Every listing is pulled directly from the company's own recruiting system through the public endpoints those systems publish for syndication — Greenhouse, Lever, Ashby, SmartRecruiters, Recruitee, Workday, and the federal USAJOBS API. We do not scrape job aggregators, and every apply link sends you to the company's own posting rather than a copy of it.
This is a reading of the paper trail, and it is built only from public and government records: WARN Act layoff notices, Department of Labor and OSHA enforcement actions, H-1B labor condition applications that disclose the salaries a company reported paying, SEC filings for public companies, and the hiring and repost patterns we observe in our own ingestion. Every point of the score traces to a specific record, and every record carries a link to its source. Where we have too little data, we show no score at all rather than a made-up one.
This is a synthesis of publicly indexed news coverage and public discussion, and it is analysis rather than fact. Every sentence in it carries a source link — if a claim cannot be sourced, it does not get written. We do not use Glassdoor, Indeed, Blind, LinkedIn or ClearanceJobs: their terms prohibit it and their review text belongs to them.
Nobody can honestly compute your chance of getting an interview. The data that would be required — every other applicant's resume, the hiring manager's intent, the internal candidate already lined up — is not available to anyone outside the company, and any product showing you “87% match” is generating a number to make you feel something. We show one of four bands and then show you the specific signals behind it, so you can disagree with our reasoning rather than trust our arithmetic.
The grade is not an AI's opinion. It is arithmetic — a weighted scorecard of measurable signals, computed the same way every time, so the same resume always earns the same grade and fixing a real weakness can only raise it. No language model assigns the number; a model only writes the plain-English notes and suggested rewrites around what the scorecard found.
Every dimension traces to a published standard, and the breakdown on your resume links each one. The rubric and its thresholds come from university career centers (Harvard, MIT), the Ladders recruiter eye-tracking study, and ATS-vendor criteria (Jobscan). Whether your resume matches a target role is measured against real occupational data — the in-demand tools and skills that O*NET (U.S. Department of Labor) lists for that job, used under its CC-BY 4.0 license.
Two honest limits: we don't have hiring-outcome data, so the grade measures resume quality against best practice, not your literal odds of being hired — anyone claiming their AI was “trained on what gets hired” almost certainly wasn't. And modern applicant systems parse better than the horror stories suggest, so the ATS rules here maximize worst-case robustness and speed the human scan rather than promising a machine won't reject you.
It lives in a private storage bucket that only your account can read, and the parsed text is stored against your account alone. It is never used to train anything, never shown to an employer, and never sold. Deleting your account deletes the files and the parsed text. We score employers, not applicants — we will never generate an assessment of you for an employer to read.
Records get misattributed, companies share names, and public filings contain errors. If something on a company card is wrong, the source link is right there — check it, and tell us. We would rather correct a card than defend it.