How a cell is computed.
version 2026-08-1
Everything below is read from the code that scores the cells. Nothing is retyped, so this page cannot drift from what is running.
The formula
Every signal gets a weight, and the score is the weighted mean of the directions. Direction is +1 when hiring gets harder and -1 when it gets easier. Magnitude runs from 0 to 1 and comes off a published ladder rather than out of a judgement.
weight = source × age × distance × saturation × sample score = Σ (weight × direction × magnitude) / Σ weight
The score is computed twice, once over the movements and once over the levels. Those answer different questions. Construction went from 7.5% to 7.4%, so the movement is nil, while 7.4% is nearly three times the European rate for the same sector. Only together are they true.
What a source weighs
Sources in use: Eurostat, CBS, UWV, DNB, ROA, CPB, SCP, ACM, Algemene Rekenkamer, EIB, Netbeheer Nederland, AZW, ABF Research, IPMERC partner data, ING, Indeed Hiring Lab.
How fast evidence ages
Weight halves after a fixed number of quarters, counted from the end of the reference period. A vacancy figure and a demographic projection do not age at the same speed, so each has its own half-life.
This is why the outlook moves when nothing is published. Evidence that fades can push a cell below the floor, at which point it goes back to reading not enough evidence.
Distance, and why sector wins
A German figure never counts toward a Dutch cell. And a borrowed figure counts toward the weight but never decides the direction while the cell has a figure of its own. Without that rule the national vacancy rate outvoted the construction rate, and the tightest market in the country came out as easing.
When a direction becomes a claim
The dead band is there so a near-zero score is never rounded into a claim. A headline also comes only from friction and demand. Knowing what labour costs says nothing about whether you can hire.
The ladders
A ladder turns a measured difference into a magnitude between 0 and 1, interpolated straight between the points below. This is the weakest link in the whole system, so it is published: anyone who disagrees can argue with a breakpoint instead of with an opinion.
vacancy rate, percentage points
0 = 0 0.25 = 0.1 0.5 = 0.25 1 = 0.4 2 = 0.7 3 = 1vacancy rate, ratio to benchmark
1 = 0 1.2 = 0.2 1.5 = 0.5 2 = 0.75 2.5 = 1share, percentage points
0 = 0 2 = 0.1 5 = 0.25 10 = 0.5 20 = 0.8 30 = 1share, ratio to benchmark
1 = 0 1.15 = 0.15 1.3 = 0.35 1.75 = 0.6 2.3 = 0.8 3 = 1index, percent change
0 = 0 5 = 0.2 10 = 0.4 20 = 0.7 30 = 1salary median, percent change
0 = 0 3 = 0.2 5 = 0.35 10 = 0.6 20 = 0.85 30 = 1required headcount, share of the relevant workforce
0 = 0 0.02 = 0.2 0.05 = 0.45 0.1 = 0.7 0.2 = 1share of the prior trend that disappears
0 = 0 0.25 = 0.3 0.5 = 0.55 0.75 = 0.8 1 = 1people required, absolute
0 = 0 5000 = 0.25 25000 = 0.5 75000 = 0.75 150000 = 1share of firms reporting, no published benchmark
0 = 0 0.1 = 0.15 0.25 = 0.35 0.5 = 0.6 0.75 = 0.85 1 = 1What this is not
Not a prediction. A cell says what the published evidence adds up to right now, and nothing about next year. Not a model that learns on its own: every number traces back through a signal to a paper, and through that paper to a primary source.
Signals are written by hand by whoever read the source, and nothing reaches a cell before a person has accepted it.