The Ingestion Verification Protocol: Does Your AI Actually Read Your Documents?
Video overview of SF0038: Ingestion Verification Protocol
About this video
You upload a document. You tell the AI to read it. It says it did. You assume the content has been fully processed and move on.
That assumption is wrong -- and it is one of the most common and least understood failure modes in AI-assisted work.
This video introduces the Ingestion Verification Protocol, a four-step method for making AI document processing observable and verifiable. Instead of trusting the AI's self-report, the protocol requires the system to show its work -- section by section -- so that processing can be checked before any downstream work begins.
The four steps
The protocol proceeds through four stages. First, scope definition establishes exactly what the document is, what the use case requires, and what level of detail is needed. Second, iterative ingestion breaks the document into segments, requiring the AI to produce a structured summary and a verbatim checkpoint after each one -- transforming an opaque process into an observable record. Third, adequacy adjudication requires the human operator to review those summaries and decide whether processing was sufficient. This step cannot be delegated back to the AI. Fourth, an optional spot-check provides additional confidence by asking targeted questions that can only be answered correctly if the document was genuinely processed.
Further reading
The full technical treatment of the Ingestion Verification Protocol is available as a published research paper with DOI. The companion Practitioner Guide covers a minimal working sequence for everyday document verification.
- SF0038: Ingestion Verification Protocol -- full methodology, formal definitions, and implementation guidance. DOI: 10.5281/zenodo.18289047
- PG-003: Verify Before You Work -- a minimal copy-ready prompt sequence for confirming document ingestion before any session begins
- PG-002: AI-Assisted Editing Without Silent Loss -- practical application of IVP principles to document editing workflows
- SF0039: Context Representation Drift (CRD) -- the related failure mode describing how constraints lose influence during extended interaction
Full framework documentation available at the Synthience Institute community on Zenodo.