Changelog

What changed, and when. Entries describe what is different for someone using the site — most of the work behind it is invisible, so most weeks this list is short.

September 10, 2026

A wording correction on the September edition: a footnote said postings were “checked daily”. Reads are attempted daily, but some days do not complete, and those days are listed on the methodology page. It now reads “re-read on a daily schedule”. No figure changed — only the claim about how often we look. /reports/2026-09
The September edition's Section 2 is back. Twenty figures — the ten longest-listed postings we name, and the ten ranked anonymously beside them — were removed from the edition on 9 September and are restored here with their original values unchanged. The edition's figures are frozen by design; that removal should not have happened, and the record of it is kept rather than quietly corrected. /reports/2026-09
Your reviews and saved application details stay private. You can see and delete your own reviews; removed-listing history is restricted to the operator. /reviews
The job-link checker now requires sign-in and checks current public listings. If a lookup cannot load, it tells you instead of treating the link as a missing listing. /answers
You can choose edition notifications and general updates independently, and unsubscribe from either or both. /email-preferences
When a company or job-board count cannot be loaded, it is marked unavailable instead of appearing as zero. Available listings and other figures stay visible, and company pages offer a way to try again. /jobs
When job listings are unavailable, Try again now reloads the board so it can recover when the problem clears. You can also return to the full board. /jobs

September 9, 2026

Company tabs now work with the arrow, Home and End keys, and keep the selected tab visible on smaller screens. /companies

September 3, 2026

The methodology page now states which postings each ranking leaves out — country, missing dates, continuous-hiring flags, bulk-dated groups, implausible dates, one per employer — and works through why the site's three different "oldest posting" figures differ while all being right. Every posting-year cohort view says the same beside its counts, and the four cohort views are listed in the sitemap with the date they were last recomputed. /methodology#exclusions
The job posting leaderboard can now be filtered by the year a posting was published: Posted 2026, Posted 2025, Posted 2024, and 2023 and earlier, beside the unchanged all-time view. Each cohort ranks active postings by days listed within that year's employer-published dates, one posting per employer, with the same flags and the same bulk-dated exclusion as the rest of the site, and states its population and employer count on the page. Cohort views have their own web address, so they can be shared. /leaderboard?posted=2025
The homepage card for the longest-open US posting now opens the leaderboard it comes from, where that posting sits at rank 1 with its dates, its flags and a link to the live listing; it used to jump straight to the employer's board. The report index's line about the oldest posting we track now opens the report's table of longest-listed postings for the same reason. Links to a live listing stay on the rows that carry the posting's own context. /leaderboard
A correction to today's earlier entry, which said three days had no complete read. Our own completeness check, which has graded every day's read since 18 August, grades six days as not complete: two had no fleet read at all (23 and 31 August), and on four (24, 26 and 28 August, and 1 September) every or nearly every board was read but the run's own record is incomplete. Figures now say so beside their window, in those two groups, and the methodology page lists each day with its cause. The earlier entry stands below as written. /methodology#missed-days
Three days in our record have no complete read of employer boards — 23 August, 31 August and 1 September — and they are now named beside every figure whose window spans them, with the cause of each on the methodology page. A posting that came down on one of those days was recorded on the next successful read; the days are listed rather than smoothed over. /methodology#missed-days
Site-wide figures now refresh only after a complete daily read of every board we track, and the "as of" date on the homepage is the date of that read rather than the following morning. On a day the read does not complete, the figures stay at the last complete read and carry its date, instead of being recomputed from a day nobody observed. /
The September report now ends with a short list of related questions — what a posting being taken down after an interview means, whether companies post jobs they do not intend to fill, and others — each linking to its own answer. They are questions the report itself does not measure, answered separately from the same data. /reports/2026-09

