A site is a hypothesis. The screen decides whether it survives.
Every project in the portfolio runs through the same four-step screen, from a coordinate on a map to a capital structure a lender can underwrite. The method is what makes the portfolio comparable, and what lets a screened-out site sit honestly next to the flagship.
Why the portfolio is ammonia-first.
The earlier approach led with methanol, which made compliant carbon a structural input. It was coherent, but it rested on one assumption: that large-scale, affordable, certified carbon would remain available across the financing life. Remove that assumption and the project is no longer an energy venture but a carbon-procurement one, exposed to a feedstock it cannot control.
Build on the abundant input rather than the scarce one. Renewable electricity, water and air scale freely; certified carbon does not, and ammonia requires none of it. A methanol slice is added only where a genuine local biogenic carbon source already exists, which is Dakhla, and on industrial CO₂, Ain Sokhna. Everywhere else the product is pure ammonia.
This shifts the dependency rather than removing it. Ammonia eliminates the carbon-supply exposure entirely; the methanol slice still depends on the biogenic-carbon premium and the RFNBO rules that define it holding across the financing life, which is a regulatory assumption rather than a settled fact. It is carried as a named, stress-tested risk, not a foundation, which is why a single site relies on it and why that site quantifies the loss of it directly.
Four gates in series. A site clears all four or it does not advance.
A coordinate enters on the left as a resource claim. Each gate takes the prior output, tests it, and either passes a tightened figure forward or rejects the site to a documented status. Nothing skips a gate. The verdict on the right is not a score; it is which of three financing realities the site lands in.
Generation is not assumed, it is derived. We fix the product output target first, then work backward through hydrogen stoichiometry to the electrolyser load and the wind and solar capacity that feeds it. When generation and output ever disagree, the output target is the anchor and generation is resized to match.
One model runs every site, so the assumptions are identical across the portfolio and a number from Nouadhibou and a number from Lüderitz can be compared directly. The model is internally consistent, not yet field-validated: no site is built, so every output is a screening figure to be confirmed by measurement and FEED, not a measured result.
A capacity factor is only as good as its evidence. We grade every resource input by provenance on a four-rung scale and treat measured data as the only basis for advancing a site. A figure that arrives one rung down is carried with a haircut; a figure that arrives at the bottom rung is held at arm's length. This is the step that separates Duqm, where station data undercut the quoted wind, from Tarfaya, where a decade of adjacent operating data backs it.
Capital is built from unit costs, component by component, not scaled from a reference plant. Generation, electrolysis, synthesis, balance of plant, owner's cost and contingency each carry their own line. The carbon chain, when a site has one, is costed explicitly rather than buried, because a green fuel model that hides its carbon logistics hides its real risk.
Every figure is a screening estimate at roughly ±25 percent, stated as such. The honest grade matters more than a false precision.
The binding constraint on these projects is capital structure, not operations. Margins are healthy; a first-of-kind, capital-heavy build simply cannot be serviced from its EBITDA at ordinary commercial gearing. The single figure that orders the portfolio is the EBITDA-to-CAPEX ratio, not the headline margin. It measures how much annual cash each dollar of capital throws off, which is what determines whether debt can be serviced at all. Dakhla leads at 2.20% on its premium carbon slice; Lüderitz follows at 1.77% on pure wind and solar. The ratio orders the sites; it does not decide them. A higher ratio buys a site a thinner grant and a lighter stack, but whether it clears is settled one level down, at the coverage test.
Under that ratio sits the test itself. Each site is run against the 1.20x debt service coverage floor, and we solve for the lowest-equity stack that holds the floor across every serviced year. That stack is development-stage by design: equity-heavy, blended with development-finance and construction grants. Where even that does not clear at base price, the site needs an offtake price floor of the H2Global or EU Hydrogen Bank type, which converts an open commodity price into contracted revenue. Dakhla, Lüderitz and Ain Sokhna clear without one. Stephenville, Nouadhibou and Tarfaya need it. That distinction is the whole portfolio in one line.
The method is the product. The sites are its output.
See how it resolves on the flagship, where all four steps are worked end to end, or read the full portfolio with every site's grade and gap stated plainly.