@eto-press/engine-eto

The flagship doctrine: one story, many mouths.

The eto engine decides what a story is — one event, told by at least two outlets that disagree — and how it must be told: a composited brief with the disagreements named in plain words, every source linked, and an end. Stages 1–9 of the pipeline live here, behind the one joint every engine shares: prefilter, judge, cluster, the density gates, cross-edition dedupe, selection, the below-the-fold nomination, the compositor, and the verification cage that checks the compositor's every quote against the accounts it was given.

It asks a model exactly three typed questions — same event, the nomination, the composite — through the platform's inference boundary, and never sees a prompt. Everything else is deterministic code.

Doctrine

One story, many mouths. The disagreement is the story. Nothing unattributed ships. The model composites; it does not comment. Incomplete beats wrong. The masthead is yours.

A paper declares it with [engine] use = "eto" in eto.toml — or by saying nothing, since it is the default. It needs at least one [[source]] in sources.toml and the two local models (eto models pull).

License: AGPL-3.0-only.