The obligation is public. The headcount that triggers it is not

Headcount is not just a size attribute. It is a tripwire. Cross certain employee counts and new legal duties attach: a works council becomes possible, a disability employment levy starts, safety officers must be appointed, audit requirements change.

At that moment a company has to buy advice and cover it has never needed, and it usually does not know the duty has attached. The “why now” is not a guess. It is written in statute.

That is the strongest version of the argument, and half of it survives contact with reality.

The thresholds are real, citable, and mostly not where we assumed. The probe was briefed with our guesses, and it corrected them, which is worth reporting because the corrections are the kind of detail that decides whether a letter is credible.

Safety officers in Germany are required at 50 or more employees, not 20. Below that the duty is conditional on the workplace risk assessment rather than on a count. The occupational-safety statute we expected to carry a headcount carries none at all. The accounting size classes sit at 50 and 250, with a further class at 10.

The disability employment levy is charged monthly per unfilled place, not annually, on a scale rising to several hundred euros a month with reductions for smaller employers. We initially read it as annual. A summarising read of the statute site said annual; the state office that administers it said monthly.

Two of the underlying rules moved recently. The safety-services regulation changed its small-business boundary at the start of 2026. The Austrian accounting thresholds on the official legal portal are superseded by a later amendment. A page like this dates faster than it looks.

To use the tripwire you have to see a company cross it. You cannot, and the reason is that the law prevents it.

There is no full-text search across German annual accounts. The register that holds them states plainly that searching across the deposits of the smallest companies is not possible on account of statutory requirements. Not a missing feature. A prohibition.

Meanwhile a complete, exact, employer-level register of headcount does exist: every German employer above twenty workplaces files an annual return for the disability levy. The labour agency holds it, and publishes aggregates only, about fifteen months later.

So the state knows precisely which companies crossed the threshold, and the visibility gap sits exactly over the band where the duty attaches.

You cannot screen filings for crossings. You can generate candidates from a commercial headcount estimate and then confirm them one company at a time, by reading that company's own published accounts.

That works. It is slow, it is per-company, and it inverts the usual pipeline: the expensive step is confirmation rather than discovery. For an outreach channel that sends tens of letters a week rather than thousands of emails, that may be an acceptable trade. For anything at volume it is not.

A crossing is not a durable state. We traced one listed German company's disclosed employee numbers across six years of its own filings. It crossed from 20 to 22, then fell back to 21. Read a single year and you would call it a company above the threshold; read the series and it is a company oscillating across it. The letter you would send differs.

The unit can change under an unchanged heading. One Austrian company's filing switched from counting people to counting full-time equivalents with no change to the label above the number. Compare those two years directly and you have invented a fall in headcount that never happened.

One question stayed open, and it decides the verdict. Small German companies must prepare an employee figure, but there is a carve-out governing what they must actually file, and whether the headcount disclosure falls inside it we could not establish. The commentary that would settle it is paywalled, and the free sources contradict the statute's own wording.

If the figure may be omitted from what is filed, then confirmation over the 20-to-50 band collapses for exactly the companies this signal targets, and this drops from weak to unusable. We would rather publish the open question than a verdict resting on it.

Three things stayed unverified. Whether the register offers bulk or programmatic access at all, since every candidate address returned a 404. The Austrian register's current fees. And any measured relationship between job-posting volume and real headcount, which is the proxy everyone reaches for and nobody appears to have quantified.