PhD Research · Critical Minerals Supply Chain Intelligence

Critical Minerals
Causal Engine

Pearl's Ladder of Causation (L1/L2/L3) over an ODE commodity model, validated against CEPII BACI bilateral trade data. Identify supply shocks, trace trade circumvention, and answer counterfactual policy questions across eight critical minerals — graphite, rare earths, lithium, cobalt, nickel, uranium, germanium, and gallium.

The pipeline · Documents → KG → ODE → Counterfactual

Step 1

Documents → Knowledge Graph

HippoRAG retrieves USGS / IEA / CEPII passages. Claude extracts (subject, relation, object) triples. 472 entities, 588 edges.

Step 2

KG → ODE shocks → Trajectory

Year-specific shares feed the {K, I, P} ODE. Pearl L2 do-calculus on substitution and fringe supply nodes. Annual price index forward.

factualno-shock

Step 3

Pearl L3 — what would have been

Abduction-Action-Prediction. Recover noise from the realised trajectory, intervene, replay. Quantifies the causal price premium of a policy.

Model validation — CEPII BACI price series

91.7%

In-sample DA

10 episodes

65.7%

Clean OOS DA

5 transfer pairs

+33.4pp

vs Momentum

directional accuracy

+23.6pp

vs Conc. Heuristic

directional accuracy

Directional accuracy (DA) on price index year-on-year moves · OOS = parameters transferred across episodes without re-fitting · Baselines receive no shock information

Tools

Causal framework — Pearl's Ladder

L1

Association

Seeing

P(Y | X=x). Correlations from BACI bilateral flows — no causal claim.

L2

Intervention

Doing

P(Y | do(X=x)). Graph surgery on the SCM + calibrated parameters (η_D, α_P, τ_K).

L3

Counterfactual

Imagining

P(Y_x | X=x′). Abduction-action-prediction with twin-network SDE noise replay.

Data sources

CEPII BACI

1995–2024 · pre-reconciled bilateral flows · ISO 3166-1 normalized

USGS MCS

2020–2024 · 5 annual volumes · 17 commodities · reserves + production

IEA 2021

Critical minerals supply chain risk · 20 chunks

DRC Reports

2 artisanal cobalt reports · 79 chunks