Press Kit
Press Kit
Download official Skelf Research brand assets for press coverage, partnerships, and publications. All assets are provided for editorial and non-commercial use.
Everything on this page may be reproduced without asking, provided the lab is not presented as endorsing the piece it appears in. If you need an asset in a format that is not here — vector, a light-background lockup, or a higher-resolution mark — ask and we will send it rather than have you rescale a PNG.
Logo
Our logo features a geometric network mark representing interconnected research nodes. Use the primary version on dark backgrounds.
Primary (Dark BG) 
Color Palette
#6ee7b7 #08080c #e8e8f0 #34d399 Typography
About Skelf Research
Skelf Research is an independent applied-research and technical-validation lab based in St Andrews, Scotland. Organisations commission it to investigate consequential AI and systems decisions; it publishes its research methods, its open-source systems and a public register of the evidence behind its claims.
Our research spans five domains: LLM Cognition & Prompt Theory, Safe & Verifiable Computing, Formal Optimisation & Decision Science, Edge Intelligence & On-Device AI, and Robotics & Autonomous Systems. Our canonical portfolio contains 19 current public products that reviewers can inspect directly; 18 carry a detected OSS licence.
How to describe us accurately
Skelf Research is an independent applied-research and technical-validation lab. Organisations commission it to investigate consequential technical decisions — whether an agent is ready to release, whether a cheaper model is good enough, whether optimised GPU kernels are silently wrong — and it delivers reproducible evidence and a recommendation (see services). Its research methods and open-source systems are public; its customer work is confidential. It is not a software vendor: the systems are instruments and evidence of capability, not a product the customer must adopt.
On the portfolio: describe it as a family of independent products at their own subdomains, not as a platform or a suite. Each one answers a separate research question and none of them depends on the others. Writing that presents them as modules of a single system misdescribes the architecture.
On licensing: check the individual repository rather than generalising. The portfolio spans MIT, Apache-2.0 and GPL-3.0, and one product (Perishable) is public-source with no licence file yet. “Fully open source” is not accurate for the portfolio as a whole, and the per-project pages state the position for each.
On status: several products are explicitly experimental at v0.1.x. If a piece depends on a capability claim, the honest source is the scope section of that product’s own site, which states what it does not do as prominently as what it does.
Suggested story angles
Publishing negative results. Several repositories in the portfolio answer their research question unfavourably and remain public with the disappointment documented. The interesting question for a reporter is why that is rare in commercial AI work rather than why it happens here.
The scope declaration as a genre. Every product states its boundaries next to its features. That is unusual enough in developer tooling to be worth asking about, and it makes the tools easier to evaluate and harder to oversell.
Measuring AI answer engines directly. Because answer surfaces rarely produce referral traffic, conventional analytics cannot see them. Our approach is to query the surfaces on a fixed prompt set and record the citations, and the method is published rather than described.
Small tools against large frameworks. The portfolio is a deliberate bet against the framework model in AI tooling. It has costs — every product is a surface that has to be kept current — and the trade is arguable in either direction, which makes it a real story rather than a press release.