Systems and tools Derived from public sources

Workforce blueprints

The blueprint estimates how many workers a planned facility will need, in which occupations and on what schedule, before the facility exists. No single source answers this, so it is built from filings, industry and regulatory standards, and comparable sites. The worked example is semiconductor fabs, with nine facility types, each with its own staffing mix and hiring ramp.

Economics · data science. This is the fourth demand layer of the labor market twin.
01

The problem

When a new facility is planned, the region needs to know how many workers it will take, in which occupations, and on what schedule. Usually the facility does not exist yet and there is no HR data to study, so we know what it will produce but not how it will be staffed. That leaves industry alliances, staffing agencies and states unable to prepare the local workforce, or even to tell whether it is ready. The same question also runs the other way, asking what the ideal staffing should be under industry and regulatory standards, such as nurses per bed adjusted for illness and mortality, or technicians per tool in a fab. No single source answers either question. Announcements give the investment and a headline headcount, staffing statistics describe the existing industry and miss a leading-edge facility entirely, and filings show scale and timing but say nothing about composition. So the blueprint has to be built from several sources.

02

How it works

01The staffing mix, per 100 workers

Industry staffing patterns are broken into job families and then reweighted with totals tuned to each facility type, because an analog fab, a leading-edge logic fab and a packaging plant are staffed very differently. Everything is expressed per 100 workers so it scales to any announcement.

Economics · data science

02Each source used where it is reliable

Each source covers different dimensions. Staffing matrices show composition but miss the newest facilities. Postings show demand but over-represent high-turnover roles. Filings show scale and timing but not the mix, so they are left out of the composition estimate. Where estimates disagree, they are combined into a band with a stated range.

Economics · data science

03The hiring ramp over time

Composition changes as a facility moves from planning to ramp-up, steady state and wind-down. Technicians are the largest group during ramp-up, and the engineering share peaks while tools are being installed. The staffing mix describes the steady state, and a blueprint that ignores the ramp misses the years when hiring is hardest.

Economics

04From facility to region

Multiplying the blueprint by announced capacity gives forward need by occupation and quarter. This is the fourth demand estimate in the labor market twin, and the only one available before any job is posted.

Economics · policy
Blueprint leading-edge logic

The staffing mix covers ten job families, measured from public staffing patterns and totals tuned to the facility type.

Example: semiconductor industry
Processing
16
Engineering
25
Software
11
Eng. technicians
8
Maintenance
8
Management
8
Other production
8
Business ops
6
Admin & logistics
5
All else
5
Each bar is a job family's share of the staffing mix, per 100 workers. Hatched marks the block unique to the sector.
Ramp select a type or a job

The workforce of one facility over its life. Band thickness is workers per line, and the total grows as the facility fills. There are nine facility types, each with its own staffing mix and ramp.

Facility type · nine identified
Composition changes over time, and differently for each facility type. Select a facility type to reshape the ramp.
What each source can see

No single source covers the whole blueprint. Each covers some dimensions and not others, and the staffing mix is built by combining them, using each source only where it has information.

MixScaleTimingFrontierPlace
Staffing matrices
Job postings
Financial filings
Trade announcements
covers partial blind
Filled means the source covers that dimension, hatched means partial coverage, and empty means none. Filings say nothing about the mix.
Derivation the full audit trail

The final per-100 share combines several sources. Each source with information on composition gives its own estimate, and the three estimates are combined into the number used in the staffing mix. Filings have no information on composition and are left out.

Occupation (SOC)StaffingPostingsFilingsTradeFinal /100
Semiconductor Processing Techs 51-914115.212.8abstains17.016.0
Industrial Engineers 17-21127.18.4abstains7.67.6
Software Developers 15-12525.98.1abstains6.46.7
Computer Hardware Engineers 17-20616.45.5abstains6.66.2
Electronics Engineers 17-20724.14.9abstains4.24.4
EE Technologists & Technicians 17-30234.03.6abstains3.93.9
Industrial Machinery Mechanics 49-90414.23.1abstains3.83.9
Maintenance & Repair, General 49-90712.72.2abstains2.52.5
Each cell is that source's estimate of the occupation's share of the staffing mix. Filings carry no information about composition. Staffing, postings and trade announcements are combined into the final share. Eight of more than forty rows are shown.
Nine types, two that dominate weight in the tracked build-out

Each facility type has its own staffing mix and ramp. The mix above is for a leading-edge logic fab.

Analog
41.5
Leading-edge logic
40.7
Memory
6.7
Compound
3.0
Advanced packaging
2.9
Mature node
2.8
MEMS / photonics
2.4
Normalized weights across the tracked build-out pipeline. Seven of the nine identified types shown.
03

Tools used

Python, pandas, numpyBuilding the staffing mix and combining the sources.
Staffing matrices, filings, postings, trade announcementsThe four public sources, each used for the dimensions it covers.
statsmodelsThe ramp and demand curves over the facility life.
The occupation resolverPosting titles standardized to the federal crosswalk.
Supabase, PostgresThe blueprint outputs feeding the labor market twin.
04

Final state

The blueprint method is built entirely from public sources, namely staffing matrices, filings, postings and trade announcements, and the staffing mix shown is a real derivation for a leading-edge logic fab. Nine facility types are identified, each with its own staffing mix and ramp. The facility scale shown here is illustrative, but the shares, and the rule of using each source only where it has information, come from the real work.

← Back to systems and tools