September 2, 2026

We found what caused the problem behind our removal-figures warning, and it was us. Some job postings had been stored under two different internal identities; in August we merged each pair into one record, keeping the earlier date so no history would be lost. That job was deliberate and correct in itself, and nobody connected it to the fact that rewriting the history of 100,000 records changes what every figure covering those weeks means. The warning page now says so, along with the two corrections it took us to get the measurement right. /how-we-checked
We said we track live postings from 1,166 employers' career sites. The true number is 1,163. The same sentence said those postings are read daily — for three of them that stopped being true on August 21, when we stopped tracking them but kept counting them. /reports/2026-09
A correction we published on September 1 was itself wrong. We said the two active-posting totals in Edition No. 1 — 503,503 and 503,499 — were one measurement read at two moments. They are not. They are two different columns of a single row recorded at a single instant, and the gap of four is the number of always-open postings. We withdraw the earlier explanation. /reports/2026-09
Edition No. 1's headline of 503,503 active postings includes 4,253 postings from three organisations we had already stopped publishing before press. The figure stays as printed, because the edition's numbers are frozen; the base it was drawn on is stated here instead. Drawn only from organisations we published, the count is 499,250. /reports/2026-09
Around press time, roughly 10,100 postings from 28 employers we publish went unobserved for between two and twelve days. One employer's board began refusing our reads on August 21 (855 postings). A database outage on September 1 blocked a night's reads for 26 more (about 9,200 postings). One employer's 4 postings went unread for nine days for reasons we have not established. Their postings aged in our figures without being re-checked. The published median is unchanged and p75 moves by one day, but the figures are not as fresh as the page implies, and this is the mechanism. /reports/2026-09
Three figures in Edition No. 1 — the share of postings carrying an employer-supplied date, and the two counts behind it — have no recorded derivation. We cannot reconstruct which postings they were computed over, so we cannot confirm them against the frozen file. The figures are unchanged and are flagged here rather than restated. One of the three, the count of 499,250, matches to the digit the published-only base stated above. We cannot confirm that is how it was computed, so the match is noted rather than claimed. /reports/2026-09
We added a warning to our removal figures for late July and August, and corrected it twice while getting the measurement right. On 10 and 11 August the stored records for 223 employers — about a quarter of the postings we track — were replaced, carrying history from before the record existed. Those employers account for well over half the removals we recorded in that period, so the removal accounting for the window rests partly on rewritten records. We could not work out which way that runs, so rather than publish a corrected count we cannot stand behind, the page says plainly what the limitation is. We now check for the same thing nightly. /how-we-checked

September 1, 2026

We now publish how often we get removals wrong. Checking 100 postings our data had marked as removed against the job boards themselves, 7 of the 92 that gave a clear answer were still live — 7.6%. At the scale of our most recent report that is roughly 14,100 postings, and it means we slightly understate how long postings stay open before coming down. We could not find a cause, and say so on the page rather than attributing it to a bug we had already fixed. /how-we-checked
A new page shows the results of an audit we ran on our own posting dates. We took 375 postings at random, published the list before checking any of them, then went to each employer's job board and compared the date it shows against ours. 96.3% of the long-open postings we could read matched the board exactly, to the day, and 92.4% of the postings we had marked as removed were confirmed gone. The page also lists every posting that did not match, what caused it, and three things the audit showed we had gotten wrong. /how-we-checked
Suggesting a company now offers the full list of industries. The dropdown had drifted to sixteen options and was missing several we already use — nonprofits, aerospace and defense, and automotive among them — so anyone suggesting a company in one of those had to pick something that was not quite right. It now offers twenty-four, including restaurants and food service, professional and business services, staffing and facilities, education, and government. /suggest
The Zeltru Report now carries a section on the wider labor market: payroll employment and the unemployment rate, job openings and turnover, and weekly unemployment claims, each with the federal release it came from, the date that release was published, and a link to it. None of those are our figures and none of them are mixed with ours — the section exists so both can be read side by side. The report’s own numbers are unchanged. /reports/2026-09
Five corrections to The Zeltru Report, listed on the report itself rather than applied quietly — no figure changed. The citation block now gives the date of the data (August 30) instead of the date it was counted; the unnamed longest-open table says which set it is drawn from, and the age it refers to is now printed; the geography section explains why its total is four postings below the headline; and two percentages that had lost their decimal have it back. The report promises its numbers will not be rewritten, which is exactly why corrections to everything around them are published in the open. /reports/2026-09
The email signup now says what it will actually send: an email when Edition No. 2 publishes on October 1, and an unsubscribe link in every message. It had promised mail “when the monthly report launches”, which had already happened, and a one-click unsubscribe that not every way we might send can carry. /reports
The Zeltru Report is published — a monthly read on how long job postings actually stay up, measured from employers’ own careers pages rather than from a survey. Edition No. 1 covers 476,618 dated openings across 1,037 employers: the median has been listed 28 days, and about one in five for 90 days or longer. Every figure is frozen at publication, so a number quoted from it stays quotable. /reports
Three employers joined the site with their openings already listed — Panera Bread with 2,732, Signet Jewelers with 2,633 and Concentrix with 1,510. Every one of those postings carries the date the employer published, read from their own careers site rather than estimated. /companies

