Geospatial and media fleets
===========================

.. note:: Planned architecture.

Agency can operate city-scale terminal fleets while keeping high-volume sensor
data outside LLM prompts and global Channels groups.

Geospatial terminals
--------------------

Fixed shop tablets and moving taxi tablets report authenticated, coalesced
position updates. Mission Control presents an authorized map with presence,
heading, active content, campaign and render status.

GeoZones and targeting rules deterministically calculate which Displays receive
a campaign. Veronica chooses the campaign intent and handles ambiguity or
failure; she does not manually select every terminal and is not invoked for
every coordinate.

Typical flow::

    position -> geofence service -> semantic transition
             -> campaign target set -> Assignment/Scope delta
             -> terminal render acknowledgement

Map subscriptions declare viewport and filters. The server clusters and
coalesces updates instead of broadcasting the entire fleet.

Camera networks
---------------

Cameras are media sources, not Properties. Channels carries small events such
as presence, motion, sample-ready and observation-ready. Frames and clips use
object/media storage with short-lived signed URLs; live preview uses an
appropriate media protocol such as WebRTC, HLS or authorized HTTP.

A monitoring Mission waits persistently without holding a worker. A semantic
event schedules a new Run, a permitted vision Agent analyses a bounded sample,
and Veronica decides whether to ignore, record, ask, notify, update a display or
dispatch more work.

Inference may remain on a local GPU, run on a Tailscale-connected private node,
or use a managed multimodal provider, subject to data-locality policy.