Better models make the data engineering matter more.

Before a model can find a signal, someone has to make sense of the data.

INSIDE THE MECHANISMFIG. 01
01RAW INPUTSNoisy inputs
02USABLE DATATime-aware data
03RESEARCHTestable ideas

What the source says

Eric Mannes describes the work behind Jane Street’s alternative datasets: resolving entities, structuring inconsistent inputs and handling revisions. The conversation connects that work to research and to the opportunities created by stronger machine-learning models.

The markets connection

Our reading: better models increase the value of getting the inputs right. A company identifier and a timestamp encode decisions about what a model knows. Those decisions deserve the same scrutiny as the model itself.

A question to take back to your work

Could you reconstruct the dataset as it was known at the time of a historical decision? A revised dataset can tell a different story from the information originally available.

Keep it in context

A practitioner interview, not an independent model-performance study. The markets connection is Market Mechanics’ interpretation.

Wrestling the World into Rows Jane Street · Signals and Threads · 2 September 2026
Article history
  • 18 September 2026 — First web edition. Original interview date retained; source checked.