At your own gas, electricity and carbon prices, what does a kilogram of turquoise hydrogen cost, and does it still pass the climate gate?

What this model cannot tell you

• It computes a levelised cost, not a profit and loss. The corpus's operating profit, payback and return figures are carried as published context and are not recomputed here, because the model has no hydrogen sale price in it.
• Each price lever moves the cost along a straight line whose slope is the published single-lever sensitivity. The published scenarios move several levers at once, and the engine does not reproduce them: it reads the published optimistic corner 0.19 a kilogram cheaper and the published stress corner 0.88 dearer. The sign changes between corners, so no single correction fixes it and none is applied. The joint-move index tells the reader when they have left the one-lever-at-a-time regime the slopes were measured in, and the published band, not the engine, supplies the comparison band.
• The base case is a first-of-a-kind plant discounted at a rate near what a fifth plant would earn and run at an availability a mature plant would achieve. The corpus says both, the model lets the reader move both, and it does not correct either by default because the report's case is the report's case.
• The energy figures for the five reactor families are recorded by the corpus as energy input without splitting electricity from heat, so the model does not use them for its electricity parameter and does not compare them with the electricity-only figures. For the same reason it draws no link between a route's readiness and that route's energy or emissions, even though the two clearly move together in reality.
• The electricity the plant draws moves what a kilogram costs, through the product the corpus prints in money rather than through a rebuilt electricity bill. The corpus publishes this plant's cost as a base case plus single-lever deltas, not as a line-by-line stack that could be re-added from first principles, so the intensity term is written as the difference between what the plant pays for its electricity at the reader's intensity and what it pays at the report's stated one. Two things follow and both are printed rather than hidden. The corpus implies three different intensities for the same plant, 12 stated, 12.5 from the sensitivity as this model ranges it and 13.07 from the operating table, a spread of 8.9 per cent; and the sensitivity route needs a price range the corpus does not print, unlike its gas line, so the model uses it as a comparison and not as the calibration. The difference between the two bases is under two cents a kilogram anywhere inside the evidence range and is shown at every setting.
• The electrolyser's electricity consumption is a comparator input and does not reach this plant's cost, because it is not this plant's electricity. It moves the saving fraction, the total-energy inversion and the volatility comparison, which are the report's own energy argument, and nothing else. It does not move the green comparator price either: the corpus prices green hydrogen as a traded index today and a projected band for 2030, not as an electrolyser cost stack, so building one in order to connect the two would be our composition rather than the corpus's.
• The two settings that decide the climate rail load at the report's own route rather than at the average for the class: a best-documented pipeline supply chain rather than the global average leak rate, and an electrified catalytic plant on renewable power rather than the literature mean for methane pyrolysis of any kind. Both are the report's own case and both are named. At the corpus's central figure for either one, the cost is unchanged and the verdict is no longer holds.
• The bottom of the sector carbon volume range is a floor on a different basis from the rest of it. Fourteen thousand tonnes a year is the observed output of the only plant at commercial scale, and the corpus publishes no measured total for the sector as it stands today; its earliest published sector figure is the 2027 projection of 30 to 50 thousand tonnes that supplies the default. Read the bottom of that range as a present-day floor, not as a published estimate of what the sector supplies.
• The report dropped a steam-reforming total-energy comparison because its source mixed bases inside itself. This model does the same: it compares total energy against electrolysis only, and never against steam reforming.
• The carbon-market rail models a price ceiling from one published product-mix curve of three points and a volume gate from one published segment cracking point applied to one published tier share. Three points do not distinguish that curve's shape from several others that would fit as well, and the tier share is held constant as volume grows even though the corpus itself shows the mix shifting toward commodity grades at scale. The model therefore understates how fast the achievable price falls at large volume, and it says so rather than adding an unsourced correction.
• The qualification clocks are not in the cost at all. The corpus records 12 to 24 months for a tyre grade and 18 to 24 for a battery grade, and the adjudicated model books the carbon revenue from day one. That gap is real, it is the report's own counterweight, and this model does not close it: it lets the reader set a lower share of carbon sold, which is a blunt instrument for a timing problem. The evidence on that clock, 12 to 24 months for a tyre grade and 18 to 24 for a battery grade, is carried on the scenario and the hypothesis that test what happens when the powder does not sell, and not on the share parameter itself, because it supports none of that parameter's three numbers.
• The carbon-removal credit lane is not modelled. The corpus prices durable removals and states the eligibility threshold but publishes no per-kilogram credit figure for this route, and records that a tonne can be sold as material or as removal but never both. A credit revenue line would be our composition rather than the corpus's.
• Leakage is applied to the delivered gas at a flat rate. Real supply chains leak at different rates in different segments, and the corpus's own thresholds are not quite consistent with one another; the model reports that inconsistency instead of choosing between them.
• The two near-zero thresholds and the two support-lane gates are written on different scopes. The near-zero thresholds are on the supply chain, the Innovation Fund gate is written on process emissions against a separate benchmark, and the model tests the fund gate on the same lifecycle basis as the others. That is a simplification and the reader should treat the fund gate as indicative.
• The permitting and funding material of the report is not in the model. Timelines, grant stacks and state-aid ceilings are dated facts and published rules rather than quantities this world computes, and forcing them into it would produce a schedule the corpus does not support.
• The readiness clock cannot tell readiness level 4 from level 5, because the corpus's scale-up path starts at pilot and publishes no duration for the step before it.
• Everything priced here is a February 2026 vintage. Fifty-seven of the report's hundred claims carry stale-risk flags, and the sharpest is the gas price itself: the corpus records a live print of about 71 a MWh on 1 September 2026, more than double the modelled basis and above the model's entire stress band. The model can be run at that price and the answer changes; nothing in the package hides it.
• There is no price history in the corpus for turquoise hydrogen, so the cost model is not backtested and not projected. What is backcast is the allowance path, the patent filing rate and the supply path, on four published points across three series. That is a thin basis and it establishes only that the drivers are consistent with the record, not that their shape is the only one the record admits.
• What the model cannot tell you: whether a specific gas contract's leakage is actually documented; whether a specific carbon buyer will qualify the powder; what a catalyst lasts at industrial scale; what a grid connection will cost; or what availability the plant will really achieve while it is learning to run. The corpus names those last four as the unknowns only a built plant resolves, and prices only the first of them.