Chemify

The first record in this lane to fall short of the top grade here, and the reason is a genuine question rather than a technicality. Chemputation is described by the company as a universal chemical programming language combined with robotics and artificial intelligence, and elsewhere in its own press releases as being itself purpose built artificial intelligence. A chemistry society describing the same platform calls it a programming language, robotics and a database of validated reactions.

A programming language is not a model and a reaction database is not a model. Apply the test this axis uses and the answer is uncomfortable: remove the machine learning and a robotic facility that executes coded synthesis routes from a validated library is still a business, and a substantial one.

The models do real work in molecular design and route selection, and three peer reviewed papers in 2026 cover discovery and literature to synthesis translation, so this is not a company wearing the label without cause. But the centre of gravity here is automation and manufacturing, and the grade should say so.

The highest autonomy in this lane paired with the most structural constraint on it. Autonomy is literal here, with instructions executing to physical product without a human performing the chemistry, which is further than any neighbour goes.

The constraint is architectural rather than procedural: the system can only execute routes drawn from a curated library of validated reactions, so the boundary of what it will attempt is set by what has already been shown to work rather than by what a model proposes. In the funded programmes the division of labour is stated plainly, with Chemify designing and synthesising while the partner runs enzyme and whole cell assays, which puts testing outside the company.

What is undocumented is everything procedural: no human approval gate is described, nothing states what happens when an automated route fails or yields the wrong product, and the validation methodology behind the reaction library is not published.

The chemistry is documented and the artificial intelligence is not, which on this index is the half that counts. The programmable synthesis concept rests on twenty years of published academic work and three current peer reviewed papers, so a reader can understand how code becomes a molecule in real detail. Set against that, nothing describes the models: no architecture, no class, no training data, no scale, no version and no benchmark.

Worse, the company's own descriptions do not agree on whether Chemputation is artificial intelligence at all, calling it purpose built artificial intelligence in its press releases while a chemistry society describes the same platform as a programming language, robotics and a reaction database. The reaction library is the load bearing asset and is described only by superlative, with no published size, composition or validation methodology behind the claim that it is the world's largest.

The intellectual provenance is the clearest in this lane and the operational provenance is absent. Everything traces to a single named academic source, the founder's group at the University of Glasgow, described by the company as more than 500 published papers over two decades, and the current platform papers are citable and public, so a reader can follow the ideas back to their origin in a way no neighbour permits. What is missing is everything a buyer would actually need.

The curated reaction library is the primary input asset and nothing states its size, its sources, or whether it was assembled from published literature, from in house experiment or from licensed data. No compute or hosting provider, foundation model, open source inventory or third party licence is disclosed, and the robotics hardware provenance is unstated, including whether it is built in house or sourced.

Stronger than most of this lane on published science and weaker on customers. Three peer reviewed papers appeared within two weeks of each other in April 2026, in PNAS, Nature Communications Chemistry and Nature Communications Biology, covering molecular discovery, automated synthesis translated from published literature, and application to cancer drug development. A physical facility has been operating since June 2025.

Two funded programmes exist with an external partner running the assays, and a malaria candidate was described in December 2025 as approaching investigational new drug enablement within twelve months. The qualifications are specific. The papers were authored by the founder with co authors from the university where he holds his chair and from which the company spun out, so they are peer reviewed but not independent of the company.

The superlative claims, the world's most advanced synthesis facility and the world's largest curated reaction library, carry no third party measurement and no published size for the library. And the malaria timeline is forward looking, with no application or clearance located as of 29 August 2026.

No protected health information exists in this workflow. The stewardship question that does apply is unlike anything else in this lane, because the output of this system is physical matter rather than a prediction or a sequence on a screen.

This is an automated facility that converts code into synthesised compounds, the company intends to replicate it as a distributed network of sites, and the constraint on what it can produce is a validated reaction library whose contents are not published. No statement of what the platform will decline to make, what controls sit between an instruction and a synthesis, or how the reaction library is governed appears on any company surface as of 29 August 2026.

Recorded as a located absence: nothing indicates a problem, the company operates under United Kingdom regulation that governs chemical facilities regardless of what it publishes, and nothing published addresses the question. The partner data separation issue common to this lane also applies and is untestable with only two counterparties named.

Graded neutrally because the obligation does not arise on any reading. The platform operates on molecular structures, synthesis routes and reaction data, with no protected health information anywhere in the workflow and no patient derived material, and the company is a United Kingdom entity whose data handling falls under British regulation rather than the United States framework this axis was written against. This is the cleanest application of the convention in the whole lane, since unlike several neighbours here there is no adjacent concern the axis is failing to capture.

Two dedicated passes on 29 August 2026 across a targeted search and a direct review of all eight navigation sections and the footer found no security page, no trust centre, no certification and no attestation of any kind. The company publishes three legal documents and none of them addresses information security beyond a standard disclaimer that the site is not guaranteed secure.

