Skip to content

jricciardi/adaptive-capacity-explorer

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

10 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Adaptive Capacity Self-Assessment Tool

Try it live

A browser-based tool that helps workers understand how prepared they are for AI-driven labor market transitions, based on peer-reviewed economics research.

What this is

This is a research translation project. Economists at Brookings measured "adaptive capacity" across 752 occupations — a composite of skill transferability, geographic labor market density, financial reserves, and age — and published the framework in Manning & Aguirre (2026). This tool makes that framework explorable for individual workers.

You select your occupation, enter a few personal inputs, and get back a composite score with a breakdown of what's working for you and what isn't. All computation runs in your browser. Nothing is stored, sent, or tracked.

Running locally

No build step. No dependencies. Just a static site:

python3 -m http.server 8000

Then open http://localhost:8000.

How it works

The composite score combines four equally weighted components, converted to Z-scores and mapped to a percentile via normal CDF approximation:

Component Source What it measures
Skill transferability O*NET skill profiles + BLS employment projections How easily your occupation's skills map to other growing occupations
Geographic density BLS OEWS + Census CBSA data How many alternative employers exist in your labor market
Financial runway User input (months of expenses covered) How long you could search for a good-fit role rather than the first available
Age factor User input Time horizon for recouping retraining investments

Transferability and density are occupation-level benchmarks from the study. Financial runway and age are personal inputs mapped to population distributions using simplified assumptions.

Data pipeline

The data/ directory contains pre-built JSON files, so running the pipeline is optional. If you want to rebuild from source:

Requirements: Python 3, numpy, openpyxl (pip install numpy openpyxl)

O*NET 30.1 database files must be downloaded separately from onetcenter.org and extracted to /tmp/onet_30_1/db_30_1_text/.

Run scripts in order:

Script What it does
01_parse_onet_skills.py Parses O*NET 30.1 skill + work activity importance ratings into 76-dimension vectors per occupation
02_compute_skill_similarity.py Percentile-rank normalizes profiles, computes pairwise cosine similarity across all occupations
03_parse_bls_projections.py Downloads BLS 2024–2034 employment projections (employment, growth, wages, education)
04_compute_transferability.py Implements the Manning & Aguirre transferability formula: employment-weighted similarity to growing occupations
05_compute_density.py Implements the Manning & Aguirre density formula: employment-weighted average log(employment/area) across metros
06_bundle_ai_exposure.py Downloads Eloundou et al. AI exposure scores from the GPTs-are-GPTs repository
07_build_benchmarks.py Merges all sources into a single per-occupation benchmark table with composite scores
08_build_filter_tree.py Builds a guided quiz decision tree for occupation selection

Data sources

Source Version
O*NET 30.1 (December 2025)
BLS Employment Projections 2024–2034
BLS OEWS May 2024
Census CBSA Gazetteer 2024
AI exposure scores Eloundou et al. (2024)

Limitations

  • Skill transferability is measured by O*NET skill profile similarity, which may not capture all relevant factors in career transitions.
  • The composite score combines occupation-level benchmarks with self-reported personal inputs using simplified distributional assumptions for wealth and age.
  • Occupation-level averages mask within-occupation variation. Two software developers may have very different adaptive capacity depending on their specific skills, employer, and network.
  • Geographic density uses the occupation's national employment distribution, not the user's specific metro. The metro selection provides context but does not change the density score.

Research

Manning, A. & Aguirre, J. (2026). "Who Bears the Costs of AI Displacement?" NBER Working Paper 34705. nber.org/papers/w34705

Eloundou, T., Manning, S., Mishkin, P. & Rock, D. (2024). "GPTs are GPTs: Labor Market Impact Potential of LLMs." Science 384(6702). doi.org/10.1126/science.adj0998

License

MIT

About

Browser-based adaptive capacity self-assessment based on Manning & Aguirre (2026), NBER Working Paper 34705

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors