The Minerals Workforce Pipeline Model¶
Can announced federal investments close America's critical minerals workforce gap in time?¶
Project FORGE — Industrial Capacity Notebook 01 (v0.2) Author: Edikan Udofia · Published: August 2026
This model was developed through the author's doctoral research in operations research. All inputs are cited public figures (VERIFIED) or explicitly flagged assumptions with sensitivity ranges. Version 0.2 completed the primary-source verification pass; remaining items for the journal version are listed in the Limitations section.
The question. In August 2026, the federal government announced $180 million for mining education, including a $100 million DOE grant program intended to "dramatically increase" the number of degree-holders in mining and critical minerals fields within two years, alongside a $32.7 million technology development hub at the Colorado School of Mines. This notebook asks a question the announcements do not answer: given the time lags built into degree production, can expanded university enrollment arithmetically close the workforce gap on the timeline national policy requires — and if not, what does the shortfall demand?
The method. A transparent stock-and-flow (cohort pipeline) model of the U.S. minerals workforce, run at two tiers: (1) the degreed engineer tier (mining/mineral engineering degrees) and (2) the skilled practitioner tier (the technician and operator workforce that constitutes the large majority of replacement demand). All inputs are either VERIFIED public figures with citations or clearly labeled ASSUMPTIONS with sensitivity ranges.
Independence note. This is independent analysis on public data. It is not affiliated with, funded by, or endorsed by the Department of Energy, the Department of Defense, the Colorado School of Mines critical minerals hub, or any government program.
1. Data layer¶
VERIFIED inputs (public, cited)¶
| # | Quantity | Value | Source |
|---|---|---|---|
| V1 | U.S. mining workers expected to retire by 2029 (">half the current workforce") | ~221,000 | SME, "Workforce Trends in the U.S. Mining Industry" (smenet.org technical briefing); cited by CSIS (Dec 2024) and in Congressional testimony (W. Copan, House Natural Resources, 2023) |
| V2 | Mining & mineral engineering degrees awarded, 2020 | 327 | CSIS, "The United States Needs More than Mining Engineers..." (Dec 2024) |
| V3 | Net drop in U.S. mining/mineral engineering graduations since 2016 | −39% | CSIS (Dec 2024) |
| V4 | U.S. mining/mineral engineering programs, decline | 25 → 14 | SME, "Maintaining the Viability of U.S. Mining Education" (smenet.org technical briefing); CSIS (Dec 2024) reports 15 as of 2023 |
| V5 | Students enrolled in U.S. mining engineering programs, 2023 (vs. ~8 yrs earlier) | <600 (vs. ~1,500) | SME/SOMP data as reported by CBS News (Aug 7, 2026); primary annual series is the SME Guide to Minerals and Materials Science Schools (obtain for journal version) |
| V6 | Mining jobs to fill per year, next two decades (replacement + growth) | 11,000–13,000 | SME (via The Oregon Group; reported by MINING.com, Feb 2024) |
| V7 | U.S. mining sector job vacancies, Apr 2023 (vs. 2022) | 36,000 (vs. 27,000) | SME reporting, MINING.com (Feb 2024) |
| V8 | Federal mining-education package announced Aug 7, 2026 | $100M DOE (PROSPECT, Notice of Intent); ~$81.3M defense package: Mines $32.7M, SD Mines $25M, Johns Hopkins $23.6M | DOE press release, energy.gov (Aug 7, 2026); press reporting (CBS, OANN, Washington Examiner, Aug 7, 2026) |
| V9 | PROSPECT's stated goal (official wording) | "double the number of graduates with degrees related to mining, minerals, and associated supply chain technologies within two years"; eligibility includes universities, community colleges, trade schools, industry partners | DOE press release, energy.gov (Aug 7, 2026) |
| V10 | Geoscience workforce expected to retire by 2029 / projected shortage | 27% / ~130,000 | American Geosciences Institute (via Grist, Feb 2024) |
| V11 | Projected annual openings, mining and geological engineers, 2024-2034 | ~400 per year (mostly replacement) | U.S. Bureau of Labor Statistics, Occupational Outlook Handbook, Mining and Geological Engineers (bls.gov) |
Verification status: V1 traced to SME's own briefing page. V2-V3 from CSIS (primary analysis). V4 traced to SME's own briefing. V8-V9 verified against the DOE press release on energy.gov (primary source). V11 traced to BLS OOH (primary source). V5 from Aug 7, 2026 press reporting of SME/SOMP data; the SME Guide to Minerals and Materials Science Schools is the primary annual series and should be obtained for the journal version. V6, V7, V10 are secondary reports of SME/AGI figures; trace to primary before journal submission.
