Hoverview

AI audit trail

AI audit trail: answer where AI was used.

A donor, reviewer or project team may need to know whether a figure came from a source document, an AI proposal or a person. A general disclaimer cannot answer that question for a particular value or report.

Hoverview stores three distinct origins: extracted from a named source document, generated as a model proposal, or entered by a person. Confirmation and later edits stay in the project history. This makes AI use inspectable where the work is recorded, rather than reconstructed when a report is already due.

From source to confirmed record

AI use remains visible at every stage

The record preserves the difference between source material, model proposal, human decision and confirmed project fact.

  1. 1

    Source recorded

    A figure begins with a named source document or a person, so its starting point remains attributable.

  2. 2

    Proposal separated

    Extracted or generated AI output keeps its origin and remains a proposal rather than becoming a project fact.

  3. 3

    Human decision

    A named person accepts, edits or discards the proposal. AI does not approve a report or own a target or baseline.

  4. 4

    History retained

    Uploads, proposals, confirmations, edits and approvals remain in the append-only project history.

AI runs for a requested task

Hoverview uses AI when a user asks it to read a document, draft a logframe from the files you upload, suggest records, check what is missing or write a first version of a report. It is not running quietly in the background on everything else your team does.

What this is, and what it is not

The provenance record states what was extracted, generated, entered by a person and confirmed. It never asserts that a report satisfies a particular donor policy.

Information sent for AI processing

When AI is used, only the information relevant to the requested task is sent to an approved processing provider under commercial data protection terms.

The clearest example is the AI-assisted logframe builder: it proposes the problem analysis and drafts the logical framework, and a person still chooses the strategy and accepts each row. This page pairs with security and data protection for the wider privacy and data-handling principles.