Methodology
What we observe
We read applicant tracking systems on a daily schedule and record what postings exist. When a posting stops appearing, we record that too. Every figure we publish is a count of those observations — never an estimate, and never a projection.
That schedule is what we run, not what we have achieved. Coverage is not yet uniform: some employers are read on most days and some on very few, and no employer in the record has been read on every day since we started. We publish the number of days we have actually observed each employer on their own record rather than implying a continuity we do not have, and any figure derived from our observations is a count of the days we were there.
We do not observe the whole job market. We observe the boards we can reach, on the platforms listed below, for the employers enrolled in our crawl. Absence from our record means we did not see something. It never means it did not happen.
What the record currently holds
These figures are 17 days old, and so is the count itself: we last totalled the record on September 20, 2026. The crawl may have run since. We are showing the last complete count rather than a partial one.
| Employers on the entity spine | 240,505 |
|---|---|
| Employers with observed job postings | 1,089 |
| Job postings in the record | 295,482 |
| Individual observations recorded | 429,838 |
| Distinct days observed | 31 |
| Postings we cannot date | 166,211 · 56.3% |
| Member-contributed sightings held across 229 listings | 282 |
| Sightings promoted into the posting record | 107,702 |
166,211 of 295,482 postings were already live when we first saw them, so we know our first sighting date and not the employer’s posting date. These are recorded as left-censored and excluded from every duration figure.
Where the postings come from
| Source | Postings |
|---|---|
| workday | 214,387 |
| themuse member-contributed | 164,040 |
| greenhouse | 6,980 |
| icims | 6,055 |
| socrata | 4,356 |
| adp_workforcenow | 2,503 |
| lever | 2,412 |
| ashby | 902 |
| workable | 93 |
| recruitee | 40 |
| member_sighting_linkedin member-contributed | 39 |
The record is not evenly spread across the platforms we watch. It is also why so much of it is undated: the Workday adapter never parses the posted date the board displays, because that field is relative text rather than a date, so every posting from it is recorded as left-censored by design.
What we hold a posted date for
| Date reported by the employer's system | 122,218 |
|---|---|
| No posted date available | 166,211 |
| System serves a date that is not a posting date | 2,200 |
| Collected by a halted backfill — excluded from findings | 4,853 |
Dates taken from Workday's detail endpoint by the one-time backfill of schema/080, which was halted on 403 responses and is unresolved with the vendor. Counted among postings because the postings are real; excluded from every duration figure and from every published finding.
Platforms we watch
12 applicant tracking systems: adp_workforcenow, Ashby, Greenhouse, iCIMS, Lever, Recruitee, socrata, successfactors, themuse, usajobs, Workable, Workday.
This list is read from the running configuration, not typed here.
What members contribute
We do not crawl LinkedIn, Indeed, ZipRecruiter, Talent.com, Jobright or Adzuna, and we have no plans to. Members who choose to can tell us what they saw on those boards while they were already there. That is a different kind of record from the one we make ourselves, and it is labelled as one everywhere it appears.
What a sighting can say: that a listing was on a board on a particular day, and the day it was first and last seen.
What a sighting can never say: that a posting closed. Nobody re-checks it. A listing missing from a later page is a different search, a different session, or the second page of results — not evidence that it came down. When we say a posting closed, it is because we watched it ourselves, twice, on healthy runs.
A listing seen on one day is not written into the record at all. It takes two separate days before a sighting becomes a posting we keep, and paid employer reports do not read this tier at all — they read only what we observed directly.
The five rules this runs under
- Facts, never prose. The title, the employer as displayed, the location, the board’s own id for the listing, the link out, and the date as shown. Never the description itself. Never the page of search results, and never what was searched for.
- You have to switch it on. It is off until you turn it on, it is its own separate agreement, and turning it off loses you nothing else.
- The record never says who saw it. We check that you agreed and that you are inside a daily limit, and then we write the listing without you attached to it. The request is not anonymous; the record is. There is no column on that table for a person, and nothing in it can be traced back to one.
- Nothing is sent while you are looking. What you see is held in your browser and sent later — after a wait of between ten minutes and two hours, mixed together and in a shuffled order, dated to the day rather than the minute. When a batch arrives says nothing about when you were reading.
- Switching it on is the moment your browser asks. The permission to read those boards is requested when you agree, not bundled into the install, and it is withdrawn when you stop.
Turning it off deletes what is still waiting to be sent. The raw daily records are deleted after ninety days; what remains is the summary — the listing, and the first and last day a member saw it.
How we decide which boards to add
We watch the applicant tracking systems listed above and we would like to watch more. The hard part is not the watching — it is knowing which boards are worth it, because nobody holds a list of job boards and every list we write from memory is a list of the ones we have heard of.
So the extension tells us which boards our members actually use. When a contributing member opens a job posting on a site we do not already know about, it records the domain, and nothing else.