A note on V9 that previews the central finding: PROSPECT's official goal is to double graduates — not enrollment — within two years. Students graduating in 2027–2028 are already enrolled today, so a four-year degree pipeline cannot double its output on that timeline under any funding level. The target is achievable only through near-completion students, certificates, and short-cycle programs — and DOE's own eligibility list (community colleges, trade schools, industry partners) reflects this. The government's program design implicitly concedes the arithmetic this notebook makes explicit.
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
plt.rcParams.update({'figure.dpi': 110, 'axes.grid': True, 'grid.alpha': 0.3,
'axes.spines.top': False, 'axes.spines.right': False})
# ---------------- VERIFIED constants (see table above) ----------------
V = dict(
retirees_by_2029 = 221_000, # V1: SME
degrees_2020 = 327, # V2: CSIS
grad_drop_2016_2020 = 0.39, # V3: CSIS
enroll_2023 = 600, # V5: <600; modeled at the bound (conservative)
enroll_2015 = 1_500, # V5: "~1,500 eight years earlier"
jobs_per_year_lo = 11_000, # V6: SME
jobs_per_year_hi = 13_000, # V6: SME
)
V['degrees_2016'] = round(V['degrees_2020'] / (1 - V['grad_drop_2016_2020'])) # ≈ 536
print(f"Implied 2016 graduations from V2 & V3: {V['degrees_2016']}")
Implied 2016 graduations from V2 & V3: 536
ASSUMPTIONS (flagged, with sensitivity ranges)¶
These are modeling choices, not verified facts. Each is varied in the sensitivity section.
| # | Assumption | Central value | Range | Basis |
|---|---|---|---|---|
| A1 | BS degree production lag (enrollment cohort → graduation) | 4 yr | 4–5 | Standard program length |
| A2 | Completion rate (entering cohort → degree) | 0.70 | 0.60–0.80 | Absorbed by calibration to the 2020 anchor (see below) |
| A3 | Share of graduates entering U.S. minerals industry | 0.80 | 0.65–0.90 | Placeholder; VERIFY (SME salary/placement surveys) |
| A4 | Engineer-tier annual demand | 400/yr | 300–500 | ANCHORED TO V11 (BLS OOH projected openings, 2024-2034); range covers demand growth above BLS's flat-growth projection |
| A5 | Grant-driven enrollment response | Doubling over 2 yrs (2027–2028), then flat | see scenarios | V9's stated policy goal, taken at face value |
| A6 | Accelerated pathway training lag (non-degree practitioner track) | 1.5 yr | 1–2 | Certificate/community-college/industry program norms |
| A7 | Enrollment path 2015→2023 | linear decline 1,500→600, flat after 2023 in baseline | — | Interpolation between V5 endpoints |
The structural split now rests on primary data: V11 fixes the degreed-engineer tier at roughly 400 openings per year, so the practitioner tier is V6 minus V11, roughly 10,600-12,600 per year, or about 97% of total annual need. This replaces the share assumption used in v0.1. A caveat for the journal version: BLS's ~400 covers the mining and geological engineer occupation under its own flat-growth projection; announced processing buildout implies demand above that baseline, which the 300-500 sensitivity range partially covers and a demand-growth module should treat properly.
A = dict(lag=4, completion=0.70, industry_entry=0.80,
eng_openings=400, # A4 anchored to V11 (BLS OOH)
accel_lag=1.5)
YEARS = np.arange(2015, 2041)
def enrollment_path(scenario):
'''Annual new-cohort enrollment (proxied by total enrollment level, A7).'''
e = np.zeros_like(YEARS, dtype=float)
for i, y in enumerate(YEARS):
if y <= 2023:
e[i] = np.interp(y, [2015, 2023], [V['enroll_2015'], V['enroll_2023']])
else:
e[i] = V['enroll_2023']
if scenario == 'S1_grant_doubling': # A5 / V9 at face value
for i, y in enumerate(YEARS):
if y == 2027: e[i] = V['enroll_2023'] * 1.5
elif y >= 2028: e[i] = V['enroll_2023'] * 2.0
elif scenario == 'S2_grant_tripling': # optimistic upper bound
for i, y in enumerate(YEARS):
if y == 2027: e[i] = V['enroll_2023'] * 2.0
elif y >= 2028: e[i] = V['enroll_2023'] * 3.0
return e
def graduates(enroll, lag=None, completion=None, calibrate=True):
'''G(t) = k * completion * cohort_share_of_enrollment(t - lag).
