Afrimintel  ›  Methodology  ›  Safeguards layer — gate-and-floors

Safeguards layer — gate-and-floors architecture

Methodology sub-page (Component E) · Editorial responsibility: Nikesh Patel · Every figure is Sourced Derived or Absent — no fourth state, nothing imputed.

§A — Architecture at a glance

Six moving parts, in priority order:

1. Critical-requirement gates (G1–G5). Context-relevant must-clears. A relevant gate that is not met flags the asset regardless of any component score.
2. Foundational-management cap (PS1). The management-system confidence score caps every dependent component — a weak ESMS pulls the whole layer down rather than being averaged against.
3. Per-pillar floors. Social, Environmental, and Governance/Management are scored independently; each must clear its own floor. A failing pillar is not averaged away by a strong one.
4. Tiered components (ordinal). Four tiers reflect where the evidence says financial loss actually originates — not an invented weighting.
5. Context-relevance + Absent. Components and gates render N/A where inapplicable and Absent (flagged) where evidence is missing — never imputed, never defaulted to a mid-value.
6. Glass-box. Every gate result and component score is shown with its basis. An optional headline figure (§L) never renders without this decomposition beneath it.

§B — Evidence base (the published derivation)

The architecture is informed by five convergent authoritative sources. It is an Editorial construction — published with its reasoning and its sources, not a black box. Sourced / Derived / Absent label individual figures and gate states; they do not apply to this design choice itself.

§C — Critical-requirement gates (G1–G5)

Each gate renders one of: Met Not met N/A Absent. N/A where the trigger condition is absent; Absent (flagged, never imputed) where the trigger is present but the evidence is missing.

Lens toggle (§J): IFC Performance Standards is the default universal lens for private / blended-finance decisions. Switch to the World Bank Environmental & Social Framework for public-sector / sovereign decisions; the two are aligned, and the toggle re-bases which framework the gates, pillars and tiers reference. A segment-aware default (DFI / public → ESF; investor → IFC PS) is a fast-follow off the access-request role.

§D — Foundational-management cap (PS1ESS1)

The management-system confidence score caps every dependent component score: a component score cannot exceed the management-system score. This is the IRMA grievance-cap analogue, and the IFC CAO data shows PS1 management-system issues are pervasive. A weak or undemonstrated ESMS pulls the whole layer down rather than being averaged against a strong component elsewhere.

§E — Three pillars, each with a floor (no cross-pillar averaging)

Three pillars are scored independently — Social, Environmental, Governance / Management — and each must clear its own floor. A failing pillar is not averaged away by a strong one. Each pillar score is Derived, carries a confidence tier, and draws only on Sourced or Derived inputs; a pillar with material Absent inputs renders its score Absent (flagged), not a mid-value.

§F — Tiered components (ordinal, high-confidence ranking)

Tier reflects how often a component binds, not its importance. A severity-gated component (Indigenous / FPIC, biodiversity / critical habitat) is dormant where it does not apply but dominant where it does: a failed FPIC gate flags the asset regardless of any other score, exactly as a failed Tier-1 gate would. Tier 3 is a statement about frequency of bind, never about the weight of Indigenous consent.

§G — E1: IBAT-derived biodiversity input

IBAT (Integrated Biodiversity Assessment Tool) is a recognised authoritative biodiversity-screening tool used at the investment / acquisition stage; its outputs are quantifiable and map to IFC PS6WB ESS6. IBAT feeds the biodiversity / critical-habitat component (Tier 3) and triggers the critical-habitat gate G5 where the asset is in or near critical habitat. It is an input to the layer, not a standalone widget. The IBAT → value mapping is documented; a missing IBAT input renders the component Absent.

§H — E5: AfDB OS7, the Africa-specific divergence

Within the Tier-2 community component, Afrimintel surfaces AfDB Operational Safeguard 7 (vulnerable groups) as the explicit Africa-specific divergence — OS7 is broader than the IFC equivalent. It is framed as a divergence the score maps to, with a Sourced definition — not as an endorsement, and not as a partnership.

§I — E4: the framework lens

The IFC Performance StandardsWorld Bank Environmental & Social Framework lens is selectable at the top of §C. IFC PS is the default (the universal private / blended-finance reference and the larger audience); WB ESF is the public-sector / sovereign reference. The two are aligned. The toggle re-bases which framework underlies the gates, the pillars and the tiers. Both lenses are documented; a segment-aware default off the access-request role is a flagged fast-follow, not a blocker.

§J — Context-relevance and Absent

Components and gates render N/A where the trigger condition does not apply, and Absent (not-demonstrated / flagged) where the trigger applies but the evidence is missing. Absent is never imputed and never defaulted to a mid-value. This is the difference between "this asset has no tailings facility, so GISTM does not apply" (N/A) and "this asset has tailings, but its GISTM conformance is not on the record" (Absent — flagged).

§K — Glass-box

Every gate result and every component score is shown with its basis and its state. No single number stands alone, and no black-box figure appears on any surface. The headline figure below (§L) is optional, secondary, and structurally subordinate to this decomposition.

§L — Optional headline roll-up (secondary, gate-governed, glass-box)

For institutions that want a single number, an optional headline figure may sit on top of the decomposition. It is governed by the gates: a relevant gate failure caps or flags the headline so it can never mislead, and it never renders without the full decomposition beneath it. The aggregation weights are a Derived, configurable, secondary roll-up — not load-bearing; the gates, cap and floors do the work. The tier-band defaults are externalised to config/safeguards-rollup-weights.json and rendered as bands, never as fixed truth:

Weights are configurable defaults pending advisory and editorial confirmation — non-blocking, because the roll-up is not the score.

§M — Worked example: Kabanga (pre-FID)

Kabanga is a development-stage Tanzanian nickel project. This example shows the architecture doing both things it must: refusing to fabricate where the record is thin, and producing a composite once the safeguards record is sourced. Sourcing the DFC Initial Project Summary (2025), the Lifezone Feasibility Study Technical Report Summary (Jul 2025) and the SLR ESIA (2025) resolved the three gates that were previously Absent.

Read this the right way. Kabanga is not "low-scoring." It is partially demonstrated: strong and Sourced on the Tier-1 land/resettlement gate, undemonstrated on tailings governance and critical-habitat screening. The gate-and-floors architecture surfaces exactly that, and declines to average it into a misleadingly clean number. Filling G4/G5 (a GISTM conformance source and an IBAT screen) is what moves this from flagged to scored — through audit-logged updates, not estimation.