Almost every real-time labour market indicator you can read today is built on job postings. They are the natural raw material: public, plentiful, machine-readable, and updated hourly. If you want a number about the market before lunch, postings are how you get one.
For a freelancer they are also the wrong number, and not by a small margin. A posting records that somebody wrote an advertisement. It does not record that a budget was approved, that a decision-maker existed, that anyone was hired, or that the work paid. Between the advertisement and the money there are several places for the signal to disappear, and freelancers live in every one of them.
This is an argument for measuring the seller side — utilisation, offers received, outlook — with the public data that shows why the buyer side is not a substitute. It is also a description of what TechBaro is trying to build, including the parts it will not be able to see.
A posting is a stock; hiring is a flow
The clearest illustration of the gap is published every month by the US Bureau of Labor Statistics, and it is free. In the Job Openings and Labor Turnover Summary for June 2026, released on 4 August 2026, there were 7.4 million job openings on the last business day of the month — a rate of 4.4% — and 5.3 million hires across the whole of that month, a rate of 3.4%.
Those two numbers are not comparable in the way people instinctively compare them, and that is precisely the point. Openings are a stock: positions still open on one specific day. Hires are a flow: everything that happened over thirty days. An opening that has been live since March counts again in June. A role filled and re-opened counts twice. A role nobody ever intended to fill counts exactly as much as one with a signed offer behind it.
In the same release, quits held at 3.2 million (2.0%) and layoffs and discharges at 1.8 million (1.1%) — a market where neither side was moving much. A postings-only view of that market can look busy while very little is actually being transacted, which is the exact condition freelancers have been describing for two years and which posting counts are structurally unable to express.
Postings indices are real instruments — and they say what they measure
None of this makes postings data bad. The Indeed Job Postings Index is a genuinely careful instrument: indexed to 100 on 1 February 2020, published as a seven-day trailing average, seasonally adjusted using the Bundesbank's method for daily series with each series adjusted separately. A reading of 101 means postings are 1% above the February 2020 level. It is fast, consistent, and it states its own base.
Crucially, it also states what it counts: job postings on Indeed. That is a measurement of advertising volume on one platform, and Hiring Lab does not pretend otherwise. The failure is not in the instrument, it is in the reading — treating "postings fell 4%" as though it meant "4% fewer people got work", which is a different claim about a different quantity that nobody measured.
An index earns trust by being narrow out loud. The ones to distrust are the ones that quietly widen the claim between the methodology page and the headline.
The failure mode is intent, not volume
Even a perfect count of postings inherits a problem it cannot fix: some meaningful share of listings have no hiring decision behind them. This is measurable, and it has been measured — by asking the people who post them.
ResumeBuilder.com, in a survey published on 18 June 2024 and fielded on 22 May 2024 (1,641 hiring managers screened, 649 completing), reported that 40% of companies posted a fake job listing that year, that 3 in 10 had one active at the time of the survey, and that 7 in 10 hiring managers considered the practice morally acceptable.
Clarify Capital, surveying 1,045 managers involved in hiring on 31 August – 1 September 2022, found that 68% had a posting live for more than 30 days, 1 in 10 had one open for more than six months, 50% keep postings open because the company is "always open to new people", and 43% were not actively trying to fill the role at all.
Both are employer self-report surveys, both skew US, and neither audits actual listings — they tell you what hiring managers say they do, which is a real but bounded kind of evidence. Even bounded, it is fatal to a naive reading of posting counts. A number that mixes real openings with listings whose purpose is to look like growth is not measuring demand; it is measuring the sum of demand and impression management, with no way to separate them.
What utilisation measures that postings cannot
Utilisation is the share of your available capacity that you actually billed. It is the number consultancies run their businesses on, and for an independent it answers the question a rate never does: did the work exist, and did you get it?
It has four properties that make it the right primitive for a freelance index.
- It is realised, not advertised. You cannot bill a ghost job. Utilisation counts outcomes on the seller's side of a completed transaction, so intent is not a variable at all.
- It is comparable across rates and currencies. A percentage of capacity means the same thing to someone billing €90 in Munich and $40 in Kraków, which a monetary figure never does.
- It is the missing half of income. Rate times utilisation is what you earn. Every published source in this space reports the first term and none report the second, so every one of them is describing half of an equation and calling it a market.
- It moves before the rate does. Rates are sticky — they are set in contracts and renegotiated reluctantly. Utilisation falls the month the work stops. As an early indicator that is the whole value proposition.
Pair it with two things a posting count also cannot supply — how many concrete opportunities reached you, and what you expect over the next quarter — and you have a picture of the market from the side that gets paid.
Four things any index owes its readers
Measuring the right quantity is necessary and not sufficient. A seller-side index built carelessly is just a differently-wrong number. Four obligations, none of which is exotic:
1. The sample size, next to the figure
Not in a methodology appendix. Next to the number, in the same visual weight, every time. A reader who cannot see n cannot calibrate anything, and a figure that hides its n is asking to be quoted by people who would not quote it if they knew.
2. A defined population
Who was eligible to respond, how they were reached, and who is therefore missing. "Freelance developers" is not a population; "people who saw this site and chose to answer" is, and it comes with obvious biases that should be printed rather than implied.
3. A publication rule fixed in advance
A threshold decided before the data arrives is a constraint. A threshold decided after you have seen the numbers is a marketing decision wearing a lab coat. TechBaro's rule is 30 reports in a rolling 30-day window, per segment, and a segment cannot borrow the overall sample to qualify — which is why regional and role cuts will publish later than the headline, and why the site currently publishes nothing at all.
4. A distribution, not a midpoint
A median with no spread behind it is a Rorschach test: every reader sees their own position confirmed. Publish the shape, or publish quartiles, or publish nothing.
What TechBaro measures — and what it refuses to
TechBaro asks contributors six questions: their role, the country they work from, how much of their capacity is currently billed, how many concrete opportunities arrived recently, how the market feels now, and how they expect the next quarter to go. Signed-in contributors can report weekly, anonymous contributors fortnightly, so that no individual can move the result.
What it will not do is as important as what it will. It does not scrape job boards, because a posting count is the thing this article argues against. It does not publish below the threshold, and it does not publish a segment on the strength of the overall sample. And it does not, ever, show an illustrative index value, a sample chart with numbers in it, or a placeholder trend line — a fabricated figure would be indistinguishable from a real one to a reader, which is exactly why it is prohibited.
The home page shows the live count toward the first publication and the form that feeds it. As of writing, the index does not exist.
What this index still will not tell you
An honest argument for a method includes the method's limits, so here are the ones that will not be fixed by more responses.
- Self-selection. Contributors are people who found this site and chose to answer. That is not a random sample of freelancers and never will be. Every figure published here will carry that caveat.
- One side of the market. Utilisation tells you what sellers experienced. It says nothing about buyer budgets, why a project was cancelled, or what a client paid an agency instead.
- Segments will lag badly. A country-by-role cut needs its own 30 reports. Most cuts will not have them for a long time, and the honest response is an empty cell, not an interpolation.
- Language and reach. An English-language site collects from people who read English-language sites. That is a real distortion in a market where the interesting regional differences are exactly where that assumption breaks.
- Small samples are directionally fragile. Thirty reports is a threshold for publishing something rather than a guarantee of precision. Early readings will be noisy, and they will be published with their sample size so that you can discount them yourself.
None of that argues for going back to counting advertisements. It argues for publishing a narrow claim with its limits attached — which is the only kind of market number that has ever been worth reading.