August 26, 2026

The navigation link that said “Reports” now says “Share experience”, which is what it actually opens — the form for telling us about an application, not a page of published reports. /report
A company’s “longest-open role” now means the same thing everywhere on the site. Talent pools and always-open listings were counted on the company page but not in the rankings, so 38 companies showed two different answers. They are left out of both now, and the page says when one was left out.
Company pages no longer describe our comparison group as “S&P 500 and Nasdaq companies”. We track far more than that now — including employers outside the US and public-sector ones — so the sentence names the number it is actually measured over.
Palantir’s headquarters read Aventura, Florida — the address on its SEC filings rather than where it operates. It now reads Denver, Colorado.
Jobs in the US capital were showing as “D.C., WA” — the city and state swapped, and the wrong state. They now read “Washington, DC”, as do the eight other spellings employers use for it.
The “report an experience” page used to show a blank dark screen until its JavaScript finished loading — on a slow connection that could be several seconds of nothing. It now draws the header, the step bar and the company search box immediately, and the form becomes usable underneath them without anything moving on the page. /report
Several pages described the site in ways the data no longer supported. The trending page's search-result summary promised rankings “as submitted by job applicants” when no reports have been submitted yet; 85 company pages said their postings were dated from the employer's own careers page when those employers publish no dates at all; and the site was still describing itself as covering S&P 500 and Nasdaq companies, which stopped being the whole roster a long time ago. All corrected. /trending
The longest-open pages said roles had been posted “more than 90 days”. The count has always included roles at exactly 90, so they now read “90 days or longer”. Same postings, more accurate sentence. /longest-open
Red Lobster's careers page reports the same posting date for every job it lists — all 3,458 of them, to the millisecond. We had been treating that as a real date and saying those roles had been open 166 days. They now read as “listed N+ days”, counted from when we first saw them, which is all we actually know. The site's count of postings listed 90 days or longer drops by about 3,500 as a result. /company/red-lobster
A new answers section takes the questions job seekers actually search for — how long until you hear back, what it means when a posting is taken down after an interview, what a ghost job is — and answers each one from the daily crawl, with the number in the first sentence and the denominator under it. Two of the pages say plainly that the question cannot be answered by anyone, including us, and show what can be measured instead. /answers
You can now paste a job posting link and find out how long it has been listed, whether it is still up, and when it came down if it is not — without an account. It sits on every answers page. /answers
The site can now say how long postings last, not just how long the ones still up have been up. Half the postings we watched come down had been listed 22 days or less, and several thousand came back after disappearing. /answers/job-posting-taken-down-after-interview
The posting-age figures — the median on the homepage, the totals on the duration page and the longest-open lists — were counting about 4,250 postings from three employers whose profiles we had declined to publish. The same three had already been removed from the job rankings; the statistics pages were a separate door and are now closed too. The headline count drops accordingly; the median is unchanged. /how-long-do-job-postings-stay-open
The email box no longer says only “no spam”. Every message we ever send will carry a one-click unsubscribe, and the mechanism behind that promise is now actually in place. We still do not name a sending schedule, because nothing sends yet.

