Skip to content

↑↓ navigate · ↵ open · esc close

Collaborators

Open problems, not job titles.

We are looking for people who want to work on systems where being right is the whole job. The fastest way to find out whether we fit is to look at what we are stuck on.

SUTP

Correlation cap: from observe to enforce

The cross-book correlation cap currently observes. Promoting it to enforce needs replay evidence across regimes and a way to size the cap that does not itself overfit.

SUTP

Named-identity audit actors

Every correction is audited, but actors are roles, not identities. Design the identity layer for a multi-operator future without weakening the single-operator present.

TaxReturnStudio

Completeness without a named input

Seven forms can be derived from the acceptance corpus; the rest are blocked on inputs the documents do not name. How does a completeness contract ask the right question of the advisor?

TaxStrategy Studio

State engines that stay deterministic

Federal and a handful of states are modelled deterministically. Extending to more states without a combinatorial test burden is a modelling and tooling problem.

Clearance Hunter

Ranking that learns from a thin outcome stream

A single operator produces few purchases and sales. What learning rule improves ranking from dozens of outcomes without chasing noise?

Studio

Release gates for a one-person fleet

Tens of thousands of tests, independent review and immutable releases already run. What is the next control that is worth its cost at this size?

How we work

  • Written spec before code; the spec names its reviewer.
  • Red first: the failing test exists before the implementation.
  • Independent review of every change, no exceptions for size.
  • A gate receipt travels with every release; change the tree, earn a new one.
  • Paper before live, and live only behind an authorisation the operator holds.
  • Credentials, account identifiers and client data never leave their host. We will not ask you to handle them.

Start here

Send us a note on one problem.

Pick one of the problems above, or one of your own that you think belongs here, and tell us how you would attack it in a paragraph. We read every one and reply within two working days.

Next stepChoose your door