- A sample, not a census. The positions come from 975 pages captured over a limited period. They describe what those listings asked for, not the whole job market, and not demand over time.
- Many positions name no skills. Only 1,265 of 4,022 positions list any technical skill. Search-result pages often show only a title, and non-technical roles rarely name tools. Shares of all positions therefore understate demand inside technical hiring; this site also reports shares of skill-tagged listings.
- Extraction is automated. A language model read every page. It can miss skills, misread a level or assign a borderline role to a neighboring family. Errors are individually small but not zero.
- Listings are not jobs. A listing describes what an employer asked for when posting, including optional extras. It does not show what the hired person actually used.
- Geography is uneven. Many sources were scoped to the United States, with smaller shares for Canada, Europe, Singapore, Australia and India.
- Employer counts reflect the sample. An employer with many positions may simply have had a well-indexed career page. Employer figures are not a measure of hiring volume.
- Small numbers are noisy. Pages flag skills with fewer than 10 positions, and breakdowns of small groups should be read as directional.
Methodology
Limitations
What this data can and cannot tell you.