Public weights. Verified evidence.
The overall score is a weighted combination of five sub-scores with fixed, public weights. When the posting states no experience requirement, that weight moves to requirement coverage. A failed screening gate caps the total, because a candidate filtered out by a screening question is unlikely to be shortlisted however well the rest matches. The score is shown with a verdict word: Strong fit from 70, Possible fit from 45, Long shot below that, and Blocked whenever a critical gate fails, whatever the number. The word leads because our benchmark measures which band a CV lands in, not single points.
One failed high-risk gate caps the score at 55, a critical one at 40, and two or more at 35. Each gate is shown as its own finding with the posting's exact wording, so you can see what to fix or whether to apply at all.
Every requirement is a quote. Every quote is verified.
A language model reads the posting in any language and lists its requirements—skills, tools, degree, years, languages, location—marking each as core, required or nice-to-have. Each requirement must come with a quote we find in the posting; anything we cannot find is discarded. A classifier model (TypeSafe) then judges, requirement by requirement, whether your CV shows it, including the years you have in the job's own field. Where the requirement's wording appears in your CV, we quote that line; where a match is judged from meaning alone, the report says no exact wording was found. Your name, email, phone number, links and street address are removed before your CV text leaves our servers, and the posting—not your CV—is all the language model sees.
Most automatic rejections are knockout questions.
Employers filter on location, work authorisation, degree, years of experience, languages and licences before anyone reads a CV. Parviso compares each stated gate with what your CV shows—dated work experience is counted from your employment history only, at month level—and flags the ones you would likely fail, quoting the posting.
Unrecognized does not mean absent.
Section headings are recognised in English, French, Spanish, Portuguese, German, Italian, Dutch and Polish, including decorative drop-cap headings. Two-column layouts are read one column at a time so sidebars do not scramble your experience—while the layout itself is still flagged, because many application systems do not read it that way. Accents and non-Latin scripts are treated as normal text.
Findings are independent checks.
Each finding has a severity, the evidence behind it and a suggested fix: unmet requirements (most important first), screening gates, template placeholders left in the CV, missing contact details, unexplained gaps, unquantified achievements, weak phrasing, and personal details such as age or marital status—flagged more strongly when the job is in a country where they are usually left off.
Tested on real CVs before launch.
Six volunteers shared their CVs (with consent, names and contact details removed) and we paired them with 17 real job posts in English, French and Spanish: 29 pairs in all. Before running anything, each pair was labelled by hand as a strong, partial, weak or mismatched fit, and its screening questions were listed. Then both engines scored every pair.
| Measure | Before (dictionary) | Now |
|---|---|---|
| Verdict matches the hand label | 8 of 29 | 24 of 29 |
| Screening questions caught | 0 of 30 | 25 of 30 |
| Score gap, strong vs. mismatched CVs | 9 points | 58 points |
| Pairs where no requirement was read | 19 of 29 | 0 of 29 |
| Problems flagged that weren’t there | — | 0 |
Run on 25 September 2026. It is a small sample: six people’s CVs cannot cover every field or country, and the labels were set by the people who built Parviso. Read it as a sanity check, not a guarantee. We will publish new results as the test set grows.
Checked against a real ATSWe ran the same CV files through Zoho Recruit, an applicant tracking system employers use. It reads CVs with the DaXtra parser and ranks candidates with its own AI matching.
- PDF with no real spaces in its text
- Zoho RecruitNo jobs and no programming skills extractedParvisoFlagged as unreadable, with the fix
- Same content, exported with real spaces
- Zoho RecruitEvery job, title and degree readParvisoRead in full, no warning
- Junior developer vs. a real post asking for 5+ years
- Zoho Recruit38% match, no reason givenParvisoBlocked, and names the 5+ years
- Skills behind the match
- Zoho RecruitAdded skills the CV never mentions (C++, AWS, Java)ParvisoEvery match quotes the CV
Run on 29 September 2026 on a trial account. It found a gap in Parviso too: until then it rated the PDF without spaces as fully readable. That check was added the same day, and the same content now scores 9 in the broken file and 41 in the fixed one against the same job. One ATS is not the whole market; we will repeat this with others.
Why it checks what it checks