Skip to main content

Methodology

How we research employment law, ingest court rulings, and apply source and publication controls.

Content updated: August 2026Learn about our content maintenance methodology

How we research

Our employment-law content is sourced from primary statutes published in the federal and state codes, supplemented by regulatory guidance, agency procedural rules, and case law. We reference EEOC.gov extensively for federal anti-discrimination procedure, and Cornell Legal Information Institute (LII) for statutory text, regulatory cross-references, and definitional context.

For state and local jurisdictions, we work directly from each state’s code (e.g. Florida Statutes Chapter 760, California FEHA, New York State Human Rights Law) and the relevant state agency’s published procedures. We do not paraphrase from secondary commentary in place of citing the underlying statute.

Our rulings database

Workers’ Rights maintains a research corpus of 144,000+ federal and state court records collected for employment-law research. This is a count of collected records, not completed analyses or published employment rulings. Not everything we collect is published: rulings our pipeline classifies as non-employment matters (an employment-law query inevitably sweeps in adjacent civil litigation), rulings with low-confidence extractions, and rulings without usable opinion text are filtered out of public search and browse results, so the published set is substantially smaller than the full corpus. The corpus is ingested daily from CourtListener, a project of the Free Law Project — a 501(c)(3) nonprofit dedicated to keeping court records accessible. We retain the original opinion text, court identifiers, and date of decision; nothing is paraphrased into the ruling record itself.

For each ruling we display, we also derive structured fields — outcome (employee win, employer win, mixed, or procedural), dominant claim type (Title VII, ADA, FMLA, FLSA, etc.), and any cited statutes — using a large-language-model extraction pipeline with confidence thresholds. Extractions below the confidence threshold display as “Unclassified” rather than guessing.

LLM-extracted fields are not authoritative. They’re a navigation layer over the underlying opinions, which remain the source of truth. Anyone who spots a misclassified outcome can submit a fix on /corrections; corrections enter our correction queue and override the extraction once verified.

Update cadence

Different layers of the platform refresh on different schedules based on how often the underlying source material changes:

  • Court rulings (CourtListener ingest): daily. New federal and state opinions appear within ~24 hours of publication.
  • Federal legal-source monitoring: daily. EEOC guidance, federal regulatory updates, and major federal employment-law developments are checked every weekday morning.
  • State legal-source monitoring: weekly. State agency announcements, statute amendments, and state-court precedent are checked once per week.
  • Local (city/county) legal-source monitoring: monthly. Local ordinances and municipal commission rules are checked on the first of each month.
  • Statute and rights pages: monitored on the schedules above. Detected statutory or regulatory changes must pass the publication controls below before they can replace a published value.

A page’s “Content updated” label is separate from the automated source-check timestamps shown on /law-updates.

Publication controls for law updates

Legal-source monitoring and candidate-update analysis are automated. A detected change does not become a published law update merely because a source page changed or a model produced a plausible summary.

For a non-sensitive field to update automatically, the candidate must clear all of these gates:

  • The monitored source is categorized as statutory or regulatory, not news or case commentary.
  • The change is classified as enacted and already in effect; proposals and future-effective changes are excluded.
  • The update identifies a concrete law record and a statutory or regulatory citation.
  • Automated confidence is at least 75%, and the merged record passes required-field and data-shape validation.

Filing deadlines, coverage thresholds, and protected-class lists are treated as sensitive fields. They never publish from the first automated pass. A second automated verification pass must retrieve controlling primary-source text and quote the span that supports the exact proposed value; even a confirmed result remains unpublished until an explicit approval action. If any gate fails, the existing published value stays in place. See our editorial standards and corrections process.

Technical detail (for the curious)

Most users don’t need to read this section. The encryption is automatic and works without any action on your part.

How your data is encrypted

Every account has its own per-user data encryption key (DEK). The DEK is what actually encrypts your timeline events, journal entries, evidence metadata, and other personal data on your device before any of it is uploaded.

We use a pattern called envelope encryption: instead of storing your DEK in plaintext, we store it in a wrapped form. The wrapping is done with a separate master key that lives only in our application server’s runtime environment — never in the database and never in source code. A database dump alone cannot decrypt those encrypted workspace contents. Account, access, and processing metadata are not all encrypted. Our application controls the wrapping key and can decrypt account data for authenticated recovery and agreed processing. This is server-managed encryption, not end-to-end encryption that prevents the service from reading content.

How cross-device works

When you sign in on a new device, the device authenticates with our server (Google sign-in or a magic link), then asks our server to unwrap your DEK over the same authenticated TLS connection. The DEK is sent over that connection and cached in your browser tab’s session storage, so it can survive a reload within that tab. It is not saved in persistent local storage for signed-in accounts. Ordinary vault reads and writes then happen locally in your browser. If you explicitly agree to secure cloud evidence extraction, the server temporarily unwraps the same DEK for that bounded processing job as described below.

Anonymous mode

If you use Workers’ Rights without creating an account, the encryption key never touches our servers. It is generated in your browser, stored in your browser, and used only by your browser. Your saved workspace stays on the device that created it. Choosing an AI action can separately send the submitted story or text for processing, as explained before that action.

Server evidence text extraction

For a signed-in account, OCR and document parsing use this server workflow only after the account holder agrees for the selected batch. The workflow receives opaque job identifiers, retrieves each already-encrypted original, verifies its ciphertext receipt, and uses a trusted application step to unwrap the account key and decrypt one file at a time in memory. Only that plaintext document is passed into an isolated, no-network temporary extraction workspace. The derived text returns to the trusted step and is encrypted again before it is stored, that temporary workspace is stopped after processing, and processing may continue after the browser closes. Signed-in accounts never fall back to the browser automatically: after a server extraction attempt fails permanently, the account holder can explicitly run supported extraction locally in the browser, and that local run never sends the file’s contents to our servers. DOCX and HTML backup files have a separate, explicit browser-only text reader that does not require a server attempt. DOCX reading is a text projection, with coverage limits shown for review against the original. Anonymous sessions remain local-only and run supported extraction in the browser while the Evidence page is open. Neither extraction path invokes an AI model.

AI document analysis

When you choose to use AI analysis (it’s opt-in), your narrative text and any document text you choose to analyze are sent to Anthropic over an encrypted connection. Its handling is governed by our API arrangement and Anthropic’s applicable provider terms. We never automatically send your evidence or journal entries; the AI seam is invoked only on the explicit actions you take in the app. See the Privacy Policy for current details.

Disclaimer recap

Everything on Workers’ Rights is general legal information, not legal advice. Employment law varies by jurisdiction, by employer size, by claim type, and by the specific facts of each situation. Reading an article here does not create an attorney-client relationship.

For advice about your specific situation, consult a licensed employment attorney in your jurisdiction. Our attorney directory is one starting point, but you are responsible for evaluating and selecting your own representation. Full terms in our legal disclaimer.