METHOD
This tape ranks companies by how many AI postings they newly published, not by how famous they are and not by how many roles happen to be open. The rules below are the rules in the code. There is no second, private classifier.
GATELESS
Every number comes from a public JSON feed that does not require an account, a token, or a session. Reading the tape does not require an account either. The same rows are at data/leaderboard.json. Pages do not load third-party scripts, fonts, or pixels.
WHAT COUNTS AS AN AI POSTING
A posting counts when the title, department, or team matches a phrase or a short token below. The job description is ignored, so a sales role that mentions the product in the body does not count unless the title or the org name does. Short tokens use word boundaries, so HTML is not ML and email is not AI.
Exclusions are applied to the title first and they win. A recruiter inside a machine-learning department does not count. An "AI Account Executive" does count, because the title says AI. That over-counts some go-to-market roles. Generic "Software Engineer" does not count unless the department or team also matches, which under-counts model work titled as ordinary software.
PHRASES
TOKENS
TITLE EXCLUSIONS
| TITLE | DEPARTMENT | RULE |
|---|---|---|
| Machine Learning Engineer | — | AI |
| Member of Technical Staff | — | AI |
| Software Integration Engineer, Self-Driving | — | AI |
| Perception Engineer | — | AI |
| AI Tutor - Croatian | — | AI |
| Digital Native AI Account Executive | Sales | AI |
| Software Engineer | Machine Learning | AI |
| Research Scientist | Research | AI |
| Forward Deployed Engineer | — | AI |
| Senior Software Engineer | Infrastructure | NO |
| Account Executive | Sales | NO |
| User Researcher | Research | NO |
| Recruiter | Machine Learning | NO |
| Barista | — | NO |
| HTML Developer | — | NO |
| Email Marketing Manager | — | NO |
| Data Analyst | — | NO |
| Carpenter (Construction) - Memphis | — | NO |
Domain words such as autonomous, perception, generative, and inference will sometimes mark a role that is not a model job. The match table is the whole rule.
VELOCITY
VEL/7D is the count of distinct AI postings whose public publish timestamp falls in the trailing 7×24 hours ending at the fetch time. The same posting id is counted once. Daily snapshots are unioned across that window, so a role that was open earlier in the week and filled before the latest fetch still counts if a snapshot saw it and its publish time is inside the window.
Roles that opened and closed between fetches are invisible until the next day's snapshot exists. The first run can only see roles that are still open. Publish fields are the vendor's, copied as UTC. Updated-at is not treated as a publish time. A missing timestamp keeps the role in AI OPEN and leaves it out of VEL.
RANK, MOVEMENT, SPARKLINE
Sort is velocity descending, then open AI roles, then name. That tie-break gives every row a unique rank. Δ is the previous snapshot's rank minus today's rank. ▲ is a move up. A company new to the tape shows —. The sparkline is that same velocity at each of the last 14 snapshots. One point means there is only one snapshot so far.
FEEDS
Zero auth. A board that does not return the expected JSON fails the run. The failure names the employer, vendor, URL, and status. The previous tape stays published. An empty job list is a real zero, not a failure.