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.
Document parse and anchor map
Every line indexed before any analysis runs
Issue detection, layered
Pattern checks for release scope, ADEA/OWBPA windows, non-disparagement teeth, cooperation duties, offset triggers
Citation verification
Every claim tied back to a line in your file
Plain-English translation
Bound to the source, no rewording drift
Report assembly and final check
No claim ships unless every prior stage passed
Document parse and anchor map
Every line indexed before any analysis runs
Issue detection, layered
Pattern checks for release scope, ADEA/OWBPA windows, non-disparagement teeth, cooperation duties, offset triggers
Citation verification
Every claim tied back to a line in your file
Plain-English translation
Bound to the source, no rewording drift
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.
- 01
Flag
A reader reports an issue. Anything: a missed clause, an unclear translation, a number that does not match the source.
- 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.
- 03
Patch
We change the system, not just the one report. A new guardrail, a sharper question, a stricter check.
- 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.
- 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.