August 25, 2026

The job rankings were including about 4,200 postings from three employers whose profiles we had declined to publish — they never appeared near the top, so the only visible sign was a slightly overstated “N results”. They are gone from the listing and the count. Separately, about 730 US postings had no country recorded and were missing from every US-filtered view; they are back. /leaderboard
Two employers joined the site with their openings already listed — Columbia Bank, the Pacific Northwest bank, with 217, and Affirm with 208. Regence (Cambia Health Solutions), which had a page but no postings, now shows its 61: we had the company all along and had not yet found its careers page. /companies

August 21, 2026

Two employers now read properly in headings and links: “Stephens (Stephens Inc.)” is now Stephens Inc., and “Susquehanna Int’l Group (SIG)” is now Susquehanna International Group (SIG). The legal names on file are unchanged — this is only what the page calls them. /companies
Two company pages now say what else the employer is called: Pape-Dawson Engineers notes that its name is also written without the hyphen, and United Talent Agency notes that it is widely known as UTA. Only where it is true — we did not add the misspellings and run-together spellings people also search for. /company/pape-dawson-engineers
Correction. An earlier entry today said the pages for CVS Health, Dollar Tree, Lowe’s and Allied Universal “now say plainly” that their careers pages are too large for us to read in one pass. They do not yet — checked again today, all four still show no postings, which is the very thing that entry described as fixed. The wording is built and shipped; it appears on those pages once the record of the incomplete read it draws from is in place. The earlier entry is left standing rather than edited, because a changelog that quietly rewrites itself is worth less than one that shows where it was wrong. /changelog
Lowe’s page now lists its openings — close to 12,000 of them, with the dates Lowe’s itself gives, half of them listed 17 days or less. Its careers page had been too large for us to read in one pass, so we had never published a count for it at all. /company/lowes-companies-inc
On a company page, the button beside the tabs now reads “Share experience” instead of “Report”. It always opened the form for telling us what happened when you applied — but next to a named employer, “Report” read as reporting the page itself.
The link to a company’s posting feed was rendering off to the side, clear of everything around it. It now sits at the foot of the page with the other links, labelled with the company’s name.
Four company pages that were headed by an abbreviation now use the full name — GD Information Technology, Nanyang Technological University, Easter Seals Southern California and Royal Melbourne Institute of Technology. Their web addresses are unchanged, so any link or search result you already have still works.
A page listed as the employer “Georgia” has been removed, along with about 1,100 postings it carried. It was not one employer: the board behind it is the State of Georgia’s shared careers system, and the jobs belong to 55 separate public employers — the Department of Labor, half a dozen technical colleges, and so on. Showing them under one name told you the wrong thing about every one of them. Where a parent company posts jobs for its own subsidiaries, that stays as a single page; this was not that.
Correction. An entry on 20 August said the University of Rochester was among eighteen company pages that had started showing open roles. That was wrong when we published it: the page had already been declined on review, so it was never reachable. Its postings were removed from the job board on 21 August and are no longer counted anywhere on this site. /changelog
A few very large employers — CVS Health, Dollar Tree, Lowe’s and Allied Universal among them — had pages saying they had no open roles. They do; their careers pages are simply bigger than we can read in one pass, and we will not publish counts from a partial read. Those pages now say that plainly, with how far we got, instead of showing a zero we never measured.
The job board was listing postings from an employer whose company profile we had declined to publish — the page itself was not reachable, but its 1,359 openings were browsable and were counted in the site-wide posting total. The board now checks that we publish a company before listing its jobs, so the total on the home page fell by about 1,400 to reflect what is actually shown. /jobs
Eleven openings at Calix were appearing under IBM, whose own postings we do not track at all. Calix now has its own company page and IBM’s no longer shows jobs that were never theirs. /company/calix
Several published figures were being calculated from only part of the companies we track, and are now calculated from all of them. The clearest one: the “median across the companies we track” quoted on every company’s longest-open page was measured over 1,000 employers and said so; it covers 1,115 now, and the median moved with it. The company ranking on the leaderboard was drawing from the same short list. /methodology
Each company’s “longest-open postings” page now ends with six other companies whose oldest open posting is closest in age to that one’s. There are 490 of these pages and until now the only way to reach any of them was through its own company profile — they had no way of reaching each other. /longest-open
Every company page now ends with a “More companies” list — six other employers we track, picked by the job board they post on, how many postings they have open, and their name. Around 555 company pages had no way out to another company at all, because the only list they had was grouped by industry and their industry was blank. Now every one of them can be reached from a neighbouring page. /companies
A company’s link to its own website is now announced by name to screen readers — “Kroger official website” rather than the address read out character by character. It affects the roughly 790 company pages that carry one.
Pasting a job link into your watchlist now finds the posting. Links copied straight from a company’s careers page — the ones carrying the job title in the address, or tracking bits added by LinkedIn and other job boards — were being turned away as postings we didn’t have, even when we were tracking them and had checked that morning. Links from a company’s own careers domain now work too. /watchlist
When a pasted link genuinely isn’t in our latest daily check, the watchlist now says that, tells you when we last checked that employer, and offers to watch the job anyway — it attaches itself as soon as the posting turns up. It used to refuse, and blame the job for being too old or already gone. /watchlist
The “longest-open roles” page for each company no longer gets its title cut off in search results. It was running to about double the length a search engine shows.

