The names, and the postings that put each one on the list.
Confidence describes the posting evidence, not the employer's intentions. Verify the interpretation before using it in outreach.
Ranked by signal strength, strongest first. Click a row to lens the whole dashboard onto that employer; click it again to release it. "See the roles" opens the postings the line was counted from.
Each row is a signal. Ask whether the roles are expansion, replacements or ongoing recruitment.
A pause in adverts does not establish a hiring freeze. Ask whether hiring has finished, moved to another channel or paused; the postings cannot settle that question.
Volume, mix and direction across the roles the filters leave in view.
Almost everything below is clickable: a month, a discipline, a seniority, a region or a slice lenses every other page onto it. Click the same mark again to release it.
| Group | Roles | Share | Employers | Remote | Median days to close |
|---|
Bars are the roles in view. The movement table below them compares each group's share of the current window with its share of the one before it, in percentage points, so a group can grow in count and still lose ground.
The place written on the posting, cleaned but not re-geocoded. A national or multi-site advert names no place and is counted in the missing share above.
Read from the posting's own wording.
Contract type as advertised. Where a posting states nothing it is counted as unstated rather than assumed permanent.
Same clock as the chart on the left, cut by level instead. A senior band with a long tail is a search problem; a junior band with a long tail is usually a pay or location problem.
A wide band is the useful part. Where p25 and p75 sit far apart the median is a weak summary and the long tail is where an introduction is worth most.
The counterweight. A title that closes in days is one the employer can serve itself, and quoting it alongside the slow list is what makes the slow list credible rather than a sales line.
Counts of roles in view, not pay. This chart does follow the filter bar. A level with a high count and no band above it has the rows but not the agreement between them; a level with neither is a level nobody is publishing on.
Movement, uncontested lanes, and the names whose collection has only just begun.
Newly-collected employers are badged : their rise is a coverage event, not a surge.
Each card carries the trajectory behind the number. A card marked New coverage means collection of that employer began inside the comparison window: there is no prior window to compare against, so the change is withheld rather than guessed at.
A discipline-and-seniority pairing with open roles at exactly one employer. Read it as category leadership or as an uncontested lane, and read a lane holding a single posting as the thin claim it is. The count is printed so you can tell them apart.
Every posting the current filters leave standing, with the account list above it.
Search and sort on this page narrow this table only. The filter bar above still governs every other page. Click an employer to lens everything on them.
One row per employer in the current view, ranked by what they still have open. "Why call" is assembled from the counts beside it and states nothing the rows do not already say. The export writes every employer in the view, not only the ones drawn here, and carries the filters that produced it in the file header.
What this build reached, what it could not, and how every derived field was derived. Read this before you read a month-on-month move anywhere else.
Every classified field carries the method that produced it. A language model reading the posting is the primary path; the ML tier and the rule tier are fallbacks used when it is unavailable or unsure, and "not derived" means the field was left empty rather than guessed. A build leaning on the fallbacks is still usable, it is simply less confident, and you can see how much.
Measured over the roles still live in this build. The bar is the share of rows carrying a value; the count beside each label is the number of rows without one. A missing field is left blank everywhere on this dashboard. It is never filled with an average, a guess or a zero.
"Reached" means the employer's careers feed answered on the last run. An employer marked failed is not an employer that stopped hiring: the collector could not read their feed. Roles collected on earlier runs stay on this dashboard with their last-collected date and may be stale; an employer with no collected roles at all is absent from every figure here until the feed answers again. The note beside each status says which of those it is. Paused employers were switched off on purpose when this market was scoped, and are excluded from the reached count.
The caveats this build generated about itself. They are worth repeating in a client meeting: stating them is what makes the rest of the numbers credible.