What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Counting roughly 600,000 live job adverts each night revealed less about how many people are hiring than about how easily job-ad data can mislead. In a September 30, 2026 account, engineer Praveen Kumar described finding stale and reposted listings, duplicate records, unused database indexes and pay information his system had failed to read. His central lesson: a surprising statistic may point to a flaw in the reader, not a surprising labour market.
What the nightly count includes—and what it cannot prove
Kumar’s crawler collects adverts directly from employer applicant-tracking systems, including Greenhouse, Workday, Lever, Ashby and Oracle HCM, among others. The reported collection spans the UK, US, Canada and Australia. The September 30 article described a corpus of about 600,000 live adverts, recounted nightly.
As an Amazon Associate I earn from qualifying purchases.
That is a changing collection, not a census of every job listing, a count of unique vacancies, or a measure of hires. The total depends on the snapshot date, which adverts are considered live, employer coverage and how duplicates are handled. TUNAI’s own published totals illustrate the point:
Free tools Windows power users keep installed
One-click scans. No signup required.
| Source and date | Reported corpus | How to read it |
|---|---|---|
| Praveen Kumar’s article, September 30, 2026 | About 600,000 live adverts | An approximate description of the collection at that time. |
| TUNAI dataset snapshot, September 29, 2026 | About 590,000 live adverts; the card also reports a 601,851-advert snapshot for some observations | Use the relevant measure and denominator stated with each statistic; the card’s figures are not one interchangeable total. |
| TUNAI insights page, updated October 5, 2026 | 666,000 live adverts | A later refresh, not a correction to or substitute for the September figures. |
These are TUNAI’s reported observations. The available account and dataset description do not constitute an independent audit of crawler coverage or every underlying advert. For readers comparing job-ad datasets, the useful questions are what geography and employers are covered, when the snapshot was taken, what counts as live, how duplicates and reposts are treated, and how each statistic’s denominator is defined.
#1 Best Overall
Old and reposted adverts are clues, not proof of a ghost job
In the September account, Kumar reported 34,000 live adverts dated more than a year earlier—6% of adverts that supplied a date. He also reported 66,000 adverts older than six months. The dataset snapshot gives the latter as 66,000, or 11% of dated adverts, across 588,202 adverts. The oldest date observed was 2010.
A date supplied by an employer is not proof that a role is still actively being recruited for. Kumar notes that some old listings may be evergreen adverts kept open to collect CVs. They merit scrutiny, but age alone does not establish that a vacancy is fake.
Rank #2
The same caution applies to repeat postings. Kumar counted 3,100 adverts reposted at least five times; one had been reposted 31 times. Repetition can indicate a hard-to-fill role or a listing kept alive. It does not, by itself, establish fraud or show that no genuine vacancy exists.
Posting patterns depend on dates and denominators
Kumar’s September article reported that 20% of UK and US adverts went up on Friday and 4% at weekends. The dataset card describes posting-day observations based on employer-stated dates over the previous 90 days. The card also reports that 31% of UK adverts carrying a posting time appeared between 2pm and 5pm; date-only midnight stamps are excluded from that time-of-day figure.
A later TUNAI insights refresh, updated October 5, reported that 4% of UK adverts and 5% of US adverts appeared on Saturday or Sunday. Those later figures use a different refresh and should not be combined with the September article’s weekend figure.
The counts also show how employers frame adverts. Kumar counted exclamation marks in 2,500 US job titles and 79 UK titles. In the September 29 dataset snapshot of 601,851 adverts, three live titles used “rockstar,” five used “ninja,” and 140 used “champion.” These are counts in that collection, not estimates for every employer or job listing.
Three pipeline failures changed what the numbers meant
Oracle HCM exposed duplicate adverts
Kumar found that Oracle HCM could expose the same requisition through multiple career sites. In 61,111 eligible Oracle adverts, he identified 43,704 adverts in 15,035 groups of byte-identical records. These are findings from his pipeline, not a general benchmark of Oracle HCM or every employer using it.
His fix was intentionally strict: connect records only when source, title, place and text all match exactly, retain both records in the underlying corpus, and publish only one public page. Broader similarity matching had previously merged hundreds of genuine vacancies. The trade-off is clear: a narrow rule may leave some near-duplicates, but it reduces the risk of erasing distinct jobs that happen to use similar wording.
A logically equivalent filter left indexes unused
The database had partial indexes declared with WHERE live, while some queries used live IS true. In Kumar’s system, PostgreSQL did not use those indexes for the relevant queries, and a match took 50 to 130 seconds.
He recommends checking the idx_scan value for partial indexes to see whether they are being used. The outcome depends on the actual database version, query and plan; this is an account of one system, not a universal rule that those predicates always produce the same result.
Pay text after character 600 escaped the reader
An earlier TUNAI insight claimed that 94% of US adverts gave no pay. Kumar questioned the result and manually checked a sample of 200 US adverts classified as having no pay. In that sample, 43 adverts stated pay in text already stored by the system; 42 placed it after character 600, beyond the point where the reader stopped.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minuteHe withdrew the 94% claim pending certification of a fix. The September 29 dataset card likewise excludes facts based on advert pay text until the US pay-reading fix is certified. The episode shows why “does the advert state pay?” is partly a data-extraction question: a field can look empty because the parser missed text, not because the employer omitted it.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How to read job-ad statistics responsibly
- Keep the date with the count. A nightly corpus changes; a later total is a new snapshot, not automatically a revised historical figure.
- Check the denominator. A percentage of adverts with a stated date or posting time does not describe adverts missing that information.
- Separate records from vacancies. Reposts and duplicate records can inflate raw advert counts, while aggressive deduplication can erase real roles.
- Treat age and repetition as signals. They may warrant checking the employer’s listing, but do not alone prove a vacancy is false or inactive.
- Validate what the parser can see. Sample records classified as missing pay or other important fields, and inspect whether the information exists outside the portion being read.
- Match claims to the source’s reach. An employer-site collection across four countries can reveal patterns in its own corpus; it cannot establish a universal labour-market rate without broader, independently verified coverage.
Kumar’s practical formulation is: “when your number disagrees with everyone else’s, check your reader before you publish the surprise.” It is a useful discipline for anyone building dashboards from messy real-world text: verify collection, parsing and denominator before turning an unexpected result into a claim.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
