Insurance Substrate — Methodology

How existing Africa mining substrate is surfaced against classes of cover — and where it stops.

Layered application, not a new dataset. The insurance view re-surfaces the same province scores, deposit intelligence, country composite, ACLED admin1 substrate, and ownership records used across the platform. No insurance-specific data is collected or modelled.

Editorial responsibility: Nikesh Patel · Platform version v2.4.3

What this methodology does not do. It does not define a rating, a premium, a PML/EML, a cession allocation, or a binding term. It defines how the platform surfaces inputs an underwriter or reinsurer pulls into their own file. Every screening signal below is ordinal triage, disclosed in full, and explicitly not a cardinal underwriting rating.

1 — Layered application

Afrimintel's substrate already answers questions an underwriting team asks at the screening stage: where is the asset, what is the security environment around it, how stable is the host country's policy and fiscal regime, who owns it, and what stage is it at. The insurance view does not generate new answers — it re-orients the existing answers around classes of cover, and draws the boundary at the point where a decision requires actuarial apparatus the platform does not have.

2 — The screening logic, disclosed

Conflict-event density (Sourced input + disclosed binning)

The conflict-density signal is the platform's ACLED admin1 substrate: the count of recorded conflict events in the deposit's first sub-national administrative unit over a trailing twelve months. The count itself is Sourced. The platform then bins that count into an ordinal flag for triage:

low  < 10 events  ·  moderate  10–99  ·  elevated  100–299  ·  high  300+

The binning is an editorial triage convention, stated here in full so it can be reproduced or rejected. It is not an insurance rating and carries no implied loss frequency, severity, or premium relativity. Where an admin1 unit is not yet in the ACLED surface, the signal is marked Pending and no value is asserted.

Two limits stated plainly. (1) Not cross-country comparable. These are raw event counts in administrative units of very different size, population, and ACLED coverage intensity. 302 in one large region and 34 in a smaller one are not on the same denominator; the ordinal flag is a within-file triage aid, not a cross-asset ranking. An underwriter normalising for exposure should treat the raw count, not the flag, as the input. (2) The counts have a vintage. They are a trailing-12-month window as captured on the platform’s ACLED admin1 substrate (mid-May 2026 capture); the window moves, and the figure should be refreshed against current ACLED before use in a live file.

Country composite (ordinal reference)

The country composite — the same governance, Fraser policy-perception, and fiscal-terms blend used in the DFI deal evaluator — is surfaced as relative country screening context. It is referenced ordinally (more favourable / less favourable relative to the platform's country set), not as a cardinal political-risk rating. A calibrated cardinal mapping to named risk tiers is deliberately not published here: doing so without an actuarial loss basis would imply precision the substrate does not carry. That calibration is deferred to a future published increment and flagged as such rather than invented.

Per-line relevance mapping

Each class of cover is mapped to the substrate inputs that bear on it. The mapping is editorial and additive — it tells an underwriter which existing platform signals are relevant to a given file, not what weight to give them.

Class of coverSubstrate inputs surfaced
Property + Business InterruptionDeposit siting, province context, ACLED admin1 density, operator capex disclosures
Political Risk (PRI / CEND)Country composite, sovereign/SOE ownership context, tenure-dispute context
Construction / Erection All RisksProject stage, FID status, infrastructure-corridor exposure, capex profile
Plant, Vehicle & TransitCorridor / logistics exposure, ACLED density along routes
Trade CreditOfftake / counterparty context, sovereign stake, country fiscal band

3 — What is Absent, and why

The boundary is the methodology. PML/EML, cession or program-weight allocation, premium and rate-on-line indications, catastrophe loss curves (including Lloyd’s RDS-format output), insurance loss / claims-history data, Solvency II / IFRS 17 reporting cuts, and physical-climate / transition-scenario exposure modelling are not produced — not because they are out of scope to build later, but because producing them from this substrate would require converting inputs the platform holds (a capex split, a conflict count, an ownership stake) into actuarial outputs they are not. A capex split is not a program-weight allocation; a conflict count is not a loss frequency. The platform marks these Absent so that the underwriting file built on top of it does not silently inherit invented figures. The substrate is the floor an underwriter stands on, not the decision they make from it.

A note on climate and model-input fit

Two boundary points worth stating directly. First, climate. The platform does not yet hold a deployed physical-climate-exposure or transition-scenario (IEA STEPS/APS/NZE) layer for the insurance view; environmental substrate is presently a single-asset SAVi-aligned pilot. Climate-risk and PRA-CBES-type applications are therefore a named future build, triggered by a customer requirement and the data to ground it — not something the substrate supports today. Second, model-input validation. The platform’s Sourced / Derived / Absent provenance standard is, by construction, the auditable-lineage form a model-input or data-validation review requires; it is a natural fit for that work even though the platform performs no validation itself.

4 — Reading a Surface 3 snapshot

On each worked dossier, the Insurance Substrate Snapshot shows the Sourced conflict-density input and its ordinal flag, the relevant classes of cover, a substrate note, and — prominently — the Absent block. The snapshot is a pointer into the fuller substrate (the dossier and the deal evaluator above it), not a standalone underwriting brief.