The absence is more conspicuous here than at the design only vendors in this lane, because this company operates a physical chemical facility where certification against recognised safety and quality standards is ordinary practice in the industry. Nothing published either claims or documents any. That is a statement about disclosure, not about the underlying compliance, which United Kingdom law governs regardless.

Nothing cleared and nothing adverse. No candidate has an approval or an investigational application on the public record, and as a Scottish company Chemify would engage United Kingdom and European regulators before the FDA. A malaria programme was described in December 2025 as moving toward investigational new drug enablement within twelve months and aiming at a single dose cure, but as of 29 August 2026 two dedicated passes located no filing, clearance or trial registration.

Worth noting for a later reader that this vendor operates a chemical manufacturing facility, which carries its own regulatory regime for process and workplace safety entirely separate from medicines regulation, and this axis does not reach it.

No governance framework, model card, evaluation charter or responsible use policy appears across the eight sections of the company website as of 29 August 2026. Partial substitution comes from the published science: three peer reviewed papers in April 2026 expose the methods and their validation to external scrutiny, and the founder's group has published extensively on the underlying approach over two decades. That is the same basis on which several neighbours in this lane hold this grade.

What none of it supplies is any statement of when the company judges its own system unreliable, which matters more here than elsewhere because the failure mode is not a bad prediction on a screen but a physical synthesis that consumed reagents and produced something unintended.

Three legal documents are published and reading them lowered this grade rather than raising it. The terms of use, marked version 1.0 and undated, are an unlocalised United States consumer software template applied to a Scottish company.

They invoke the American Arbitration Association and its consumer rules, United States export control law, the Digital Millennium Copyright Act, California civil code and the California Department of Consumer Affairs, and they name no British jurisdiction anywhere despite the company operating under Scots law. The chosen court is a county in California that does not exist, a known artefact of template text left unedited.

Liability is capped at fifty United States dollars for a company that has raised roughly $107 million and physically manufactures chemical matter. Notices are directed to a university chemistry department rather than to either corporate address in the site footer. Several clauses are grammatically incomplete, and the document governs an account and user content on a site that offers neither.

None of it reaches the questions that matter for this business: who owns a designed molecule, what warranty attaches to a synthesised compound, and who carries the consequence when the product is wrong.

No electronic health record surface exists and none would be appropriate, so this is graded neutrally under the convention for this lane. Recorded in its place, because it is the most interesting version of this question in the whole cluster: the interoperability here is with the chemical literature itself.

One of the April 2026 papers demonstrates automated synthesis translated from published scientific literature, which means the integration surface is every procedure ever published rather than a partner system. That is a genuinely different model of interoperability from an application programming interface. No software product, connector or programmatic access is offered to customers.

Graded neutrally under the convention for this lane, now the eighth consecutive record where the convention rather than the evidence sets this grade, though here the axis means something unusual. Deployment for this vendor is physical: one operating facility in Maryhill, Glasgow, a stated intention to build a network, and a Silicon Valley site planned, which is a real and disclosed deployment footprint of a kind no neighbour has. What is absent is the data half.

Nothing states where a partner's molecular designs are held or processed, under what separation, with what retention, or what happens to the synthesis data a run generates. A planned expansion into the United States by a United Kingdom company would raise cross border questions that nothing published anticipates.

Better than the lane on figures and worse than the lane on customers. The disclosed numbers are unusually concrete and, importantly, are money received rather than milestone ceilings: $1.6 million in follow on grant funding from the Gates Foundation in December 2025 against $3.2 million cumulative, a $16 million facility cost, a $43 million Series A with its lead named and $9 million of it identified as a United Kingdom government programme, a Series B above $50 million, and a headcount above 200.

Almost nothing else in this lane states what it actually received. Against that, the only named counterparties are a philanthropic funder and a small biotech collaborator; no pharmaceutical company, commercial contract or licence appears anywhere. The founding year contradicts the company's own earlier announcement, 2022 on the current site against 2019 in its 2023 funding release. And one report renders the cumulative grant total in sterling where the company's own release states dollars, so a reader checking the figure will find two.

The widest coverage graded in this lane, on two dimensions. Across the workflow it is the only vendor here that spans design through physical synthesis and into manufacturing at experimental scale, closing the make step that its neighbours hand to somebody else. Across application it reaches beyond medicine by design, with the company naming catalysts, electronic materials, agriculture and green energy alongside drugs, and disease work covering tuberculosis, malaria and cancer.

Held below the top grade because the reach is asserted more widely than it is evidenced: no materials or industrial customer is named, nothing has reached the clinic, and a stated intention to build a global network of facilities currently rests on one operating site.