Invarra field notes

Making agent actions and AI behavior easier to examine.

Practical explanations of Phalanx execution control, alongside research on meaning, measurement, and reliable evidence.

When an AI agent can take action

A practical introduction to task authority, protected credentials, and the boundary between a proposed action and an executed one.

Read the explanation

Research field notes

These articles explain the ideas behind LIP and CSR. They provide research context and do not define Phalanx V1's runtime dependencies.

Why one correct response leaves an important measurement question unanswered.

How a shared semantic unit changes the way we organize an evaluation.

What differing outcomes can reveal when the underlying meaning stays fixed.

Why an evaluation must define what can change and what must stay the same.

A careful starting point for claims about targets we observe through representations.

How separating the measurement layers helps locate a system's sensitivity.