How H2Hub verifies hydrogen intelligence
We do not resell public databases. Every figure in the H2Hub verified layer has been cross-referenced against at least two independent primary sources and reviewed by ReneEnergy technical staff as part of H2Hub's internal data-validation process. That review is a data-quality check — it is not a licensed professional-engineering certification. This page documents exactly how the 2,934 → 160 verification funnel works, who does the work, and what we refuse to publish. Licence terms for each source are listed in the data licensing register.
The verification funnel
Global Hydrogen Review 2026 project database (announced through decommissioned).
Aliased operator names, MW/MWel unit reconciliation, geocoding, product classification.
European Hydrogen Observatory (July 2026) primary + operator-issued FID / commissioning notices.
Two-source verification of capacity, technology, offtake, timeline; edge cases resolved by manual engineering review.
Process
- 1Ingest
IEA GER 2026 project database and EHO July 2026 export are loaded as immutable snapshots.
- 2Normalise
Units reconciled (MW vs. MWel vs. t/y H₂), aliases collapsed, geocoding validated.
- 3Cross-reference
Each candidate is matched against ≥ 2 independent primary sources: operator press release, FID filing, EHO, or regulator disclosure.
- 4Technical review
ReneEnergy technical staff manually resolve conflicts on capacity, technology, and status as part of H2Hub's internal data-validation process. This review is not a licensed professional-engineering certification.
- 5Publish
Only fields where two independent sources agree are surfaced in the verified layer; disagreements are logged in the reconciliation dashboard.
What we refuse to publish
- Single-sourced capacity claims (press-release-only, no FID or regulator filing).
- LCOH figures without a documented CAPEX / electricity / capacity-factor stack.
- "Announced" projects older than 24 months with no construction milestone.
- Company-supplied numbers that contradict regulator filings without a written reconciliation.
Who does the work
H2Hub is built and maintained by ReneEnergy — a systems-first hydrogen engineering practice. We cross-reference published figures against the primary sources listed in the dataset lineage and methodology above, and we apply the resolution rules documented in the verification funnel when sources conflict.
Refresh cadence
Model card: AI Advancement Classifier (FID-probability framing, withdrawn)
The homepage previously reported an "FID probability" derived from a binary logistic regression. Our own corpus audit invalidated the framing, so the probability output has been retired. The stage-distribution benchmark that replaced it is documented on the homepage; this card records what the retired model actually was.
- Model
- Multinomial logistic regression, reduced to binary Advanced vs. Stalled.
- Features
- Technology, product, region only. No capacity, CAPEX, power price, offtake or developer signals (all 0% coverage in the source).
- Label
- Current status from the IEA snapshot: Operational / DEMO / FID-Construction ⇒ Advanced; all else ⇒ Stalled. No status transitions were observed.
- Validation
- 5-fold cross-sectional CV on the same snapshot: 73.7% accuracy, ROC-AUC ≈ 0.74. No held-out temporal test set.
- Left-censoring
- The observable ~4% cancellation share is severely left-censored — most cancelled projects leave the IEA tracker before appearing here.
- Why retired
- A cross-sectional stage distribution is not a hazard rate. Reporting it as an FID probability implied a forecast the data cannot support.
- Reinstatement criteria
- Requires retrospective IEA vintages (2023–2026) with observed status transitions before any probability output is reintroduced.