Anyone who has costed a process knows what happens to a headline figure on contact with a real plant. The number was computed at a capacity factor nobody achieves, or at an energy price from a different year, or on a basis that quietly excludes the largest cost in the system. The figure is not dishonest. It is simply unusable, since the assumptions that produced it were never stated, therefore it cannot be recomputed for your case.
Serinyx published three reports today, and they are built the other way round. Every figure states its basis. Every model input is bounded by the extremes of published estimates rather than by round numbers chosen to make a demonstration look tidy. Each report says explicitly what would have to be true for its own conclusion to fail. This note walks the three of them at the level a technical reader will want, and then invites you to attack them.
Methane pyrolysis, and the two tests it has to pass separately
Hydrogen's Dark Horse Is Turquoise models methane pyrolysis, which produces hydrogen and solid carbon instead of carbon dioxide.
The energy comparison is the first thing worth getting right, since it is usually reported wrongly. Pyrolysis needs roughly 12 kilowatt hours of electricity per kilogram of hydrogen, against 52.4 to 55 for electrolysis. On total energy, however, pyrolysis is the hungrier route, since it also consumes about 59 kilowatt hours as methane. What it saves is electricity, and only electricity. If you are constrained by grid connection rather than by primary energy, that is decisive. If you are not, it changes the answer.
The economics turn on the solid carbon. A kilogram of hydrogen costs 5.24 euros before any carbon revenue, 2.44 euros with the carbon sold at 700 euros a tonne, and 5.39 euros if the carbon has to be buried. The alternatives are worth setting out on the same basis. Turquoise hydrogen with the carbon sold runs 1.51 to 4.29 euros a kilogram, grey with its carbon bill 2.77 to 3.69, blue with capture 2.93 to 5.01, and green today 6.03 to 10.21.
Then there is the second test, which is separate from cost and does not care about it. A methane leak rate above 0.63 per cent invalidates the climate case, and both published figures fail that threshold. A configuration can therefore be the cheapest option in the model and still not qualify. The model computes its own sensitivity ranking. The carbon co-product price moves the cost by 5.20 euros a kilogram across its range, which is more than the gas price, electricity price, capacity factor, cost of capital and build cost combined.
The report also carries the constraint that most turquoise-hydrogen analysis omits. The carbon has to be sold to somebody, and the market for it has a size. Selling more carbon at 700 euros a tonne than the market can absorb is not a scenario, and the model treats it as a contradiction rather than an input.
Battery recycling, where the route decides the answer
Europe's Most Valuable Mine Is a Scrapheap is a materials and process argument before it is a policy one.
Feedstock is arriving on dates fixed by vehicles sold years ago: 230,000 to 420,000 tonnes in 2030, rising to 1,500,000 to 2,100,000 tonnes by 2040. About three quarters of the 2030 feedstock is factory scrap rather than end-of-life packs. That distinction matters, since scrap and dead cells are not the same material and do not behave the same way through a plant.
The refining route decides whether the lithium survives. Smelting recovers 0 to 30 per cent of the lithium and loses the rest to slag, while recovering 85 to 95 per cent of the cobalt and nickel. It is also the only route at full commercial readiness. Hydrometallurgy recovers 90 to 98 per cent of the lithium and 95 to 99 per cent of the cobalt and nickel, at early commercial readiness. Direct recycling retains 95 to 99 per cent of the cathode material as cathode material, and is at pilot scale.
From August 2031 recycled content becomes mandatory, at 16 per cent cobalt, 6 per cent lithium and 6 per cent nickel, with quotas roughly doubling by 2036. That is the point at which the route that destroys the lithium stops being commercially rational regardless of its margin, since the material it destroys is the material the law requires. Meanwhile the metal itself has become cheap. Lithium carbonate ran at about 85,000 dollars a tonne at the 2022 spot peak, about 46,000 in 2023, 12,374 in 2024, and 10,542 forecast for 2025. A seller of black mass keeps three to five per cent of the lithium price. Announced European refining capacity for 2030 is roughly double the top of the expected feedstock band, and some plants report utilisation around ten per cent.
Humanoid robotics, and the difference between a zero and a gap
The Demo Economy is the report a technical reader will find most familiar in spirit, since its subject is a measurement problem.
The public record for 2025 holds 292 disclosed award projects, one fully documented acceptance chain, no verified payment remittances, and no verified repeat purchases anywhere. The most-cited Western deployment metrics appear twelve times across the reviewed record and collapse to a single origin, which is the vendor's own pilot dataset. No flagship programme publishes an autonomous share of task time or a measured human-minutes-per-robot-hour figure, and seven classes of public register that would carry such figures were checked and found silent.
The report is then careful in the way an engineer would want. Publishing those deeper stages is not generally mandated for ordinary goods procurement in China, therefore those zeros are ceilings on what can be known rather than statements about what happened. The audited economics that do exist are reported as they stand. One issuer sold 3,551 units in the first nine months of 2025 against 5 in 2023, at a gross margin of 62.91 per cent down from 87.67 per cent. The average selling price fell to 167,600 yuan from 593,400.
The models, and the invitation
Each report is built on a working model of its subject. Every input in those models carries three things. There is a default, which is the report's own central case. There is a range whose ends are the extremes of published estimates rather than convenient round numbers. Finally there is a note saying which published source each end came from. Where a value is an assumption rather than a measurement, it says so in its own first sentence.
The models also state where they break. Each carries a small set of pre-registered hypotheses with explicit thresholds, so the conclusion is reported as holding, straining, or failing rather than as a verdict with no failure mode. Every equation names the published method it comes from. Where the evidence will not support a forecast, the model refuses to produce one and records why, which happens more often than a marketing document would like.
An accompanying interactive version of each model lets you set the inputs to your own case and watch what happens. We would rather you did that than take our word for any of it. If you find a configuration that is defensible from the published record and that overturns a conclusion, we would like to know, and the report will be corrected.
What these three reports are not
Each is a summary of something much larger. The hydrogen report runs to about 7,700 words and draws on approximately 2.08 million words across 1,051 documents and 16 research domains, which is roughly one word in two hundred and seventy. The recycling report is about one word in a hundred and thirty of its research. The robotics report summarises a working corpus of about 50.1 million words in 29,111 files.
They also cannot draw on operator data or confidential submissions, and sources that cannot be publicly linked are listed without a URL and say so. A twenty-minute document written for a general reader is not the instrument you would use to decide on a plant. It is the instrument you would use to decide whether the people who wrote it are worth talking to.
If a question in front of your team looks like one of these, and the analysis you have been handed does not show its assumptions, we should talk.