[ reference · glossary ]

AI deployment and MPS glossary

These terms describe boundaries, not guarantees. Their meaning depends on the data flow, operational conditions, and evidence supplied for a particular system.

On-device AI

Inference or related processing executed on an end-user device. It does not necessarily mean no data leaves the device.

Cloud AI

Inference or related processing performed on remote infrastructure accessed over a network.

Hybrid architecture

A design that assigns different steps to local and remote systems, ideally with an explicit data and escalation boundary.

Latency

Time from a user action to a useful result. It can include device work, network transfer, queueing, retrieval, inference, rendering, and retries.

Resilience

Ability to continue an acceptable function through a disruption. It requires stated conditions and fallbacks; it is not the same as independence.

Telemetry

Operational or analytics data emitted by software, including events, diagnostics, and sometimes metadata about use.

Device eligibility

The hardware, OS, storage, memory, accelerator, and operational conditions under which a feature is supported.

Provenance

Information that helps a reader inspect where a claim or artifact came from, its status, limitations, and review history.

MPS

The Maha Provenance Standard: a claim-level tagging framework for AI-assisted nonfiction. It is not a truth certification.

Preflight

MPS’s separate private automated claim-review workflow. It is triage, not human certification or primary-source verification.