The Prior Art Decides What Is Worth Filing

Run the prior-art search early and the result that matters is never the one you wanted. Not the empty column. The paper that was there first, in a journal nobody in the room reads, published eleven months before the first internal memo. In April 2025 a research group published in Renewable Energy, volume 243, a biochar-supported nickel-iron catalyst synthesised under a carbon-dioxide atmosphere. The figures were complete and specific. 3,172.62 millilitres of gas per gram of biomass at 750 degrees Celsius. A hydrogen yield of 1,452.72 millilitres per gram. Surface area near 330 square metres per gram. Syngas above 82 per cent by volume, with tar handled through the water-gas shift reaction. Read that list again the way an examiner would. Every line of it was printed, dated and indexed before the commercial conversation started. Nothing was stolen. The work went into the open literature, which is a permanent record with a timestamp, and a patent claim is a legal argument about what that record does not already contain.

Investors ask whether an idea is novel. Novelty is not a judgement and it is not a matter of taste. It is a comparison against everything disclosed anywhere in the world before a filing date. Novelty is a date, not an opinion. The European Patent Convention is unforgiving here. Article 54 sets an absolute standard with no general grace period for your own publication. The exceptions in article 55 cover disclosure through abuse and a narrow list of official exhibitions. Publish your own work in a journal and you have described it to the world. The United States is kinder and not kind. One year exists after an inventor's own disclosure, which is why the American route sometimes rescues a paper that has already ended the European filing. A conference poster counts. So does a thesis sitting in a university repository, a supplier data sheet, and the student project nobody bothered to read. A patent attorney will charge for the search and then hand you a document that says the same thing. The difference is the invoice.

The record is crowded, and the crowd is specific. Chemical Engineering Transactions, volume 125, published a 2026 study on rice-husk biochar: roughly 71 per cent toluene conversion at 850 degrees Celsius, an activity span of only 20 minutes, and surface area moving from 17 square metres per gram raw to 440 under carbon-dioxide activation and 426 under steam. Silicon made up 94 per cent of the inorganics, and that one fact is why impregnated metals underperform on that feedstock. Journal of Cleaner Production described a sorbent built entirely from carbide slag and dolomite at a ratio of 90 to 10, holding 0.52 grams of carbon dioxide per gram after 20 cycles. Separation and Purification Technology published steel slag reconstructed into a CaO adsorptive phase and an FeOx catalytic phase, and named the two dead phases, Ca2Fe2O5 and Ca2SiO4, that had to be engineered around. Then the count. PatSnap recorded 3,001 active patents in carbon capture materials as of April 2026, with 655 new filings in 2024 alone and a steady drift from amine solvents toward solid sorbents, membranes and electrochemical routes. An idea is not worth money because it is good. It is worth money because the record does not already contain it.

Blocked ground is not the end of a search. It is the shape of what is left. Published work from the last two years leaves three kinds of room. A composition the literature has not described, which is the narrowest and the most valuable. A method of operating something already known, which is where most live filings sit. And an integrated system that arranges known parts against a constraint nobody has solved. The third kind is usually the open one. Calcium looping reached readiness level 7 at a 200 kilowatt thermal circulating fluidised bed pilot in Stuttgart, with capture above 90 per cent across 120-hour campaigns run between a carbonator at 650 degrees and a calciner at 890 to 930 degrees under oxy-fuel firing. The pilot is not the news. Sorbent deactivation is the named bottleneck in the same literature, and a bottleneck described in public is a problem still waiting for an owner. Frontiers in Chemistry added a smaller detail in July 2025: two per cent water vapour raises the carbonation rate of calcium oxide between 400 and 650 degrees, with a pseudocatalytic effect that fades as temperature climbs. The commercial record has already moved too. WtEnergy's SynTK plant in Barcelona, launched in August 2025, is the first gasification-to-cement-kiln installation on the record: 30 megawatts thermal of fuel substitution, 90 per cent chlorine reduction, 95 per cent alkali removal and 55,000 tonnes of carbon dioxide avoided a year. Every line above is a claim some industrial group could have filed and did not.

Novelty is a date, not an opinion.

A grant panel is not a patent office, and the two disagree about what counts. The Innovation Fund, Horizon Europe and the national programmes underneath them all ask for freedom to operate, and the answer comes out of the same search. A published paper is a citation. A granted claim is an asset. Treating the first as the second is how a technically strong application reads as commercially unowned. The probability that a competent examiner finds the paper you hoped nobody would find is not zero, and most companies price it at zero. That gap is the risk. Investment committees behave the same way. Diligence in a deep-tech round now opens with a patent landscape rather than a financial model, and a portfolio that maps onto somebody else's publication reads as a research programme.

Search before the deck, not after. Run the query your future examiner will run, in the same databases, and read the figures instead of the abstracts. Then decide what you are filing. The ambition is not a claim. The composition, the operating window, the cycle regime and the integration are claims, and the record has to be silent on the one you choose. Four questions separate a filing from a hope. What is new, in one sentence an engineer would accept? What is the closest published work, and what does it fail to do? Which numbers in your own data are absent from that paper? And who owns the space around you if the examiner reads it the way you did? Answer those and the patent falls out of the work. Skip them and the paper you never read is already standing where your claim was going to be.