PersonaOSSettings
Product context
New grounding rounds are drafted for this product by default.
Data in this browser
Grounding rounds, grounded versions and recently opened personas are stored locally; the public persona population is never written to.
How the population was built
- 1Title space first
A taxonomy of industry → sub-area → role family → base role was multiplied across specializations and ten seniority scopes (Junior, Associate, Mid, Senior, Lead, Principal, Regional, Head of, VP of, Global): 122,596 unique titles, exhausted before any firmographic expansion.
- 2Firmographic expansion to 3M
Each title was expanded breadth-first across company size band × industry segment, so persona t + 1 + k·122,596 is always variant k of title t. Every row is unique.
- 3Expertise last
A modeled expertise score was assigned after the full population existed, and the global tiers were cut from its real distribution.
Data sources
What each layer is loaded from, and its licence.
- O*NET 30.1 database
U.S. Department of Labor / O*NET Center, CC BY 4.0. Occupation data, work context, tasks, knowledge, skills, interests, work values, work styles, technology skills, job zones, education. Each element keeps its domain source (Incumbent, Analyst, Machine Learning, AI/Expert).
- BLS OES, May 2025 national
Bureau of Labor Statistics Occupational Employment and Wage Statistics: employment, mean and median annual pay, 10th and 90th percentiles, matched on the SOC code (falling back to the broad group where a detailed code is not published).
- PersonaOS title space
3,000,000 rows across 122,596 titles, 20 industries, 58 sub-areas, 90 role families and 407 base roles, expanded across four firm size bands and industry segments, with a modeled expertise score and global tiers.
What is real, what is modeled, what is synthetic
- Personas are synthetic, generated from a structured taxonomy modeled on O*NET/SOC occupational structure. They are not a government title list and not scraped from any database. The occupation layer, by contrast, is the public O*NET and BLS data itself, shared by every persona on the same base role.
- Responsibilities, goals, frustrations and expertise profiles are template-derived, keyed to role family and seniority: internally consistent, not individually authored.
- The expertise score is a modeled ranking device for the tiers, not a measurement of any real person.
- Display names are generated deterministically from the persona id so a persona reads as a person rather than a record. No name here belongs to anyone.
- Voice, context modulators, constraints and the trait sketch are derived by fixed rules and carry the Generated label. A grounding round is the only way an attribute becomes human-answered, and those carry the quote.
- The dataset is served as ~40 MB of static files: a persona packs into 9 bytes because every free-text column is a function of the title, of the seniority band, or a pick from a small phrase pool. No server, no model.