Docs

How the instrument reads an agent, end to end.

Start with what Holonograph measures and why, then follow the method into attribution, variance, coverage, and the SDK that carries it.

Overview

What Holonograph is, the substrate it builds, and the reading it produces. The right first page.

Every topic in a single document, for reading end to end or searching in one place.

Where the lens sits, and why it captures every model call in both directions.

The structured record every reading is computed from.

How a moment of the system is frozen for comparison over time.

What a surface declares it will do, checked against what it did.

Splitting a shifted score into agent, model, eval, and noise.

Telling real change apart from sampling and chance.

Declared surfaces, observed coverage, and the size of the blind spot.

Whether a judge actually ran, or silently defaulted.

How findings turn into tightened checks over releases.

Production, evaluation, and the provenance every call is tagged with.

Comparing vendors on identical terms, one lens across all.

The client contract, and the reliability loop: attach a claim, and the lens checks your judge.

Emitting to the lens without changing your request path.

Fire a signed webhook when a predicate over the lens's own verdicts crosses a line.

The overview walks the whole instrument in a single read, from substrate to reading.