Severance ClarityStart a review

Why this site, and not the obvious alternatives

We’re not a law firm and we’re not a chatbot. Here is what we are and why that matters for severance-agreement review.

Why a lawyer is not always the right call first

Why an employment lawyer is the wrong tool here

An employment lawyer is the right call if you have a discrimination, retaliation, or wage claim you want to preserve. Asking one to read a severance agreement is $500 to $800 an hour, runs into your seven- or twenty-one-day review window, and produces a verbal answer you can’t reference at 11pm when you’re rereading the release. Severance review is structured: release scope, carved-out claims, ADEA/OWBPA timing under 29 USC 626, non-disparagement reach, cooperation duties, post-employment restrictions, severance triggers and offsets. We give you a written breakdown, every clause anchored to the page it came from. Bring the lawyer in once you know what to push back on.

Why ChatGPT is not a substitute

Why ChatGPT is the wrong tool here

Severance agreements are written to be skim-friendly and are not. The release on page 3 likely waives claims you don’t yet know you have. The non-disparagement on page 5 may have a carve-out for SEC/EEOC reports, or may not. The ADEA review window only applies if you’re over 40, and the seven-day revocation window is non-waivable if it does. A chatbot will summarize the cash number, miss the carve-outs, and confidently misstate the ADEA timing. We’ve signed severance agreements. We’ve watched colleagues sign worse ones. The chatbot hasn’t done either.

How we build quality in

A quality system, not a prompt

Borrowed from manufacturing: a serious quality program is not “we tried hard.” It is layered checks, a measured defect rate, and a patch loop that improves with every report.

1GUARDRAIL

Document parse and anchor map

Every line indexed before any analysis runs

2FILTER

Issue detection, layered

Pattern checks for release scope, ADEA/OWBPA windows, non-disparagement teeth, cooperation duties, offset triggers

3GUARDRAIL

Citation verification

Every claim tied back to a line in your file

4FILTER

Plain-English translation

Bound to the source, no rewording drift

5GUARDRAIL

Report assembly and final check

No claim ships unless every prior stage passed

Reader flags something? We patch the stage, not the report. Every reader after benefits.

Our patch loop

Every flagged issue runs the same five steps. The loop is what raises the floor — not heroics on any single report.

  1. 01

    Flag

    A reader reports an issue. Anything: a missed clause, an unclear translation, a number that does not match the source.

  2. 02

    Diagnose

    We trace it back to the part of the system that produced it — the prompt, the schema, the merge logic, the rendering layer.

  3. 03

    Patch

    We change the system, not just the one report. A new guardrail, a sharper question, a stricter check.

  4. 04

    Verify

    We re-run the patched system against the original input and against a regression set, so the fix does not break anything else.

  5. 05

    Ship

    The patch goes live for every future reader. The bar moves up once and stays up.

OUR COMMITMENT

If you find a defect, we fix the system. Then we tell you what we changed.

Every flagged issue feeds the patch loop. Every patch raises the floor for the next reader. That is the only quality program worth running.

A service of Misphat LLC. Every flagged defect feeds the patch loop; every patch raises the floor for the next reader.
Start a review · $19.99