what this pipeline actually did

We publish the number, whatever it says.

In the 7 days up to our last completed run we screened 176 newly collected roles and cleared 1 of them as genuinely work-from-anywhere.

176roles screened in our last 7-day window
1cleared as work-from-anywhere

Measured by the run that finished on . “Screened” counts what we collected and read in that window — not when the role was posted. Some of it is older than the window.

01 · the method

One pipeline, four checkpoints.

Every listing passes the same four stops. Nothing skips a stop, and nothing reaches a person until it has passed all four.

1

Collect

Every night we pull new postings from job-board and applicant-tracking APIs — RemoteOK, Arbeitnow, Himalayas, Working Nomads, JSearch, Greenhouse, Lever and Ashby. We do not scrape sites whose terms forbid it.

2

Classify

A keyword pass is only the first filter, and it is deliberately generous. The decision is made by a model that reads the whole description and assigns the role a category and a remote scope.

3

Verify

A role that scores well against a member’s CV gets a second pass, which re-reads the description for hiring geography and work authorisation before the role is delivered.

4

Deliver

What survives is scored 0–100 against a member’s actual CV and lands on their dashboard, with email and WhatsApp alerts on the paid plans.

02 · the definition

What we mean by “work from anywhere”.

A role is judged on what the posting itself says — not on the word “remote” appearing in its title.

What the model is shown
One model call decides this, and the posting reaches it as a short list of fields: title, company, listed location and source board, and the first 8,000 characters of the description. The rest of the description is not sent, and no other part of the row we stored is sent. No person reviews the answer, and we never ask the employer.
What the number is
A model’s reading of that text. It is not a guarantee and it is not a promise from the employer. Where the text gives no reliable signal either way the classifier is instructed to answer unclear, and unclear is not counted — a rule that pushes this figure down rather than up.
Where this figure is soft
The shape of the error, rather than a claim that there is none. A role inside this figure can carry an employer’s general note that some of its roles have location-based eligibility requirements; we do not read that as a restriction on the role in front of it. A role inside this figure can also have a listed location that names a country or a city — some of them say hybrid — while its description reads as open, and the place is sometimes named in the job title rather than in the location field at all. On postings like those the model can be wrong, and this figure carries that error. Read the posting before you apply.

Both numbers above are read from one row this pipeline wrote about its own run — the same row the member dashboard reads. There is no second, friendlier figure kept for this page.

03 · where we look

8 boards and ATS APIs.

Public APIs and public applicant-tracking endpoints. This list is read from the same configuration file the nightly collector obeys, so a board we have switched off cannot stay named here.

Aggregated from

RemoteOK Arbeitnow Himalayas Working Nomads JSearch Greenhouse Lever Ashby

See them scored against your own CV.

A free account scores every role we clear from 0–100 against your actual CV, and includes 1 CV review a week.