Sources · Video 05

Why you can't find a job: 2,118 openings, 70 a day

Every claim in the video, with its method and what is wrong with it. Measured 7 September 2026 and re-measured every day since. Cost: nothing.

1. What we set out to test

The advice everyone gives is apply to more jobs. It assumes a pile of openings exists and that you are not reaching into it often enough. We counted the pile — and then counted how much of it appeared recently.

What we cannot see, and never claim to: whether a job was filled, how many people applied, or whether anyone read an application. No public source carries that. Everything below describes what a platform serves to a candidate, and nothing else.

2. Method

Five occupations across three cities — fifteen independent cases. Not New York, not San Francisco, whose markets would carry a result on their own.

For each case we read the counter three times: the full stock, the stock posted in the last 30 days, and the stock posted in the last 24 hours. The share older than 30 days and the « days of stock » are ours, derived from those three numbers.

We checked the source before trusting it. Our previous video used a board that advertised 90,057 listings and served only 2,000. On the search we tested here, Adzuna served everything it claimed: 2,118 listings across 43 pages, the 44th empty. Pages 1, 10, 20, 30, 42, 43, 44, 50 and 100 were each requested individually. That is one query, not a guarantee about every query.

3. The counts

job-market-flow-daily.csv — one row per case per day, updated daily. Columns: date, occupation, city, stock, posted in the last 30 days, posted in the last 24 hours, share older than 30 days, days of stock.

Median share older than 30 days: about half. Fourteen of fifteen cases fall between 36% and 51%. We give a range rather than a single figure on purpose — see « the number moves » below.

4. The case that disagrees, kept in

Warehouse associate in Chicago: 240 listings, and 172 of them appeared in the last seven days. Only 20% are older than a month. That market refills almost entirely every week.

It is one case out of fifteen, and it is in the video for the same reason it is here: not every field behaves the same way, and a result that fits everywhere without exception should make you suspicious rather than confident.

5. The number moves — measured, not assumed

Two collections on the same day, two hours apart:

Nothing collapses, but a figure that moves in two hours cannot be quoted as a constant. That is why the video says « about half » and not « 46 percent ».

6. Days of stock — our weakest number, and we say so

Stock divided by one day of new listings. Median across the fifteen cases: about a month; fastest 12 days, slowest 60.

It assumes new listings arrive at a steady rate, and they do not. That is the same flaw as our own « views per month = views ÷ age », which we criticise in our other videos, so we will not pretend it is different here. The count was taken on a Monday. A Monday is not a Saturday. We now run the same count every day, weekends included, and every day lands in the CSV above.

7. Repeated adverts — an upper bound, not a measurement

Of 1,500 sampled listings, 1,132 descriptions were distinct. So up to one listing in four is the same text posted again.

Up to, because we compare only the first 400 characters of a description that the source itself truncates at 500. Two adverts that differ after that point look identical to us. The real figure is somewhere below 25%, and we cannot narrow it with this source.

The same sample carried 777 distinct employers, the largest holding 3.9%. That matters: in our previous video one employer held 66.5% of the sample, which forced us to weight every figure by employer. Here it changes almost nothing — which makes the numbers above stronger, and is worth saying plainly.

8. What we tried to measure and could not

We wanted to tell you how many listings are actually open to you — how many demand experience you do not have. We could not.

This source cuts every description at 500 characters — twenty of twenty we checked, all cut at exactly 500 — and the experience requirement almost always sits further down. We could read it on 19 listings out of 1,500.

One percent is a finding about the truncation, not about the job market. It is not published as a number anywhere, and it is described in the video as a failure rather than quietly dropped.

9. What this does not say

• Not that the jobs are fake. An old listing can be perfectly real: some roles are hard to fill, and some employers keep a vacancy posted deliberately. We measured age, never intent.
• Not that the job market is bad. We measured what one aggregator serves to a candidate. Sliding from that to the state of employment is exactly the move we criticise.
• Not a national picture. Five occupations, three cities, one country, one aggregator. Fifteen cases is enough to stop us publishing an accident. It is not enough to describe a labour market.

10. Run it yourself

You need a free Adzuna API key. The script is recherche/etape4_debit.py in our repository.

python recherche/etape4_debit.py collecter
python recherche/etape4_debit.py analyser
python recherche/etape4_debit.py serie

90 API calls for a full collection, 45 for the light daily one. Your numbers will not match ours — they will carry the day you run them, and that is the point. A dated disagreement is information; a frozen copy is not. If your figures contradict ours, we would like to see them.