Hey. Starting Vixr has been a bit of a blur.
It started with a question: when an SEC filing drops and the language clearly signals something, why does the market sometimes take hours, or days, to fully react? That gap between information release and price absorption felt like something we could measure.
That question turned into months of work. Testing approaches that failed. Finding that LLMs can't predict prices (nobody can). Discovering that they can read quantitative signals and filing text to assess whether information is priced in. And eventually building a pipeline that shows a preliminary directional signal on a small dataset, verified against a random control.
We want to stress: this is early-stage research. Two months of data. A small sample. Promising initial results that need much more validation before we'd call them conclusive.
Why publish this
The signal intelligence space has a credibility problem. Black boxes. Unverifiable claims. "90% accuracy" marketing that falls apart under scrutiny.
We're going to publish the real methodology, the real numbers, and the real failures. If our approach works, the data will show it over time. If it doesn't, we'd rather know. And we'd rather you know too.
What to expect
- Model updates. What we changed, why, and what the numbers say.
- Product updates. What we're building on the platform.
- Research. Methodology deep-dives and signal breakdowns.
- Roadmap. Where we're headed and why.
We're starting late in the journey, so the next few posts will catch you up on how we got here. After that, updates will be real-time as we iterate.
Let's go.
— The Vixr Team