Enrollment level -> entering-cohort size via division by program length
(steady-state approximation), lagged, discounted by completion, and scaled
by a single calibration constant k chosen so the model reproduces the
VERIFIED 2020 graduation count (V2) exactly. k > 1 indicates the degree
series (V2) covers a somewhat broader population than the enrollment
series (V5) -- a commensurability gap documented as limitation L0.'''
lag = lag or A['lag']; completion = completion or A['completion']
cohort = enroll / A['lag']
g = np.zeros_like(cohort)
g[lag:] = cohort[:-lag] * completion
if calibrate:
i2020 = np.where(YEARS == 2020)[0][0]
k = V['degrees_2020'] / g[i2020]
g = g * k
return g
# Report the calibration constant once, transparently
_raw = graduates(enrollment_path('S0_baseline'), calibrate=False)
_i = np.where(YEARS == 2020)[0][0]
K_CAL = V['degrees_2020'] / _raw[_i]
print(f"Calibration constant k = {K_CAL:.3f} "
f"(raw model 2020 output {_raw[_i]:.0f} vs verified {V['degrees_2020']})")
for s in ['S0_baseline', 'S1_grant_doubling', 'S2_grant_tripling']:
g = graduates(enrollment_path(s))
i2020, i2030, i2035 = [np.where(YEARS == y)[0][0] for y in (2020, 2030, 2035)]
print(f"{s:20s} grads 2020={g[i2020]:5.0f} 2030={g[i2030]:5.0f} 2035={g[i2035]:5.0f}")
Calibration constant k = 1.347 (raw model 2020 output 243 vs verified 327) S0_baseline grads 2020= 327 2030= 141 2035= 141 S1_grant_doubling grads 2020= 327 2030= 141 2035= 283 S2_grant_tripling grads 2020= 327 2030= 141 2035= 424
Anchor check and calibration. The raw pipeline under-predicts the verified 2020 count (243 modeled vs. 327 verified), implying the CSIS degree series covers a somewhat broader population than the SME/SOMP enrollment series. Rather than silently tuning parameters, the model applies one transparent calibration constant (k ≈ 1.35) so 2020 output matches V2 exactly; the commensurability gap is logged as limitation L0 and must be resolved with a unified primary dataset (ASEE/IPEDS) for the journal version. All scenario comparisons are unaffected, since k scales every scenario identically.
2. The demand side¶
Total replacement + growth demand is V6: 11,000–13,000 jobs/yr. A4 splits this into the degreed-engineer tier and the practitioner tier. Demand is held constant across scenarios — deliberately conservative, since announced processing buildout (the $12B stockpile program, new domestic processing facilities, the Mines hub's commercialization mission) implies demand growth, not stasis.
demand = {}
for name, total in [('lo', V['jobs_per_year_lo']), ('hi', V['jobs_per_year_hi'])]:
demand[name] = dict(total=total,
engineer=A['eng_openings'], # V11: BLS OOH openings
practitioner=total - A['eng_openings'])
dm = pd.DataFrame(demand).T
dm.index.name = 'demand case'
print(dm.round(0).to_string())
total engineer practitioner demand case lo 11000 400 10600 hi 13000 400 12600
# ---------------- Figure 1: Engineer tier — supply vs. need ----------------
fig, ax = plt.subplots(figsize=(9, 5))
styles = {'S0_baseline': ('Baseline (no policy response)', '#7a7a7a', '-'),
'S1_grant_doubling': ('Grant response: enrollment doubles by 2028', '#1f5fa8', '-'),
'S2_grant_tripling': ('Optimistic: enrollment triples by 2028', '#1fa86b', '-')}
for s, (lbl, col, ls) in styles.items():
g = graduates(enrollment_path(s)) * A['industry_entry']
ax.plot(YEARS, g, ls, color=col, lw=2, label=lbl)
ax.axhspan(300, 500, color='#c0392b', alpha=0.15,
label='Annual engineer-tier openings (V11: BLS ~400/yr, band 300-500)')
ax.axhline(A['eng_openings'], color='#c0392b', lw=1.2, ls='--')
ax.axvline(2026.6, color='k', lw=0.8, ls=':')
ax.text(2026.7, ax.get_ylim()[1]*0.05, 'Aug 2026\ngrants', fontsize=8)
