Systems and tools Illustrative · synthetic data

Opportunity, and what moves it

Every occupation is scored on five published pillars, and policy levers are attached only where the research supports them, each with a confidence band.

Policy · economics. Live in one state across 227 scored occupations. Architected for a ten-state cohort.
01

The problem

A workforce board mostly wants to know which occupations to invest in and what would help them grow. Composite scores usually let them down in two ways. The score is a black box whose weights no one can defend, and the interventions come without evidence grades, as if a pilot study and a randomized trial carried the same weight. Both problems can be fixed, and most of this work goes into fixing them.

02

How it works

01Five pillars, all shown

Demand, wages, access, transitions and resilience are each measured from public data and shown next to the score, so no one has to take the index on faith. A wage floor is reported alongside as metadata, outside the weighting.

Economics · data science

02Stability, tested

Rankings are stress-tested by shifting the normalization anchors. If an occupation's rank stays put it can be trusted, and if it flips, the index flags it. Rank stability stays above 98% under these tests.

Econometrics

03Levers only with evidence

Interventions come from a curated research corpus. Each one carries its effect range, its interval, and a breakdown of the evidence into causal, quasi-experimental and observational studies. Null results are kept, and every magnitude comes from a published study.

Policy · economics

04Synthesis where decisions happen

The index, the levers and the forward-looking signals come together in standing dashboards on one shared data lake, with transition pathways for each occupation and state-level views a board can put in front of its members.

Policy · product
Opportunity index click an occupation

Every occupation is scored on five pillars and the index ranks them, with the pillars shown so the score can be checked.

OccupationIndex
Pillar values here are illustrative and synthetic. The method, the pillar definitions, and the rule that pillars are always shown come from the real system.
Policy levers select a lever

Estimated effect on pipeline completions, from the research literature, each shown with its confidence band.

Ranges come from the published literature. The evidence bar splits the study base into causal, quasi-experimental, and observational.
Observatories three studios · select one

Three standing dashboards run on the same data lake, behind one selector and one layout. Each studio covers one topic with its own set of live views, and each view shows the level it runs at.

The views listed are live, each on the shared data lake. The figures show the level each view operates at and do not report results.
03

Tools used

Python, pandasPillar construction and the composite score.
scikit-learn, statsmodelsNormalization and the rank-stability perturbation test.
Curated research corpus, controlled vocabulariesThe lever evidence library, tagged by study design and setting.
Supabase, PostgresThe shared data lake behind every surface.
Dash, PlotlyThe index, the levers, and the observatory studios.
04

Final state

The index runs in production in one state, with 227 occupations fully scored, and is built to roll out across a ten-state cohort. The lever library holds a curated corpus of about 1,200 research findings mapped through controlled vocabularies, with evidence tiers attached and null results kept. The occupations, pillar values and lever magnitudes shown here are synthetic, so that no client's numbers are shown.

← Back to systems and tools