Skip to main content

​
Anya

Anya is the analyst. Where Terminal gathers evidence, Anya reads it and forms a judgement about what it means. That division is deliberate and it is enforced in the product, not just described here:

Terminal finds evidence

Sourced facts, tables, and figures, each carrying its provenance. It does not tell you what they mean.

Anya forms the judgement

A reasoned read of that evidence, marked as inference and never dressed in the authority of a source it did not come from.

​
Claim types

Everything Anya writes is typed, so a reader can tell what kind of statement they are looking at: The distinction between the first two is the whole point. An inference presented as a sourced fact would be the product lying about where its confidence comes from.

​
Honest degrade

Anya refuses rather than fills gaps. When the evidence floor is not met, a brief is withheld and says so; it does not produce a plausible-looking narrative from thin data. If a figure has no primary or official evidence behind it, it is blanked rather than estimated. This is why a run can complete and still deliver nothing, and why that is reported as a hold rather than a success.

​
Relationship to Workflows

A workflow can include an Anya investigation as a step. Two things are worth knowing before you do:
An Anya investigation is metered separately from the workflow that triggers it, against your Anya research allowance. A workflow’s own estimate does not include it.
Anya steps are not marked safe for unattended use, so a workflow containing one cannot be published on a schedule or an event trigger. It runs on demand, with a person present.

​
Anya’s own API

Anya is addressable under /anya on the same host as Workflows:
It is a separate surface from Workflows, not a legacy one, and it stays that way deliberately. A workflow can invoke an Anya investigation as a step, which covers the case where analysis is one stage of a scheduled pipeline. It does not cover most of what Anya is. Universes and positions, cases with handoffs and outcomes, beliefs and their revision history, alert rules, calibration and track record: these are stateful, conversational, and reviewed by a person as they go. A composition graph models a run with a beginning and an end, so expressing any of that as a workflow would mean expressing it badly. The practical read: reach for Workflows when research should happen on a schedule and land somewhere without anyone present. Reach for the Anya API when a person is working a position, a case, or a question over time.
This surface is not yet documented page by page here. It is reachable today with a credential carrying the anya entitlement, and through the web application. Reference pages follow.