← All lab updates
Our numbers are now on GitHub, next to every article
We opened a public repository, SiliconPsycheLabs/splabs, as the lab's home on GitHub. For every article that shows a result, it holds the same figures as a readable table and as a data file, so anyone can check, quote or plot them without copying them off a web page. It also holds the SDK that connects agents to PSA, our open data, and the link to our open models.
## What is in it
- **The numbers behind our articles.** Every article on this site that shows a chart has a folder in the repository with the same figures: a table you can read, and the same numbers as a data file. At the bottom of each of those articles there is now a link that takes you straight to its folder.
- **The index of everything we have published**, by section, with a one-line summary of each article.
- **The PSA SDK** for Python and TypeScript: the adapters that connect agents built with LangGraph, CrewAI, AutoGen, MCP, A2A, LangChain.js, the Vercel AI SDK or plain HTTP to PSA.
- **Our open data**, each dataset with a card that says how it was built and what it must not be used for, and the link to our open models on Hugging Face.
## Why we did it
We say that we read model behaviour and then correct it. A claim like that is only worth something if it can be checked. Each article already states its method, the bar we set before the test and what the result does not show yet. Now the figures themselves sit in a public place, in a form a program can read, and they are kept in step with the site: when we publish a new result, its folder appears with it.
## What stays private
The test items, the training data and the correction recipes. A test set that is published stops being a test, and the recipes are what we sell. What we publish is what lets you check our claims: the aggregate figures, the method and its limits, and which public benchmarks we used, so the outside parts can be reproduced.
## Where to find it
[github.com/SiliconPsycheLabs/splabs](https://github.com/SiliconPsycheLabs/splabs). The link is also in the footer of every page of this site.
## In plain words
The numbers in our articles are now also on GitHub, in a form anyone can download and check, next to the tools that connect your agents to PSA. Each article with a result links to its own numbers.
This update is not available.