What is recorded, and what cannot be
- The domain, and only the domain. Never the address of the page, never the part after the question mark, and never what you searched for. There is no column for any of them, and the one that holds the domain refuses a value with a slash in it — so a page address does not fit in the record even by mistake.
- Only job postings. A page has to have an apply control and a job heading, and must not look like a list of results or a search. Ordinary browsing records nothing at all.
- Only if you have switched contribution on. The same agreement as above, and turning it off deletes what is still waiting.
- No column for a person. The record says a board was used, never who used it — and unlike a sighting it does not even carry the identifier that groups one person’s sitting together. There is nothing in it to trace back to anyone.
- Dated to the day. The first day, the last day, and how many days in between anyone met a posting there. When you were reading is not answerable from it.
We also record whether the page published its job details in a structured form. That is not about you — it is about what the board would cost us to support. A board that publishes structured data needs no code at all, so we can rank candidates by how much they are used against what they would cost, rather than by use alone.
You can also just tell us. Open a posting, press You should watch this board, and we get the domain and the fact that a person asked for it. That needs no agreement and no observation of anything: it is the highest-signal thing we receive, and it is the one we would rather have.
Thresholds
| Threshold | Value | What it governs |
|---|---|---|
| minimum_closed_postings | 5 | How many postings must have been seen to close before we will describe an employer’s follow-through at all. A median needs closes, not elapsed time. |
| minimum_history_days | 30 | A floor, not the gate. However many closes arrive, we will not characterise an employer we have watched for less than this. |
| thin_postings_at_or_below | 3 | At or below this many postings ever, we describe the record as thin. Posting rarely is not a red flag; it is a fact about how an employer hires. |
| employer_index_min_observations | 5 | How many postings we must have observed before an employer page is listed in search results. Below it the page is still reachable and still accurate — it is simply too thin to be a useful search result. A coverage threshold, not a confidence one. |
| link_publication_confidence | 0.9 | How confident a link between a public filing and an employer must be before we show it. A record that cannot be matched unambiguously is attached to nobody rather than guessed at. |
These are provisional. The ledger is weeks old, so none of them has been calibrated against outcomes yet, and all of them are expected to move. When they do, this page changes and the change is listed below.
What we refuse to say
We do not publish claims about intent. We record that a posting was up for a number of days, that it closed, or that it reappeared. We do not say why, and no figure we publish should be read as saying why.
We do not score or rank employers. There is no model behind any page on this site, and where a reader might expect a verdict they will find the observations instead.
We do not record which postings a person looked at. The extension asks about a page you have open and sends no identifier with the question.
Where each row came from
Not every row in the record came from our crawler, and the record says which is which on every row rather than flattening them together. Where a fact came from is part of the fact.
LinkedIn is the case worth stating plainly. Our crawler never fetches that host. There is no crawl target for it and no adapter that resolves it, so there is nothing in our system that could fetch it even by mistake. The rows we hold from it arrived from members who switched contribution on, through the extension, about pages they had opened themselves. The same is true of Indeed, ZipRecruiter, Talent.com, Jobright and Adzuna, where we hold no rows at all.
A member-contributed listing becomes part of the posting record only after it has been seen on two separate days, and when it does it is marked as member-contributed permanently and held at our lowest source authority. Paid employer reports do not read that tier at all.
Dates
Every date on this site is UTC, stored and displayed. Not converted to the reader’s local time: an observation dated September 6, 2026 is dated September 6, 2026 for everybody, because the date is part of the record rather than part of your experience of it. A reader in one timezone and a reader in another have to be able to cite the same page.
Where an employer’s system publishes a posted date, we record it and label it as theirs. Where it does not, we record the date we first saw the posting and label it as ours. The two are never mixed, and a figure derived from our first sighting is always described as a minimum: a posting we first saw 12 days ago has been open at least 12 days.
Some systems serve a “posted” date that changes when the listing is refreshed. We do not treat those as posting dates, and we do not republish them as ours.
What a close date is
A close date is the first time we observed the posting to be absent, not the moment the employer removed it. The actual removal happened somewhere between our last sighting of the posting and that observation.
How wide that window is depends on how often we read that employer’s board, which is not uniform — see above. On an employer we read daily the window is about a day. On one we have read three times in three weeks it is much wider, and the employer’s own record shows how many days we have actually observed them so the window can be judged rather than assumed.
We require the posting to be absent on two consecutive healthy reads before recording a close at all. A single failed read makes every posting at a source look absent at once, so a run we cannot vouch for is excluded rather than counted.
Changelog
- 1.1 — . Corrected the crawl cadence: version 1.0 said the platforms were read every morning and that we polled once a day. That was our schedule and not our record, and it was not true of any employer we hold. Added what a close date is and is not. Added that all dates are UTC. Added the source provenance section. Added the live coverage figures.
- 1.0 — . First publication.
When this page changes, the previous version stays reachable so that a figure cited under it can still be checked against the rules that produced it. Previous versions.