How the Ghost Job Index is measured

Where the listings come from

Myro reads job listings from employer hiring systems directly rather than from job boards. Each listing keeps the address it was read from, so it can be re-checked at that same source later.

The two signals

The hiring system. A verifier requests the listing at its own address. A listing is closed only on an explicit answer: the page is gone, or it says in words that the role is no longer open. A request that is blocked, rate-limited, redirected or timed out produces no verdict. An employer's system refusing our check is not evidence a role ended.

The careers feed. Separately, the listing is looked for in the feed the employer publishes. That gives two facts: when the feed last carried the listing, and whether it has since stopped carrying it.

What counts as still advertised

A listing is counted as still advertised when its hiring system has conclusively stopped serving the role, and the employer's feed has never since been seen without it. The clock runs from the close to our most recent sighting in the feed, which makes the reported duration a floor rather than a total: the ad may still be up now.

Only listings watched on both sides can be counted either way. That population is much smaller than the number of closed roles, and the index prints it beside every rate.

What is excluded, and why

Roughly half of all closure evidence ever recorded is thrown away before the index is computed. Between July and August 2026 our verifier followed a class of listing address that was missing part of its path. Such an address cannot reach any listing, so the "not found" it returns is a fact about the address, not about the role. Those records were reverted where they had affected what users see, and they are excluded here by a rule about the evidence itself rather than by date, so the same defect cannot re-enter the index if it recurs in another form.

When a figure is withheld

An employer or sector is published only once at least 20 of its closed roles have been watched on both sides. Below that, a share is noise wearing a percentage sign, and printing it beside a company name would be an accusation the evidence cannot carry. Counts are still published in that case; only the rate is withheld.

What the index cannot see

• Intent. The index reports that an ad is still up after a role closed. It does not and cannot say why. An oversight, a feed that syncs slowly, and a deliberately kept ad look identical from outside.
• The employer's own timeline. We know when we first saw a role close, not when the employer closed it. Every duration is measured from our observation.
• Roles that reopened. A role that closed and was later reposted at the same address may read as a long-standing ad.
• Employers we do not track, and every listing whose checks never reached a verdict. Both counts are published on the index.

Corrections

If a figure is wrong we will check it against the stored evidence and publish the correction. Write to hello@himyro.com.

Version history

• v2, current. Counts a listing as still advertised only when the feed has never since been seen without it.
• v1, never published. Counted any feed sighting dated at or after the close, which our own re-check schedule made true for almost every listing. It measured our timing rather than the employer's, and was replaced before the index was released.