Alion Labor Market Data
Collection
Job postings read from employers' own hiring systems: ghost jobs, pay, who is hiring, extraction labels. • 8 items • Updated
Datasets:
You need to agree to share your Hugging Face information to access this dataset
This repository is publicly accessible, but you have to accept the conditions to access its files and content.
This dataset is free. Alion asks who uses it so we can tell you when a new monthly version or a related dataset comes out. The data is licensed as stated below; please credit Alion with a link to alion.io.
Log in or Sign Up to review the conditions and access this dataset content.
Ghost Job Detection
Will a job posting that is already old ever come down? Real lifecycle labels, not guesses: 20,000 postings that were live on employers' own career sites on 2026-09-23, described as they looked that day, and what happened to each by 2026-10-28 - closed, or still up.
Every other public "fake job" dataset is a hand-labelled snapshot. These labels come from Alion re-reading each employer's own applicant tracking system every 48 hours and journaling when every posting went up, came down and came back.
- Rows: 20,000 (train 16,090, test 3,910 - split by employer, no employer on both sides), sampled from 290,586 postings live on 2026-09-23 on platforms that publish the original posting date.
- Still up at the outcome date: 66.2%. Flagged as a ghost by Alion at the outcome date: 12.5%.
- Observation moment = the 2026-09 edition of the Ghost Jobs Index.
Columns
| Column | Meaning |
|---|---|
| posting_id, alion_url, title | the posting (its live page on Alion) |
| role_family, seniority, work_mode, country | classified from the posting |
| employer_size, staffing_agency, ats_platform | the employer and its hiring system |
| salary_stated | the advert states pay |
| posted_date, age_days_at_observation | the employer's original publication date and the age on the observation day |
| reopened_before_observation, reposted_before_observation | taken down and put back / re-published before the observation day |
| employer_live_postings_at_observation, employer_share_over_90_days_at_observation | the employer's board that day |
| employer_median_days_to_close | the employer's usual time to close a role (computed over its history, so it also sees the months after observation) |
| still_up_at_outcome | the target: still listed on the outcome date |
| closed_date, days_from_observation_to_close | when it came down |
| alion_verdict_at_outcome | Alion's ok / stale / ghost verdict for postings still up (method) |
Use
from datasets import load_dataset
ds = load_dataset("alion-io/ghost-job-detection")
Typical tasks: predict still_up_at_outcome (survival of old postings), time-to-close regression, calibrating ghost-job heuristics.
Limits
The employers Alion tracks, weighted toward technology and toward companies that run their own hiring system. A posting "closed" when the employer's board stopped listing it - that is not proof of a hire. Licence CC BY 4.0: credit Alion with a link.
More data
This is a frozen sample. The full, daily-updated data (live postings from 80,000+ employers' own hiring systems, companies, pay, hiring signals) is on alion.io and through the Alion API and MCP server.
Other Alion datasets: Ghost Jobs Index · ghost-job-detection · job-posting-extraction · salary-benchmarks · whos-hiring · role-technology-taxonomy
Questions, custom cuts, bulk access: info@alion.io
- Downloads last month
- 13