Mixing these up is the most common misreading, so they are labelled separately everywhere they appear.
| Score | What it measures | Where the data comes from |
|---|---|---|
| Openness | How much an employer tells you up front, before you apply | The employer's own job listings |
| Response | Whether they replied | Outcomes logged by people using Recede |
An employer can score well on one and badly on the other. A company that states salary on every listing and then never answers anybody is exactly the case these two numbers exist to separate.
Computed from listings we have collected from public job boards and company careers pages. No employer submits anything, and nobody's opinion is involved.
| Term | Definition | Measured over |
|---|---|---|
| pay_pct | Share of listings that state a salary or a range | Every listing held: open now, or closed in the last day |
| facts_pct | Share that state something checkable: years of experience required, contract type, whether equity is included | Every listing held: open now, or closed in the last day |
| stale_pct | Share still open after more than 60 days | Listings open right now |
| median_age_days | Age of the typical listing | Listings open right now |
What an employer is willing to state is a property of everything they publish, so pay and facts are measured across every listing we hold from them: the ones open now and those that closed in the last day, after which a closed listing is deleted. A role they posted last month is still evidence of how they write job adverts.
How fresh their board is can only be read from what is live. A closed listing goes on getting older after it closes, so counting retired roles as "stale" would penalise an employer for taking a filled job down, which is the behaviour this score ought to reward. Both denominators are shown on screen next to the score.
Listings older than 180 days are not kept, even while the employer still shows them. Over the 18 days of closures we hold, boards took down 21% of their week-old listings and 2.3% of their year-old ones: an advert that survives half a year is far more often left up than waiting to be filled. So stale_pct counts listings between 60 and 180 days old, and a board full of year-old adverts reads as fresher than it is by that much.
Pay carries half of it because an unpriced listing is the single thing job seekers complain about most, and because it is the one piece of information that decides whether applying is worth anyone's time. Freshness is next, because applying to a three-year-old advert is time nobody gets back. The remaining fifth covers everything else a listing could have told you and did not.
Refusing to score is not the same as having nothing to say. Most employers in the catalogue have only one or two listings, and showing them as a blank would be silent about the majority of companies you could apply to today.
So those employers are described in counts, never rates: "their only listing states a salary", "none of their 2 listings state a salary". The sample is inside the sentence, so the figure cannot be mistaken for a percentage. No score is shown, and no percentage is even stored for them: leaving a 100 in the database would be a correct-looking number waiting to be displayed somewhere it should not be.
As of 21 October 2026, of the employers currently in the catalogue: 14,444 employers carry an openness score and a further 12,676 carry counts without one, from 658,041 observed listings between them. An employer showing neither is one we have not observed enough of to describe at all. That is an absence of evidence, not a bad result.
Those are the employers you can look up today. We hold scores for 23,670 in total, but 9,226 of them have no listing in the catalogue now, which keeps only listings from the last three weeks, and you would not find them, so counting them here would describe coverage you cannot use. The earlier version of this page quoted the larger figure.
Built only from outcomes people logged themselves: an application was sent on a date, and either a reply arrived or it did not. There is no scraping of anyone's inbox and no guessing.
Public job boards and employers' own careers pages: the same pages any person can open in a browser. We collect the listing text an employer chose to publish. We do not collect anything from behind a login, and we do not collect personal data about anyone at the employer.
Every role we show or count was posted in the last three weeks. Older ones are not shown, counted or kept, except one you saved or applied to, which stays yours.
The roles we keep are counted directly. The job engines we search while you look (Adzuna, Careerjet, EURES, Germany's Jobbörse and the others named on their listings) are counted by their own totals for those three weeks. Where an engine can only be asked for set windows, such as one week, two weeks or a month, we read three weeks between the two nearest. Careerjet dates a listing by the day it last saw it open, not the day it was posted, so its listings count as seen open in that time, and say so.
Where two engines are searched together in a country, we measured how much they share: we read both engines' full answers to narrow searches, one occupation in one city, and compared them listing by listing. That country counts as the larger engine's total times the share we measured, and never more than the two added together. Where we have not measured, only the larger engine counts. The figure on the sign-in screen is the total of all of this, rounded down.
Employers write location as free text, so the same country arrives spelled a dozen ways. We resolve it once, against published references rather than a list of our own: ISO 3166 for countries, GeoNames for cities, UN M49 for regions, and the IANA time zone database. About 93% of listings resolve to a country.
The other 7% are left blank on purpose. A listing that says only “Remote”, “Hybrid” or “Multiple locations” does not say where the work is, and guessing would put a role in front of somebody who cannot take it. A role explicitly open worldwide is recorded as exactly that, and appears under every country, because it is a job anyone can do.
Salary figures are checked against what an annual wage plausibly is in that country, using ISO 4217 currencies and World Bank income data. This is a filter for obvious errors (a listing claiming a fourteen-million-dollar salary), not a judgement about whether pay is good.
The application letter is written in the language the employer advertised in. Where a job board states the language of the posting we use that; otherwise we detect it from the text, and where neither is clear we write in English rather than guess. Letters are written in English, German, French, Spanish, Portuguese, Italian and Dutch; an advert in any other language gets an English letter.
Of the listings Recede stores (measured on 25 September 2026 from a 5% sample), 79.7% are in English, 6.6% Swedish, 3.7% German, 3.0% French, and under 1% each Portuguese, Spanish and Dutch; for 4.8% the language could not be determined. The listings searched live (Germany's Jobbörse, and EURES for the rest of Europe) are mostly in their own country's language.
What we translate is the scaffolding of the letter: the opening, the closing, the sentence that states your experience. Your own words are never translated. Your job titles, your employers and your description of what you did appear exactly as you wrote them, and where your description is plainly in a different language from the letter we leave the sentence out rather than produce something you did not write.
Employers included. Every figure here is checkable, which means it is also correctable.
If a weight, a threshold or a definition changes, this page changes with it and the date at the top moves. The scores are only worth anything while the method behind them is public and stable.