Methodology

Rebuilt: A figure for a past night that we worked out afterwards from our record of every posting's changes, because that night was not recorded as it happened. Rebuilt figures are marked rebuilt wherever they appear.

Careful figure: The share of dated postings listed 90 days or longer, with bulk-dated postings not counted as long listed. It is shown first, and the share counting every posting is shown beside it.

Bulk-dated: A posting that shares one identical posted date with 49 or more other postings from the same employer. A date shared that widely may record an upload, not the day each job was first advertised.

Completeness check: A test of each night's reading of the boards we track. A night that passes is published. A night that fails is held: the previous night's figures stay up, and the page says which night was held.

Fixed panel: The 1,032 employers the Zeltru Report holds fixed from one month to the next, so a change in a panel figure is a change at the same employers, never a change in which employers we read. On any night, the panel employers with dated postings are the ones counted, and each figure says how many of the 1,032 that is.

Every posting's age is counted from the date the job board itself reports, whenever we have stored one:

• Greenhouse, Ashby and iCIMS boards publish a publication date with each listing. Where we have it we use it directly — “Listed 200 days” means the employer's board says it was published 200 days ago. Some iCIMS boards serve a date that is not a real publication date, and we set those aside.
• Lever carries no posting date, and we no longer publish an age for its listings. The only date Lever's board offers is the date the posting record was created, which is not the same thing: we have watched a listing whose record dates from 2017 appear on a board we were reading daily, and one record created in 2016 is an internship asking for candidates who graduate in 2028. Until September 17, 2026 we counted that value as a posting date. We do not any more. Those listings stay on the site, with their links, marked as having no posting date.
• Workday boards don't put a real date in their list, so we read it from each job's own detail page. At the employers we read before September 24, 2026, that gives us a real posted date for nearly every Workday role.
• Employers we began reading on or after September 24, 2026: posting dates are not yet stored for most of their postings, on every platform. They are planned to arrive in a later change. Until then those postings say “Posting date unavailable” and are left out of the age figures, so the median time listed and the share listed 30 days or longer come mostly from the employers we read before that date. Edition No. 2 shows the share of each platform's postings carrying a board date at the end of September.

One limit on the boards that do publish a date, stated because it is ours to state: on Greenhouse the published date is the date a listing was first published, and it survives a listing being taken down and put back up. For about 5% of the Greenhouse listings we can check, the date therefore counts from an earlier publication rather than from a continuous run on the board. We have not found the same pattern on Workday, Ashby or iCIMS, where it affects under 1%.

Collection is scheduled daily, but saved active records can persist during collection gaps. An active flag or an employer posting date does not confirm a recent successful check.

“Listed N days” is a posting's age, counted from the employer's posted date on its current public listing. When no permitted employer date is available, the posting says “Posting date unavailable”; we do not substitute a private first-observed date.

“Listed N days or longer”, also written “listed N+ days”, as in “listed 90 days or longer”, describes postings whose employer-dated age is at least N days, so a posting listed exactly N days is counted. Postings without an employer date are left out of those figures.

Some older pages and saved records used “listed N+ days” for a minimum counted from first observation when no usable employer date existed. Keeping that older definition for those figures does not publish private observation dates or individual history, and it does not change historical aggregate formulas.

The site-wide age figures are stated one way everywhere on this site: of currently listed postings with employer-published dates, X% have been listed 30 days or longer — and the same form for 60 and 90. That is a composition figure. It groups saved active records by employer posting age; it does not prove they were checked today. It is not a survival rate, because we have not followed a set of postings from the day they appeared to the day they came down. “Or longer,” not “more than”: a posting listed exactly 30 days is counted.

These figures and the “postings listed 90+ days” counts use only board-dated postings — real employer dates. Undated postings (“Posting date unavailable”) are excluded from these percentages, so they can never be inflated or deflated by our own first-seen clock as roles age.

We also exclude continuous-hiring roles — per-diem/PRN and float pools, talent pools, freelance rosters, all-shifts requisitions, and postings an employer labels evergreen — because they stay listed by design rather than going unfilled.

Company medians are shown only when a company has at least five dated postings listed; below that we state the count and posting age directly.

A company's share listed 90 days or longer follows the same rule: it is shown only when at least five of its listed postings carry an employer posting date, and below that we state the count alone. On the leaderboard, whose shares are US-only, at least five of its US postings must carry one.

