RoleCraft reads occupational data the way a pilot reads weather: not to predict the exact future, but to know how much runway you have before the work changes. Here is how the score is built and what the labels mean.
01
Runway score
Each occupation gets a 0–100 score. Higher means more of the work is expected to stay human; lower means more of the underlying tasks are exposed to near-term automation. The score is not a forecast of job losses — it is a reading of how much of the role still has a protected runway.
0–100. Higher = more of the work stays human.
02
Job-level view: exposure bands
At job level the map uses three bands derived from the runway score: High AI exposure (runway below 40), Medium (40–69), and Low (70 and above). Words like Automate, Augment and Anchor describe tasks inside a job, not the whole job — so they never appear as job-level labels here.
High < 40 · Medium 40–69 · Low 70+.
03
Task-level view: the three zones
Inside each occupation, every O*NET task sits in one of three zones. Automate: observed AI usage on the task is automation-flavoured — AI executes, a person checks. Augment: observed usage keeps a person in the loop. Anchor: no observed AI usage has reached the task. The dominant zone and the full split are shown in the watchlist and detail panel.
Automate + augment + anchor always sum to 100% of a role's tasks.
04
Score composition
The score blends what AI is observed doing (Anthropic Economic Index usage, log-scaled, plus Microsoft's applicability reading) with what it could do (the Eloundou capability ceiling) and how automation-flavoured current usage is. Six signals carry it: automation momentum, task coverage, AI traffic, augmentation tilt, capability ceiling, and second-source confirmation. Each bar in the composition chart is a signed contribution to the score, not the raw level of the signal. A low AI-traffic level with a positive contribution is correct: little traffic protects the score.
Bars are signed contributions to the score, not levels.
05
Trend, pressure line and dates
The pressure line sits at 40; occupations below it lead the watchlist. The 13-week trajectory is a slice of the path between the two Economic Index vintages (2025 and 2026) — an interpolation between two real measurements, not a weekly sensor. Inflection dates project when a declining trend crosses the line, which is why they appear as coarse periods rather than exact days.
Pressure is exposure, not observed displacement.
06
Estimated rows and Born since 2024
169 occupations carry an ESTIMATED chip: the usage data has never observed them — janitors, construction labourers, welders — so they are scored from the capability ceiling alone, with no trend and no AI-traffic reading. They are included so the map shows the whole measured economy, not only the AI-visible part. The four Born-since-2024 rows are AI-native roles found in job postings with no official name yet in SOC or O*NET; they carry no measured employment.
Estimated = ceiling only. Born since 2024 = too new for official statistics.
07
Sources and scope
United States, national level. Anthropic Economic Index (observed usage; May 2026 vintage, with the 2025 vintage powering the trend), Eloundou et al. 'GPTs are GPTs' (capability ceiling), Microsoft 'Working with AI' (Tomlinson et al., observed Copilot usage), O*NET task statements, OEWS May 2025 employment counts, Lightcast Open Skills (August 2026 capture), combined by the RUNWAY-V1 scoring model.
Sources: AEI · Eloundou · Microsoft WWAI · O*NET · OEWS May 2025 · Lightcast · RUNWAY-V1.
The score is a model, not a verdict. Use it to start conversations, not to make final decisions about hiring, training, or workforce planning.