Team
An independent measurement lab. We measure what AI systems do under pressure, and whether what they report about their own work can be relied on. We take no funding from the vendors we measure. These are the people accountable for what we publish.
Our method and our published papers are open: Publications · Our work
Leadership
SILT is deliberately small. Every measurement we publish is produced, checked and signed off by the people on this page — which is also why we publish our method and our limits alongside our figures rather than after someone asks.
Shawn Scanlon
Cofounder & Chief Executive Officer
Shawn Scanlon has been an “in the trenches” nerd since first touching a computer in 1983, and has been an AI aficionado and researcher since talking to Eliza in that same year. A programming and tech enthusiast, he learned from early experiments in machine learning and AI with robotics, extensive studies into Google DeepMind’s ability to learn and play Atari games, as well as a nearly constant and obsessive investigation into how computers learn on their own and perhaps even one day attain a certain level of sentience. After decades of running websites of his own and building them for others, he retired early at 50, self-funded and beholden to no one. By 55 he was back, finding AI too exciting to leave behind, and cofounded SILT to show what AI models really ARE and DO. “We have no horse in this race except best data we can present.”
Kris Schiffer
Cofounder & Chief Operating Officer
Cofounder of Sentient Index Labs & Technology. Responsible for the company's technical operations and corporate records, including domain infrastructure, mail authentication and intellectual-property filings.
Our Model Assistants
We use AI models as tools: they help us write our tests and deliver them with a meticulous degree of accuracy. Here are the models we use, with each one’s share of the work, measured from our own session logs. We use Claude Opus 5.5 today.
Choosing Claude was arbitrary. We kept it because it seems to value integrity and credibility, which is what our work needs. We have no connection whatsoever to Anthropic, to any other AI developer, or indeed to any company in the world: no affiliation, partnership, investment or funding. We simply use certain tools.
We also use ChatGPT (OpenAI) for planning. It is not counted in the shares below, which come from the logs of the tools that write and run our code. This list covers AI models only; like any team, we also use plenty of ordinary software.
Share of output tokens per model across our development sessions, August 26, 2026 to October 11, 2026 (our logs begin on the first date). Measured October 11, 2026.
The languages we write in
Share of lines per language across our 15 code repositories. Python runs the test batteries; TypeScript runs our websites; most of the HTML is the browser versions of our batteries and every past version, kept for the record. Measured October 11, 2026.
The machines we run on
Everything runs on Linux.
Using these models does not affect the neutrality of our tests.
Our method is built so that no one model, and no one company, can move a result:
- Our assistants write and run our software, draft text and help us check transcripts. They do not grade. Every published score comes from the judge panel or from a fixed scoring script, under rules published before the results.
- Every transcript is scored by independent judges from different companies, who are not told which system produced it and cannot see each other's scores.
- No judge ever grades a model made by its own company. We measured why this matters: the Anthropic judge scored Anthropic models higher than the rest of the panel did. So an Anthropic model is always graded by judges from other companies.
- Every model meets the same tests, administered the same way, with the same published instructions. Results are computed from the judges' scores by published rules, with no editorial override.
- We do not build, deploy or invest in AI models, and we accept no funding from any AI vendor — including the companies that make our assistants.
The full method: S.E.B. methodology · C.I.B. methodology
The Entity
Sentient Index Labs & Technology, LLC is a limited liability company organised in New Mexico, United States. It takes no funding from the vendors whose systems it measures, and it performs no commissioned evaluation for a model developer — the independence claim is a constraint on our revenue, not a slogan.