HOW THE SCORE IS BUILT.

The formula.

For every job in the dataset:

That's the whole model. Two numbers come from a 2023 Stanford and OpenAI study of which job tasks AI can do. Two come from Anthropic's quarterly data on what AI is being used for at work. One is a small adjustment per industry estimating how the score is likely to drift over 1, 3, and 5 years. Every weight is published. Every source is named below.

What each term means.

Eloundou α (alpha) — "Can AI do this job task directly?"

From Eloundou et al., "GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models" (OpenAI / Stanford, 2023). For each job, α is the share of tasks an AI can do directly, on its own — without any extra software wrapped around it — and cut the time the task would take by at least half. Range: 0–1.

Eloundou β (beta) — "Could AI tools do this task soon?"

From the same study. β is the share of tasks an AI can handle when paired with software someone could realistically build today. β is always at least as big as α. Range: 0–1.

Anthropic Index intensity — "Are people already using AI for this work?"

From Anthropic's Economic Index, refreshed every 3 months. We match a job to the U.S. Department of Labor's task list, then to Anthropic's data on which tasks AI is actually being used for today. Higher = real workers are using AI for that kind of work right now. Normalized to 0–1.

Anthropic automation share — "Is AI replacing or helping?"

Also from Anthropic's Index. Of the real AI usage observed, how much is AI doing the task for a worker (replacing) vs helping a worker do the task themselves (augmenting). Jobs where AI is mostly replacing score lower (more affected). Jobs where AI is mostly helping score higher. Range: 0–1.

Industry drift — 1, 3, and 5-year projections

Every job belongs to an industry. Each industry carries a 5-year drift number — how much the score is expected to drop if current AI progress and adoption continue. The drift values:

1-year drift is 25% of the 5-year value. 3-year drift is 65%. (A simple front-loaded schedule.)

What the score bands mean.

• 80–100 · HIGHLY RESISTANT. Mostly physical, hands-on, or relationship-based in ways AI doesn't threaten in 5 years.
• 65–79 · RESISTANT. The paperwork and routine parts are exposed, but the core is in-person, hands-on, or judgment-based work AI can't easily do.
• 50–64 · MIXED EXPOSURE. Several valuable tasks are in AI's reach, but the role still depends on skills AI can't copy.
• 35–49 · EXPOSED. A real chunk of the role is being done by AI today. Senior workers usually survive — but the path in for new hires is changing.
• 0–34 · HIGHLY EXPOSED. Reserved for jobs where AI can already do most of what creates value. As of this release, no job is in this band.

What the score is not.

The score is information, not personal advice. It doesn't know your specific employer, your city, how senior you are, or any of the dozens of other things that make a real career decision real. It's a starting point for a conversation, not the conversation itself.

It also doesn't say whether AI is good or bad for your work. Just a measure of how much AI is changing the work.

What we update, and how often.

• EVERY 3 MONTHS The two Anthropic numbers (intensity and automation share). Whole site is re-scored against every release.
• ONCE A YEAR Eloundou α and β, as new versions of the Stanford / OpenAI study come out.
• ONCE A YEAR U.S. Department of Labor task lists, when a new version is released.
• AS NEEDED Industry drift values, if two Anthropic releases in a row show our estimate is meaningfully off.

The honest caveats.

• Anthropic's data shows one company's AI usage, not all AI usage. It's the best public source we have. When other AI companies publish similar data, we'll blend it in.
• The model assumes AI keeps getting better at its current pace. If progress stalls — or speeds up dramatically — the 5-year projections will shift. We re-check every quarter.
• Some niche jobs use an estimated number rather than a directly measured one, when there isn't enough data on that exact role yet. Those jobs are flagged in the data file.
• The industry drift values are judgment calls, anchored to evidence but not yet derived from a formal statistical model.

The dataset.

Every number that produces every score is public. The current dataset is at /assets/data/occupations.json. If you spot a number that looks wrong, email admin@tradesmanpass.com with a source link. We'll fix it and credit you in the changelog.

Citations.

• Anthropic. Anthropic Economic Index — March 2026 release. huggingface.co/datasets/Anthropic/EconomicIndex. Licensed CC-BY.
• Eloundou, T., Manning, S., Mishkin, P., & Rock, D. (2023). GPTs are GPTs. arXiv:2303.10130.
• U.S. Dept. of Labor. O*NET Database, version 28.0. Public domain.
• U.S. Bureau of Labor Statistics. Occupational Employment and Wage Statistics.

Want to challenge a number?

Email admin@tradesmanpass.com with the occupation, the input you disagree with, and a published source. Real arguments improve the model. Public credit for substantive contributions.