The Hiring Signal Index,
and why there is no number yet.
The design is finished. The data is not. This page sets out the calculation, the cohort and the limits first, so that when a figure appears it can be checked against what was promised rather than explained afterwards.
The collection has completed one run, producing 12 hiring findings across 19 companies. The index measures change, and change needs a prior state to measure against. Until a second run covers the same cohort, any score would be arithmetic on one observation.
Functions that opened roles, against functions that stood still
Not a count of postings. A count of postings tracks company size and hiring season; the ratio tracks where budget moved.
- For each company, classify every open posting into a function: engineering, sales, marketing, product, finance, operations, support.
- For each function, compare the count against the same company and function in the previous run. A function is opening when the count rose, standing still when it did not.
- The market figure is the share of company-function pairs that opened, across the market’s cohort. One number per market per period, with the pair count shown beside it so a reader can see how thin or thick it is.
- A cluster rule sits on top: three or more roles in one function inside a fortnight is reported separately as a project, because that is the shape worth acting on and it is lost in a ratio.
A company with four hundred staff posts more roles than one with forty, in every period, for reasons that have nothing to do with intent. The ratio normalises that away and leaves the thing a reader wants: whether budget is moving, and into what.
What is actually being read
- Cohort: the 19 companies on the open watchlist, across four technology markets. The same cohort as the print-rate note already on the site, so the two are comparable.
- Sources: six applicant tracking systems that companies post their own adverts on, read directly rather than through a third-party index. The dates are the company’s own and the text is unedited.
- Unit: one posting, on one company board, with the date the board gave.
- Geography: wherever the posting says. The cohort is United States and European technology companies, and the index would be reported per market rather than per country.
- Cadence: the index would be computed per run, and a run covers the fortnight before it.
Exactly what has to exist before a figure can stand
In order. Nothing here is a judgement call; each one is a condition that is either met or not.
- A second run over the same cohort. The index is a change measure and there is currently one run to work from. This is the blocker, and it is the only one that cannot be worked around.
- Function classification on every posting. The collection currently keeps the findings worth printing, not a function-level census of all postings read. The index needs the census, because a ratio over the printed subset would measure what the suppression stage chose rather than what the companies did.
- Enough company-function pairs per market to report one. Three companies in a market cannot carry a market figure. The honest floor is a stated minimum below which the market is reported as too thin rather than given a number.
- A fixed cadence. Two runs a variable distance apart are not comparable periods. The index needs the interval fixed before the first figure, not after.
These do not go away when the data arrives
- Postings are not hires. A posting is evidence that a budget was approved and a role was opened. Whether anyone was appointed is not on the board.
- Six boards are not every board. A company hiring through an agency, a referral or a system outside the six is invisible, and that absence is reported rather than estimated.
- A reposted role can look like a new one. Reposts after a failed search inflate an opening count, and the cluster rule is the partial defence rather than a cure.
- Nothing about individuals. The index counts postings; it holds no candidate data and names no person.
- One index is not a labour-market statistic. It describes this cohort in this period and nothing wider.
Where several outlets report the same hiring story, that is one underlying item carried several times, not several independent confirmations. The index counts postings on company boards for exactly this reason: the board is the company speaking for itself, once.
The method, and the figure that already stands
The research method in full
The five stages, what each is allowed to conclude, and the live arithmetic behind the print rate.
How much market research is noise
A figure that already stands, because it needed one run rather than two, with the four pages a reader can add up themselves.
Hiring research runs on the Talent lens. What the platform does with it.
Run this on the companies you care about
The same six boards, read overnight, on your own watchlist rather than ours.