Bristol Airport has selected HyFlux to build BRSupercool under its Airport Carbon Transition Programme, alongside projects from Seamach Energy and Equilibrion. This is what we are building, and why we think it is the missing piece.
Two different problems wearing the same name
Almost everything written about hydrogen aviation is about the aircraft. Tank gravimetric efficiency, fuel-cell power density, whether you burn it or run it through a stack, how much the airframe grows to carry it. That work matters and there is a great deal of it.
Then there is the other problem, which gets a fraction of the attention: what has to happen on the ground, every day, for that aircraft to depart on time.
How long does the turnaround actually take when you have to chill a coupling down, purge it, move several tonnes of cryogenic liquid, purge it again and disconnect — and can you still board passengers while that is happening? How much liquid hydrogen does one departure need, and where does it come from? What does the airport have to build, and how much of its land does that take? And at the end of it, what number does the airline put in its emissions report, and can anyone check it?
These are not aircraft questions. They are airport questions, and they decide whether hydrogen aviation is operable or merely possible.
Why it has not been modelled
Not because it is uninteresting, but because it falls between disciplines. The turnaround is an operations-research problem. The cold chain is a thermodynamics and logistics problem. The emissions accounting is a carbon-reporting problem with its own standards and boundaries. Each has a literature; they are rarely joined up.
And they interact. A delayed aircraft on stand is not just a schedule problem — it is hydrogen boiling off, which is energy to re-liquefy and mass that has to be replaced. A faster fill rate shortens the turn right up until it does not, because at some point the fuelling stops being the slowest thing on the stand and you are simply buying capacity you cannot use. You cannot see either effect by modelling the pieces separately.
What BRSupercool is
A digital twin of the chain from the aircraft fuelling coupling outwards: production and compression, road haul, liquefaction, storage and boil-off, transfer to the stand, and the turnaround itself. It couples a discrete-event simulation of the turnaround to an energy and emissions balance of the cold chain, and reports per-departure Scope 3 factors against a Jet A1 baseline across five hydrogen sourcing pathways.
Aircraft performance inboard of the coupling is deliberately outside the boundary. We do not model mass, aerodynamics, fuel cells, inverters or batteries — and rather than approximating them, the model refuses.
Why "we don't know" is a feature
This is the part we would most like other people to copy.
Every consequential number that leaves the model carries a status saying where it came from, and that status is never quietly upgraded:
- Verified — checked against an external authority: NIST REFPROP, the EU Joint Research Centre.
- Derived — computed by a tested model from documented inputs.
- Assumption — a stated input with no model or test behind it, labelled as such.
- Missing — we have no value, and will not invent one.
That last category is the important one. Ask our model for something it cannot compute and it returns an insufficient evidence response listing exactly which inputs are absent — not a plausible-looking figure. Twenty-six aircraft-level quantities are returned that way, each with a stated reason.
It would be easy to fill those in. Every one of them has a defensible-sounding number somewhere in the literature. But an airport is going to quote these figures to an airline and to a regulator, and a number nobody can trace is worse than no number at all — because it looks like knowledge.
The test we set ourselves
If a reviewer picks any figure in the model at random, can we say where it came from, which test protects it, and what would have to change for it to be wrong? If not, it does not ship as a result. Roughly 165 automated tests exist to make that answer stay true — including ones that fail deliberately if a headline figure moves without someone intending it.
Where the surprises have been
We expected the hard part to be the aircraft interface. It has not been.
The interesting constraints have turned out to sit further back in the chain — in the equipment that refills the refueller, in the minutes of a fuelling operation that have nothing to do with how fast hydrogen flows, and in the difference between where you liquefy and where you burn. More than once the model has produced an answer that inverted the assumption we started with, and on at least one occasion it caught a figure of our own that was wrong.
We publish those corrections. A model whose value rests on traceability has to report its own errors the same way it reports its results, or the traceability is decorative.
What happens next
The model will be independently reviewed by our validation partner PXL-ICE before the final report in November 2026. We have asked them to try to break it rather than to confirm it, and we have given them a written list of the places we already believe it is weakest. A validation that finds nothing would be a failed validation.
Bristol Airport keeps the tools at the end: the turnaround simulator, the cold-chain model and the Scope 3 factor pack, as documented Python packages, so the ACT team and successor projects can re-run the scenarios without us.
The working is public
The dashboard, the Avonmouth-to-stand corridor map, a full-day operational twin, the systems model, the live workplan and the complete source register — every source recording what we took from it. Challenge any of it.