August 20, 2026

Company page headings now use the name people actually write — “Northrop Grumman” rather than “NORTHROP GRUMMAN CORP /DE/”. The full legal entity is still on the page, now stated outright as its own line in the company details rather than shouted in the heading.
Company pages now open with more of what we actually know: how long half that employer's postings have been listed, dated to our last read, and how many carry a date from the last month. The heading says which company's job postings you are looking at.
Eighteen company pages that had been sitting empty now show their open roles — among them Walmart, Dollar Tree, Citigroup, Thermo Fisher, AdventHealth, SSM Health and the University of Rochester. They were always listed here; we had not found their job boards until now. /companies
Seventy of the largest employers we could find now have pages with live postings, including Lowe's, T-Mobile, U-Haul, Booz Allen Hamilton, Five Below, Meijer and CarMax. Twenty-two of those boards refuse to list more than 2,000 postings at once however many are open, so those pages read “at least 2,000” rather than a count. /companies
Three company pages that had been sitting empty — LMI, Wipfli and AmTrust Financial Services — now show their open roles. They were always listed here; we simply could not find their job boards until now. /companies
A hundred and forty more employers now have pages with live job postings, among them WinCo Foods, Duluth Trading, The Container Store, Old National Bank, Envoy Air, Stifel and Fred Hutchinson Cancer Center. Their boards do not publish a total, so each page reads as a floor: at least this many roles are open, and the real number may be higher. /companies
Forty more employers now have pages with live job postings — among them Ryder, Pediatrix, The Salvation Army, Planet Fitness, Natural Grocers, MV Transportation and Rural King. Their boards do not publish a total, so each page reads as a floor: at least this many roles are open, and the real number may be higher. /companies
Fourteen more employers now have pages with live job postings — among them Planet Fitness, Pet Supplies Plus, Rural King, SP Plus and Cellular Sales. Their boards do not publish a total, so each one reads as a floor: at least this many roles are open, and the real number may be higher. /companies
The methodology page now says exactly what “jobs tracked” means: postings currently listed on an employer's own board, counted one per posting, with postings that have since come down reported separately. Where a board will not list everything it holds, the figure is marked as a floor rather than quietly topped up with the board's own estimate. /methodology