first_gain = 2027 + A['lag'] # first year a larger cohort can graduate
ax.axvline(first_gain, color='#1f5fa8', lw=0.8, ls=':')
ax.text(first_gain + 0.1, ax.get_ylim()[1]*0.75,
f'{first_gain}: earliest year\nexpanded cohorts\ncan graduate (A1)',
fontsize=8, color='#1f5fa8')
ax.set(title='Figure 1 — The lag is structural: degree supply vs. engineer-tier need',
xlabel='Year', ylabel='Graduates entering industry per year',
xlim=(2016, 2040))
ax.legend(fontsize=8, loc='upper left')
plt.tight_layout(); plt.show()
# ---------------- Figure 2: Cumulative engineer-tier gap, 2026-2035 ----------------
w = (YEARS >= 2026) & (YEARS <= 2035)
fig, ax = plt.subplots(figsize=(9, 5))
rows = []
for s, (lbl, col, _) in styles.items():
g = graduates(enrollment_path(s)) * A['industry_entry']
gap_lo = np.cumsum(np.maximum(300 - g[w], 0))
gap_hi = np.cumsum(np.maximum(500 - g[w], 0))
ax.fill_between(YEARS[w], gap_lo, gap_hi, color=col, alpha=0.18)
ax.plot(YEARS[w], (gap_lo + gap_hi) / 2, color=col, lw=2, label=lbl)
rows.append(dict(scenario=lbl, cum_gap_2035_lo=int(gap_lo[-1]),
cum_gap_2035_hi=int(gap_hi[-1])))
ax.set(title='Figure 2 — Cumulative unfilled engineer-tier openings vs. V11 demand band, 2026-2035',
xlabel='Year', ylabel='Cumulative shortfall (positions)')
ax.legend(fontsize=8, loc='upper left')
plt.tight_layout(); plt.show()
print(pd.DataFrame(rows).to_string(index=False))
scenario cum_gap_2035_lo cum_gap_2035_hi
Baseline (no policy response) 1847 3847
Grant response: enrollment doubles by 2028 1338 3338
Optimistic: enrollment triples by 2028 986 2829
3. The practitioner tier — where the real arithmetic lives¶
The engineer tier is the visible gap; the practitioner tier is the decisive one. Roughly 97% of the 11,000-13,000 annual need (V6 minus the V11 engineer tier) is technicians, operators, and skilled trades — a population university degree programs do not serve at all. The question for this tier is throughput: what annual training capacity, starting when, does closing it require?
# ---------------- Figure 3: Time-to-impact — degrees vs. accelerated pathways ----------------
fig, ax = plt.subplots(figsize=(9, 5))
years_fine = np.arange(2026, 2036, 0.25)
def output_start(start, lag):
return np.where(years_fine >= start + lag, 1.0, 0.0)
deg = output_start(2027, A['lag']) # cohorts expand 2027 -> first grads 2031
acc = output_start(2027, A['accel_lag']) # accelerated cohorts 2027 -> output mid-2028
ax.fill_between(years_fine, 0, acc, color='#1fa86b', alpha=0.25,
label=f"Accelerated pathways (A6: {A['accel_lag']}-yr lag) — producing from mid-2028")
ax.fill_between(years_fine, 0, deg, color='#1f5fa8', alpha=0.25,
label=f"Expanded degree programs (A1: {A['lag']}-yr lag) — producing from 2031")
ax.set(title='Figure 3 — Time-to-first-output: the 2028-2031 window belongs to accelerated training',
xlabel='Year', ylabel='Producing trained workers (indicator)',
yticks=[0, 1], yticklabels=['not yet', 'producing'])
ax.legend(fontsize=8, loc='center right')
plt.tight_layout(); plt.show()
# Required accelerated throughput to hold the practitioner tier level, 2028-2035
need = demand['lo']['practitioner'], demand['hi']['practitioner']
print(f"Practitioner-tier annual need (V6 minus V11): {need[0]:,.0f} - {need[1]:,.0f} per year")
print(f"=> Accelerated/vocational pathways must sustain roughly that annual throughput;")
print(f" degree expansion contributes ~0 additional workers to ANY tier before {2027 + A['lag']}.")
Practitioner-tier annual need (V6 minus V11): 10,600 - 12,600 per year => Accelerated/vocational pathways must sustain roughly that annual throughput; degree expansion contributes ~0 additional workers to ANY tier before 2031.