When we add a company to tracking, its entire current job board appears in our data at once. That's us catching up — not the employer posting a wave of new jobs. We exclude these onboarding-day imports from “roles added” and growth numbers, so a day we onboard a hundred companies reports the real organic activity, not a one-time import spike.

Two different numbers could be called “jobs tracked”, and they are far apart, so this page says which one we mean. The headline figure is active tracked roles: saved postings marked active, counted one per posting. It excludes records marked inactive; all-time observed includes historical records and is a different population. Active records can persist during failed or incomplete collection and do not prove a recent successful check.

We also only count postings we actually enumerated. Where an employer's board refuses to list everything it holds, we count what we could read and mark the figure as a floor. We never substitute the board's own estimate of its size for a count we could not take — an estimate answers how much of a board we can see, not how many jobs are open.

A failed or incomplete board read can leave saved active records in place. Those records do not tell us whether the job is still listed today, and an employer posting date is not a recent observation date. A company explicitly marked as no longer tracked is handled separately from an individual collection gap and is excluded from freshness-dependent activity and alerts.

We do not mark postings as taken down merely because a read failed. A failed read does not establish removal. Recording thousands of removals on a day nobody checked would be inventing an event. When we can read the board again, we compare it to what we hold and record what actually changed.

A few employers' job boards limit how many postings they will list at once, no matter how many they actually hold. Target's board answers 2,000 while its own category counts add up to about 12,100. We can read the 2,000 it will serve and no further, so we say “at least 2,000” rather than stating a number we know is short. Eight of the boards we track behave this way.

For these companies we also leave out anything counted from postings appearing and disappearing — new posting counts, how long roles last before coming down, and the “since you last looked” figure on companies you follow. On a capped board a posting can drop out of view simply because newer ones pushed it past the limit, and the employer never touched it. Rather than report that as a job being taken down, we leave the figure out for those employers and say so here.

Applicant reports have moved to the private interview timeline, where your record of your own process is private to your account. The reported no-response rate, recruiter score and the breakdowns built from the old report form are retired, and no page publishes them.

A listed role is every active posting on the employer's own careers page, as of our saved snapshot, including continuous-hiring postings, such as talent pools and per-diem pools, and postings an employer labels evergreen. A figure taken over a narrower set says so where it appears. The share of postings listed 90 days or longer, the oldest posting and the age distribution leave out continuous-hiring and evergreen postings, because they stay listed by design, and a company's median age counts only its dated postings, excluding likely evergreen listings.

Four company counts appear across the site and they count different things. Companies indexed is every company with a profile page. Career sites we track is the subset with at least one live posting.

On a benchmark — the median share of postings listed 90 days or longer — the companies we track with at least five dated postings listed names the sample that median is taken over: only companies whose own share can be stated, which is smaller again, because most boards publish no posting date. And companies with employer-supplied posting dates is the posting-duration dataset, which excludes boards that publish no date. A figure on this site should always say which of these it is.

A company page compares an employer with its industry only when the industry has at least 20 employers with five or more dated postings listed each. The page then sets the employer's median days listed beside the median of those employers' medians, and its share of dated postings listed 90 days or longer beside the median of their shares.

A table lists employers in the same industry of a similar size, ordered by postings listed and nothing else. Each figure is the one that employer's own page states. The comparison describes how long postings have been listed, not why.

It is the median age of postings we watched come down, counted from the employer’s own posted date to the day we recorded the posting gone. Both endpoints are real dates, so unlike a “days we tracked it” figure it is not truncated at the day we started watching.

It is not a survival statistic. A survival statistic follows a cohort from posting to removal and reports what fraction remains at each age. We cannot do that: a posting removed before our observation window opened is invisible to us however long it lived. The bias runs one way — postings with very long lives are under-represented, because we only see the ones whose removal happens to fall inside the window — so the median is a floor on typical lifespan rather than an estimate of it. We state that rather than correcting for it, because correcting would need the cohort data we have just said we lack.

A removal carries no reason. We can observe absence from the employer’s own listing after a complete board read. A detail URL may still answer after delisting; a failed or incomplete read does not establish removal. We cannot see whether the role was filled, cancelled, paused, or moved to another board, and we do not infer it.

