Evidence first
Every claim we make about a paper or a stack is traceable. If you cannot quote it, you cannot say it. Our pipeline literally enforces this on auto-published critiques.
About
A small consultancy with strong opinions and a public method.
We exist because too much LLM and graph-AI work ships without anyone asking the boring evaluation questions. We are small on purpose; depth beats coverage.
Operating principles
How we work.
Public retractions
When we are wrong, the retraction is on the same page as the original entry, with strikethrough and a note. The retraction log is open.
Tools over decks
The deliverable is usually code or data, not a slideshow. Our open tools (toslop, GraphSlop, the tracker) are how we keep ourselves honest.
Local where possible
The research pipeline runs against a local model. We pay the cost of running our own inference because it changes what we can publish openly.
Read next
Where to start.
If you are evaluating us for an engagement, the most efficient sequence:
- Read the method page — how the auto-publishing actually works.
- Skim the latest research digest.
- Open GraphSlop and pick any entry; that is the shape of a critique we publish.
- Then tell us what you are trying to ship.