The six-week window, and where the number comes from

We say "six weeks, not six months" in almost every conversation. It is the sharpest thing we say and it is the reason municipalities take a second meeting. So it deserves a proper account of where the number comes from, and an equally proper account of what is still unproven about it.

The finding

There is an interval between the moment a change becomes visible in municipal administrative data and the moment a response stops being prevention and starts being crisis management. In the Catalan municipalities we have modelled, that interval is roughly six weeks. The national statistics a director is actually working from arrive between twelve and twenty-four months after the period they describe.

The gap between those two numbers is the entire product.

How the interval was derived

Three inputs, none of them secret.

Service response latency. From the point a municipal social services team decides to reallocate capacity, standard practice in Catalan municipalities puts a measurable change in provision somewhere between three and five weeks later: a staffing decision, a contract variation, a redirected outreach programme. That is the floor. A signal that arrives with less lead time than that cannot be acted on, only reported on.

Signal formation. A neighbourhood-level shift in service demand becomes statistically distinguishable from noise once it has persisted for two to three weekly reporting cycles, given the cell sizes typical of a barrio of eight to fifteen thousand residents. Below that, we are looking at variance.

The outer edge. Beyond roughly eight weeks, the changes we are tracking have generally already produced the secondary effects a municipality would have wanted to prevent: waiting lists, a school enrolment problem, a housing pressure that is now in the local press.

Six weeks is the usable middle. It is long enough for the signal to be real and short enough for the response to matter.

What a municipality is currently working from

The comparison is not flattering to the instrumentation, and it is not the municipality's fault.

Source	Geographic resolution	Typical lag
National demographic series	Province or municipality	12 to 18 months
Regional labour market series	Comarca or municipality	6 to 12 months
Municipal padró extract	Neighbourhood	3 to 9 months, depending on processing
Own administrative caseload	Service point	Real time, but not aggregated or compared

The last row is the interesting one. Municipalities already hold a real-time signal. It sits in the caseload of individual services, in separate systems, unaggregated, unjoined, and with no comparison against a baseline. Nobody is hiding it. There is simply no layer that reads it.

What this does not mean

It does not mean that six weeks of warning prevents every outcome. Some pressures build over years and no lead time changes them. Housing supply is the obvious one: knowing six weeks earlier that a neighbourhood is under housing stress does not produce housing.

It also does not mean that acting on a six-week signal is always right. A signal is not an instruction. The Observatory produces informational alerts and never prescriptive ones, and a signal that a service is under pressure is not evidence about why. That is a separate piece of work, and it involves talking to people rather than reading a chart.

And it does not mean the interval is the same everywhere. A municipality of 300,000 with a well-staffed innovation team and a municipality of 55,000 with one data analyst do not have the same response latency. The floor moves. Part of what the Diagnostic Sprint does is measure the actual interval for the specific institution, rather than assuming ours.

What we are doing about the uncertainty

The first municipal pilot is designed to test the number rather than to confirm it. Specifically:

1. We record, for each alert raised, the date the signal crossed threshold and the date a decision was taken. That gives a measured response latency for that institution rather than a modelled one.
2. We record the alerts that were raised and not acted on, and why. A prevention tool that is right and ignored has a different problem from one that is wrong.
3. We publish the distribution, including the cases where the window was shorter than six weeks and the response did not land in time.

The third point is the one that matters. Any vendor can publish the cases where their tool worked. The methodology is only worth something if it survives publication of the cases where it did not.

The honest summary

Six weeks is a working figure, derived from service response times and data release schedules, and it is currently modelled rather than measured. It is defensible enough to plan an engagement around and not yet strong enough to be quoted as a finding. We are telling you which of those it is, because the alternative is that you find out later.