Every page quoting this figure carries the observation window beside it, because a window is what tells you what the number excludes. Postings we can identify as continuous-hiring requisitions are excluded, and so are boards that publish no posted date — those are left out rather than given an invented one.

Since 18 August 2026 a completeness check has graded every day’s read of the boards we track. 6 days since then are not graded complete. 2 had no fleet read at all: a removal that happened on one of those days was recorded on the next successful read, a day or two later than it happened — a small bias that runs one way, and is only knowable if the days are listed. On the other 4, every or nearly every board was read but the run’s own record is incomplete, so the day is not confirmed complete; the observations from the boards that were read stand. Any figure whose window spans one of these days says so beside the window. The days:

• August 23, 2026 — no fleet read, 0 of 1,124 boards read: the collector was upgraded and split into shards to stay inside its host's run limits, and the fleet did not walk during the transition.
• August 24, 2026 — not confirmed complete, 1,120 of 1,124 boards read: the first sharded night; four boards went unread and five of six shard records reported incomplete.
• August 26, 2026 — not confirmed complete, 1,123 of 1,124 boards read: one board went unread and two of six shard records reported incomplete.
• August 28, 2026 — not confirmed complete, 1,124 of 1,124 boards read: every board was read, across three dispatches that day, but two of fourteen shard records reported incomplete.
• August 31, 2026 — no fleet read, 2 of 1,124 boards read: the shard planner counted boards where the runner counts employers; the first employer with two boards appeared in the plan twice, and every shard was refused.
• September 1, 2026 — not confirmed complete, 1,139 of 1,139 boards read: every board was read, but a host outage lost the run records of 26 boards, so the day cannot be shown complete.

Days before 18 August were observed daily but not graded — the check did not exist yet — and we do not grade them after the fact. A day with no observation is not a quiet day: recording zero removals for it would be inventing an event, the same rule we apply to a single board we cannot read.

Every ranking on this site starts from the same tracked postings and applies a short list of rules. The list is not the same on every page, so two of our “oldest” figures can differ by thousands of days while both are right. The rules, and where each applies:

• Country. The leaderboard’s lists default to US postings. The figures at the top of the leaderboard and on the homepage (postings tracked, the median time listed and the share listed 30 days or longer) count every country, and so does the report. On the homepage they sit under the headline on a computer and under the search on a phone. A posting whose board publishes no country is in the report and in no US view.
• Undated postings. A posting whose board publishes no date says “Posting date unavailable”. We do not substitute the date we first saw it, so it carries no age: the leaderboard lists it without one, and posting-year cohorts and the report’s dated figures use employer-published dates only.
• Continuous-hiring postings. Evergreen requisitions, talent pools and the like are flagged and hidden by default everywhere. The all-time leaderboard has a toggle to show them; cohort views and the report’s figures do not include them.
• Bulk-dated postings — 50 or more from one employer sharing one identical posted date — are excluded from cohort views and from the report’s “excluding bulk-dated” figures, and included, unmarked, on the all-time leaderboard.
• Implausible dates. We never drop a date from an aggregate calculation solely for being old. Public role lists use current eligible records and employer dates. Historical individual examples are withheld; aggregate populations and values remain unchanged.
• One per employer. Cohort views show each employer once, its longest posting in the cohort. The all-time view caps an employer at three postings per page unless you filter to that employer.

The report preserves its original aggregate populations. Individual historical examples previously used to illustrate these rules are withheld from public delivery.

Private posting event history begins July 27, 2026, when daily observation started. Individual removal, reappearance and observation history is operator-only; public pages do not provide a role timeline. Events before collection began were not observed, so absence of a recorded event is not evidence that nothing happened.

Dates recorded before August 18, 2026 are accurate to within about two days. Until then a repeated check on the same day could advance a posting’s clock twice, which could retire a posting from our records a day or two earlier than it actually left the employer’s board. Frozen originals and historical aggregate definitions are preserved.

Public reports retain aggregate statistics. Individual historical examples, including named and de-identified examples, are withheld. The original editions are retained privately; posting age does not create a public naming exception.

A long-listed posting isn't proof of anything on its own — roles stay listed for many legitimate reasons (backfills, high-volume hiring, slow internal processes). We report counts and durations, dated from each company's own job board. Active status does not confirm a recent successful check. Every listing links back to its source so you can verify it.