August 19, 2026

When we cannot read an employer's job board any more, its postings now say what date we last read it and stop counting up in age, instead of quietly looking current. We do not mark them as taken down either — we did not see them come down, we stopped looking. /methodology
Eight employers' job boards refuse to list more than a set number of postings at once, however many are open. Those companies now read “at least 2,000 roles” rather than “2,000 roles”, and we no longer show when one of their postings came down — on a board like that, a posting can drop out of sight because newer ones pushed it past the limit, which is not the employer taking it down. /methodology
The methodology page now explains when a monthly report names a specific company: only when one of its postings passes 5,000 days listed, checked automatically when the report is taken rather than chosen by anyone. /methodology

August 18, 2026

You can now follow a company from its page. Your dashboard then shows its open postings, how long its listings usually stay up, and how many postings have first appeared on its board since you last looked. /dashboard
The Watch button now tells you when a role could not be saved, instead of showing “Watching” either way. If you have watched a role and later found it missing from your watchlist, this is why. /watchlist
The job posting statistics pages now update within minutes of the daily refresh instead of trailing it by up to a day, so they no longer show a smaller dataset than the homepage was already reporting. /how-long-do-job-postings-stay-open
Everywhere we report how long postings have been up, we now say it the same way — “listed 90 days or longer” rather than “open more than 90 days” — and the methodology page defines that sentence once for the whole site. The figures have not changed: a posting listed exactly 90 days was always counted, and the wording now says so. /methodology
The share of postings still open at 30, 60 and 90 days is now measured the same way everywhere on the site, so the homepage and your dashboard no longer quote it a fraction of a point apart.
Your dashboard now opens with what the tracker saw today — how many postings we follow, how long the median one stays listed, the share listed 30, 60 and 90 days or longer, and the companies sitting on this week's longest-open roles — whether or not you are watching anything yet. /dashboard
Each role you watch is now drawn against every posting we track, so you can see where its age sits among them rather than reading the number alone. /dashboard
Job postings now show when we last saw them on the employer's own site, and postings we no longer find say so with the date.
New page: job posting statistics — the median posting age, the share listed over 90 days, and how many companies those figures cover, each stated with the population it is measured over. /job-posting-statistics
Every company page now has a feed you can subscribe to.
This page. /changelog

August 17, 2026

Company pages now open by answering, in one sentence, how long that company's postings stay open — how many we track, how long the median has been listed, the oldest, and when we last checked. Companies we have no posting data for say nothing rather than guessing.
A company's figure and the site-wide median now appear side by side, without either being characterised.
Every company count now states which population it counts, and the methodology page explains the four different ones. /methodology
New page: how long job postings stay open, measured across everything we track, with the full distribution and an industry breakdown. /how-long-do-job-postings-stay-open
The longest-open ranking moved to a name that describes what it measures. Old links still work. /posting-age-leaderboard
Reporting an incorrect listing now carries the listing's title into the message, so you do not have to describe which one you meant.
The email signup no longer promises a weekly email or an unsubscribe link that does not exist yet.
Company pages no longer cut a company's name in half in search results. Long names now get a shorter title rather than a truncated one.
Page 2 and beyond of every listing now tells search engines what it actually is, instead of claiming to be page 1.

August 13, 2026

A watch control on every job-board card, next to the posting's age. /jobs
A dashboard showing what happened to the jobs you are watching.
Duplicate company entries awaiting a merge no longer appear in lists.

August 12, 2026

Each company's longest-open postings now have their own page.
Job postings are browsable by location. /jobs/locations
An A-Z company index, with numbered pages. /companies
A readable error page instead of a blank screen when something fails.

August 11, 2026

Homepage counts now come from live data. A posting count that had been wrong is gone.
Signing out is immediate.

August 10, 2026

Search job postings by title, and the company directory is paginated.