Field Notes · 2026
An independent research program

Most of what a model thinks never mattered.

RLTS studies the internal economics of large reasoning models. What a model genuinely deliberates, what it merely rehearses, and what it can afford to forget.

Research in progress · 2026

We think the current generation of reasoning models is wildly inefficient with its own tokens, and that most of the gap between a frontier model and a small one lives in deliberation, not knowledge.

Our work is aimed at closing that gap with methods, not scale.

N° 001
On the structure of a reasoning trace, and what a trace is actually made of.
In preparation
N° 002
The cost of thinking out loud when nobody is listening.
In preparation
N° 003
Distillation without dilution: teaching a student to carry a teacher's doubt.
Forthcoming
N° 004
A negative result, held back until we are sure it is the right negative result.
Held