Accuracy and limits
Rung writes a resume from what you tell it, using occupation data published by the US Department of Labor. This page says what Rung will and will not say about you, where it can be wrong, and what has been measured.
What Rung will not say about you
- It does not invent experience. A job's real duties are offered to you as lines you can keep or leave, and nothing reaches the resume as your work unless you keep it. If you are aiming at work you have not done yet, that work is never placed in your history.
- It does not decide your occupation quietly. When a job title clearly matches one occupation, Rung names it and leaves it open to change. When it could be several, Rung asks. When nothing fits, Rung writes from your own words and grounds on nothing.
- It does not score you. There is no rating, no percentage, and nothing that judges how employable you are.
Where the words come from
Duties, skills, and job titles come from O*NET, the occupational database published by the US Department of Labor, covering about 1,016 occupations. Pay ranges come from the Bureau of Labor Statistics, schools and programs from the National Center for Education Statistics, and places from the Census. None of that is invented. Credits lists every source and its date.
The sentences on the resume are written by a language model from your own answers and from that grounded data, then shown back to you as plain lines you can change before anything is kept.
Where Rung can be wrong
- Speech. Speech recognition is a weaker technology for some people than for others. In the leading published study it made about twice as many errors for Black speakers as for white speakers, and more again with stronger accents and dialects. The limit sits in the technology, not in the person speaking, so Rung never leans on it alone: anything you say is shown back as text you can fix, and voice is never the only way to answer.
- Wording. A model can word something in a way you would not, or make a line sound larger than the work was. Every line is shown back for you to change or remove, and a line that does not read like real resume content is flagged with the reason.
- The occupation match. A typed title can sit close to more than one occupation. Where the match is not clear, Rung asks instead of choosing.
What has been measured
The occupation match is measured against a labelled benchmark of 154 job titles, weighted toward the hard ones: modern tech titles that O*NET files under older occupations, military codes, titles too broad to place, and entries that are not occupations at all. It runs again on every change to the data or the matcher.
| What was measured | Result |
|---|---|
| Cases where Rung silently used the wrong occupation | 0 of 154 |
| When Rung asks, the right occupation is among the choices | 96% |
| Entries that are not real occupations, correctly grounded on nothing | 89% |
What is not measured yet
Two gaps, named rather than implied. Rung has not yet been tested at scale with the people it is built for, so the evidence behind the design is published research rather than Rung's own field results. And speech accuracy has not been measured on Rung specifically across accents and dialects, which is one reason every spoken answer is shown back as text before it is kept.
If Rung gets something wrong
Tell us and it gets fixed. Email hello@rung.guru and say what you were doing and what Rung produced. A person will reply.