matches a hand count from one page, and that any printed point total equals the sum of the rows.

Agency charge files and employer position statements

The agency file is the record of what each side said before the lawsuit, and it is mostly typed, the kind of page recognition handles best. The charge, the employer's position statement and its exhibits are the parts to make searchable first. Expect exhibits that duplicate pages elsewhere in the production, sometimes in a slightly different version.

For a federal charge, the Equal Employment Opportunity Commission explains the position statement in its guidance on effective position statements. The page tells employers to give "specific, factual responses to every allegation of the charge", to describe the applicable practices, policies or procedures, and to submit supporting documents. It also says the agency "may also release your position statement and non-confidential attachments to Charging Parties and their representatives". State and local agencies run their own processes, so check the one that handled your charge.

Getting the federal file itself has conditions. The agency's questions and answers on FOIA requests for charge files state that "Only the Charging Party and Respondent may obtain copies of the charge file, and only if certain requirements are met", and set out the timing for each. The agency's internal Section 83 rules on disclosure of information in charge files list material that staff remove before release. Neither page says what file format a released charge file arrives in. Test whatever you receive: search for a word you can see on the page, and run OCR on a copy if nothing is found.

Why the position statement gets converted first

The position statement is the employer's first written account of why it acted. Later accounts, in interrogatory answers and depositions, get compared with it. With the statement as searchable text you can:

• search it for each reason given for the termination and copy the exact sentence with its page number;
• search the rest of the production for the same phrases, to see which documents the statement drew on;
• list every person the statement names, as a starting list of witnesses and comparators;
• list every exhibit it cites and find where the same page appears in the litigation production.

In the Calder file, the position statement says Ms. Okafor was dismissed for "repeated failure to complete end-of-shift reconciliations". A search for "reconciliation" across the converted production finds the phrase in two write-ups, both dated after her complaint, and in none of her four prior reviews. That is a pattern to show the lawyer, and it surfaced because the text was searchable.

Can you search printed employee emails and chat exports?

Yes. Printed emails are clean, computer-generated type, so converted text of them is searchable for names, dates and phrases. What you lose is everything the printout never had: the metadata, the attachments as files, and any reliable way to sort by sender or date without reading the headers.

What happens to headers

Each printed email starts with a block of From, To, Cc, Sent and Subject lines. OCR reads those as ordinary lines of text. They are not fields. A search for "From: D. Pruitt" will find messages he sent, provided the header prints his name the same way every time, which mail programs do not guarantee. Search for the surname, then for the email address, then for any nickname the display name uses.

What happens to threads

A printed reply carries the earlier messages quoted beneath it. Print ten replies in a thread and the first message appears ten times. Three consequences:

• Hit counts overstate. Forty hits for a phrase may be four messages, each quoted ten times. Count distinct sent dates, not hits.
• The top message is the new one. On each printout, the header at the top of the first page belongs to the message that printout was made from. Headers further down belong to quoted history.
• Quoted text can be edited. A sender can change quoted text before replying. Where wording matters, find the printout of the original message and cite that.

Chat exports

A chat export prints as a long run of short lines: a name, a timestamp, a message. Check the first page or the footer for the time zone and the date format the export used. Reactions and emoji may print as small pictures or as bracketed words, and edited or deleted messages may print as a system line. Converted text keeps the words and drops the pictures, so note any page where a reaction or image carries meaning.

When to ask for something better

If the messages matter, searchable printouts are a stopgap. Whether you are entitled to electronic copies with their metadata is a question for the discovery rules, your requests and any agreed protocol, covered in the post on making a document production searchable. Convert the printouts now so that work can start, and keep the request alive.

Building a comparator table from converted employment records

A comparator table lines up the plaintiff against other employees on the same facts: job, supervisor, conduct, discipline and outcome, each with a page reference. Converted records make the table quicker to fill, because you can search every file for the same policy name, form title or infraction code.

"Comparator" is commonly used to mean an employee outside the plaintiff's protected group who was in a similar position and was treated differently. How similar the person must be is a legal standard that varies by court, so confirm it against the law that governs your case. The preservation regulation uses related language when it describes the records an employer must keep once a charge is filed: those "relating to the aggrieved person and to all other employees holding positions similar to that held or sought by the aggrieved person".

A layout to start from:

Filling it in

1. List the conduct the employer relies on, in the words of the position statement.
2. Search each comparator's converted records for those words and for the policy number.
3. Open every hit on the page image and read the whole document, not the matching line.
4. Enter one row per incident, not one row per employee, so that repeat conduct shows.
5. Fill the source column as you go. A row without a page number gets deleted.
6. Mark any cell you typed from handwriting, so the reviewer knows to recheck it.

The table is only as complete as the production. If a comparator's file has reviews and no discipline forms, the table should say "none produced" and not "none".

Which pages of a personnel file carry medical or identity data?

Medical data sits in leave requests, accommodation paperwork, fitness-for-duty and drug test results, injury reports and benefit forms. Identity data sits in the application, tax withholding and direct deposit forms, employment eligibility papers and benefit enrollments. After OCR all of it is searchable text, so the converted files need the same care as the paper.

Federal disability regulations expect an employer to keep employee medical information apart from the general file. 29 CFR 1630.14, at (c)(1), says such information "shall be collected and maintained on separate forms and in separate medical files and be treated as a confidential medical record", with listed exceptions. A production may therefore deliver medical pages as a separate set, or mixed in if the employer's filing was loose. Check both.

A page-by-page guide to what to look for:

For court filings in federal civil cases, Federal Rule of Civil Procedure 5.2, at (a), limits what a filing may show: "the last four digits of the social-security number and taxpayer-identification number", "the year of the individual's birth", "the minor's initials" and "the last four digits of the financial-account number", unless the court orders otherwise. State courts have their own rules, and a protective order in your case may add more.

Using the text layer to find them

Searchable text makes these pages easier to locate. Search the converted production for "SSN", "Social Security", "date of birth", "DOB", "routing", "diagnosis" and "restriction", and log each page found. The search is a first pass. A number written by hand, or printed without a label, will not be found by any of those terms, so page through the form-heavy groups by eye as well. The post on OCR for documents with Social Security numbers covers the working copies.

Before any of this goes to a converter, read what the service says it does with uploads. Schema Law's commitments are listed on its home page under security, and its pricing page shows the page limits per document on each plan.