# ---------------- Sensitivity: conclusions under assumption ranges ----------------
sens = []
for lag in [4, 5]:
for comp in [0.60, 0.70, 0.80]:
for need in [300, 400, 500]:
g = graduates(enrollment_path('S1_grant_doubling'), lag=lag, completion=comp)
g = g * A['industry_entry']
w = (YEARS >= 2026) & (YEARS <= 2035)
gap = np.maximum(need - g[w], 0).sum()
sens.append(dict(lag=lag, completion=comp, eng_need_per_yr=need,
first_gain_year=2027 + lag,
cum_eng_gap_2026_2035=int(gap)))
sens = pd.DataFrame(sens)
print(sens.to_string(index=False))
print(f"\nCumulative engineer-tier gap (S1, 2026-2035) ranges "
f"{sens.cum_eng_gap_2026_2035.min():,} - {sens.cum_eng_gap_2026_2035.max():,} "
f"across all assumption combinations; first supply gain never precedes "
f"{sens.first_gain_year.min()}.")
print("Note: calibration to the 2020 anchor absorbs the completion-rate parameter "
"(identical gaps across A2 values above) -- supply level is pinned by verified "
"data, so results are driven by lag (A1) and demand share (A4), which is exactly "
"where the uncertainty belongs.")
lag completion eng_need_per_yr first_gain_year cum_eng_gap_2026_2035 4 0.6 300 2031 1338 4 0.6 400 2031 2338 4 0.6 500 2031 3338 4 0.7 300 2031 1338 4 0.7 400 2031 2338 4 0.7 500 2031 3338 4 0.8 300 2031 1338 4 0.8 400 2031 2338 4 0.8 500 2031 3338 5 0.6 300 2032 1528 5 0.6 400 2032 2528 5 0.6 500 2032 3528 5 0.7 300 2032 1528 5 0.7 400 2032 2528 5 0.7 500 2032 3528 5 0.8 300 2032 1528 5 0.8 400 2032 2528 5 0.8 500 2032 3528 Cumulative engineer-tier gap (S1, 2026-2035) ranges 1,338 - 3,528 across all assumption combinations; first supply gain never precedes 2031. Note: calibration to the 2020 anchor absorbs the completion-rate parameter (identical gaps across A2 values above) -- supply level is pinned by verified data, so results are driven by lag (A1) and demand share (A4), which is exactly where the uncertainty belongs.
4. Findings (v0.2)¶
- The lag is structural and no funding level removes it. Enrollment growth beginning in 2027 cannot increase graduate supply before 2031 (A1). This holds across every scenario and every assumption combination tested. The binding constraint on the degree pipeline is time, not money.
- Even doubled enrollment leaves the engineer tier short through 2035. Under the policy's own stated goal (V9), the cumulative engineer-tier shortfall from 2026-2035 runs between roughly 1,300 and 3,300 unfilled openings across the V11 demand band (300-500 per year), against BLS's central estimate of ~400 openings annually.
- ~97% of the need is outside the university pipeline entirely. The practitioner tier requires on the order of 10,600-12,600 trained workers per year (V6 minus V11) — a scale only accelerated, non-degree pathways (certificates, community colleges, industry upskilling, open training programs) can address, and the only channel that can produce output inside the 2028-2031 window.
- Policy implication. Degree-program investment is necessary for the long run but arithmetically cannot be the near-term answer. The near-term answer is training architecture: high-throughput, short-lag practitioner pathways feeding into the same progression. (This is the design thesis of Project FORGE, and finding 3 is why.)
5. Limitations and the road to the paper¶
- L0: the enrollment (V5) and degree (V2) series are not perfectly commensurable; the k ≈ 1.35 calibration must be replaced by a unified ASEE/IPEDS dataset.
- A4 is now anchored to BLS OOH projected openings (V11); the journal version should add a demand-growth module reflecting announced processing buildout above BLS's flat baseline, and reconcile occupation-code coverage with the degree series.
- Demand is held flat; a demand-growth module tied to announced processing capacity (stockpile program, new facilities) would strengthen the result, not weaken it.
- Enrollment response to grants (A5) takes the policy goal at face value; historical elasticity of enrollment to funding shocks should be estimated for the journal version.
- Completion and placement rates (A2, A3) need primary sources (ASEE, SME surveys).
- The cohort model is deterministic; the paper version should add stochastic completion and a proper age-structured retirement module calibrated to V1.
Publication path: notebook (this) → Payne Institute commentary → journal manuscript (Resources Policy / Mineral Economics), targeting co-authorship with faculty in Mineral & Energy Economics and Operations Research.