Delphidata Launches Project Viability Score for Low Carbon Molecule Projects

The low carbon molecule sector has seen a rapid expansion of announced projects over the past decade. However, historical data consistently shows that a large share of early-stage announcements do not progress to construction or commercial operation. For investors, offtake partners, technology suppliers, and policymakers, distinguishing credible projects from speculative announcements represents a significant analytical challenge.

The Project Viability Score was developed to address this gap with a systematic, data-driven approach. Rather than relying on qualitative assessments or static checklists, the model draws on the full historical record of projects that have either reached operation or been cancelled, and applies the learned patterns to the current active pipeline.

The score is the output of a statistically calibrated ensemble of five group-specific logistic regression models, each tuned to the structural characteristics of a distinct technology family: electrolysis-based green hydrogen, blue and specialty hydrogen routes, carbon capture projects, biomass, biofuels and sustainable aviation fuel (SAF), and infrastructure assets. This group structure reflects the fact that an alkaline electrolyser project and a direct-air-capture plant are evaluated by fundamentally different development economics and risk profiles.

Six features drive each model: project capacity (technology-specific), EPC contract commitment, offtake agreements, grant and subsidy linkage, registered FID date, and geographic attributes. Each feature receives a data-driven weight estimated by maximum-likelihood logistic regression. To address data sparsity in smaller technology categories, the model applies adaptive regularisation: L1 (LASSO) for data-poor groups and L2 (Ridge) for groups with sufficient observed outcomes.

Raw probabilities are post-processed using Platt calibration to ensure that predicted values are interpretable as empirical frequencies rather than raw model scores. A probability ceiling of 0.9 is applied uniformly across all groups, reflecting the epistemic reality that no announced project in an early-stage industry can be rated at absolute certainty of completion.

The calibrated probability is translated into a five-level score for dashboard use, ranging from Score 1 (below 0.20, very low likelihood) to Score 5 (above 0.80, very high likelihood). Score thresholds are fixed and do not adapt to the portfolio distribution, ensuring that a Score 4 project today is directly comparable to a Score 4 project from a prior scoring cycle.

The model is trained on the Delphi Low Carbon Hydrogen and Low Carbon Fuels Project Database, a proprietary dataset maintained by Delphidata. As of April 2026, the database contains approximately 9,000 hydrogen and derivative project records worldwide, including green and blue hydrogen, sustainable aviation fuel (SAF), biofuels, biomethane, and carbon capture projects. Records include structured metadata on project status, technology category, capacity metrics, contractual partners, financial milestones, and geographic attributes. Cancelled projects are explicitly flagged and retained in the training set rather than removed, ensuring that the model learns from negative outcomes as well as positive ones.

The model is retrained continuously as the database is updated. As more projects either reach commercial operation or are formally cancelled, the training set grows and the model's calibration improves. This self-reinforcing design means that accuracy is expected to increase systematically as the low carbon hydrogen industry matures and observed outcomes accumulate.

In the scoring cycle preceding April 2026, the ATOME Villeta Green Fertiliser Project in Paraguay received a predicted probability of 0.80, placing it at rank 17 out of 2,650 projects globally and as the highest-scoring project across all of Latin America. At the time of scoring, the project held FEED status.

In April 2026, ATOME publicly announced that the project had reached Final Investment Decision (FID). This outcome is classified in the methodology as an exogenous, ex-post validation event: it occurred independently of the model, after the score was assigned, and therefore constitutes a genuine out-of-sample test of the model's discriminative ability.

The methodology documentation notes that a single case is illustrative rather than conclusive. Systematic model validation relies on aggregate diagnostics including ROC-AUC, Brier scores, and Events-Per-Variable ratios computed across the full training set for each technology group.

The Project Viability Score is available as a free beta for all subscribers of the Delphi platform. Users can filter the ranked pipeline by region, technology, project stage, and capacity, and export results for further analysis. The full methodology documentation (Version 5.1, April 2026) is available upon request.

Delphi Data Labs GmbH

Meiereistrasse 14
1020 Vienna
Austria

https://www.delphidata.com/

Helene Arnhof
Climate Tech Analyst
helene.arnhof@delphidata.com

Delphidata is a Vienna-based market intelligence company specializing in the global decarbonization and critical infrastructure sectors. Founded to address the growing need for structured, asset-level data in the energy transition, Delphidata serves project developers, investors, strategy teams, EPCs, and technology providers navigating the hydrogen, CCUS, ammonia, methanol, and bio-/synfuel markets.

At the core of Delphidata's offering is 'Delphi', an interactive intelligence platform tracking 8,000+ decarbonization projects across five sectors, continuously updated from 150,000+ sources. The platform combines interactive global infrastructure maps, a real-time FID Tracker, competitor benchmarking, investment tracking, and dynamic forecasts, replacing static quarterly reports with live market intelligence.

Beyond this dashboard, Delphidata operates a suite of complementary products:
- Serapis, a real-time enterprise API delivering structured, asset-level data for analytics and AI workflows
- Theia, an AI-curated newsfeed monitoring market signals across 30+ critical infrastructure sectors
- Aletheia, a grounded AI chat retrieving verified insights from connected market data with full traceability.

Delphidata publishes regular thought leadership content including quarterly market reviews, sector-specific strategy reports, and the Strategosphere report series, trusted references for energy transition professionals worldwide.

This release was published on openPR.