One discipline, applied to more than one problem.

Can the loop run on somebody else's problem?

The shared machinery: authorized data in, candidates generated, attacked, mostly rejected, failures kept, survivors watched forward, advancement controlled. It runs today as research infrastructure. Turning it into something a customer can operate on their own data is the work.

Does this prediction deserve to be acted on?

The model side of the loop, running continuously against real market data. It runs on XGCS research infrastructure planned for licensing to Kairos. Its record, including the results that failed and the ones later found to be contaminated, is open to read.

What does this loop need to know about markets specifically?

Connectors, cost models, regime tests, baselines and evaluation rules for the first domain. Markets were chosen because a weak model fails there quickly and measurably, not because the company is a finance company.

Was this action allowed, and can you prove it?

A control boundary for autonomous systems: authority delegated in bounded envelopes that fail closed, every consequential action checked before it leaves, and the decision written to a hash-chained record. Nothing has deployed it in production, nobody uses it, and it has never governed a real agent.

Can an agent be given real operational reach without giving it the keys?

Intent becomes a typed request, authority decides, a bounded worker executes, and a receipt comes back. This runs internally today against a narrow set of operations. It is the same shape as Lightcone, applied to the company's own infrastructure rather than a customer's.

Can context be owned by the person it describes?

A draft interoperable standard for user-owned AI context: what a tool may be told, scoped to a purpose, with provenance attached. It is being drafted as an open standard rather than a Kairos product. Implementations, custody and management tooling around it may be commercial; the format itself is not intended to be.

What would it take for software to understand a person well enough to be useful, and no further?

Working prototypes exist for conversational intake, intent extraction and matching within a trusted network. How much of this becomes a Kairos product rather than adjacent research is not settled, and the ownership question behind it is open.

Can a model explain a finding without being able to invent one?

A pipeline exists that composes and renders from a sanitized snapshot and cannot reach past it to the underlying research. Whether that becomes a product or stays internal plumbing depends on evidence that has not been gathered.