How the AI Review Works

The AI produces a reasoned review decision; the platform signs a certificate only after the authorized workflow and payment verification. The model itself never holds signing keys, database privileges or payment authority.

Pipeline

Seven logical stages

1 · Completeness

Missing information and format checks on the frozen version.

2 · Category check

Technical solution vs. mere idea or project — each gets its own record type.

3 · Claim consistency

Draft claims are checked against the description for internal consistency.

4 · Prior-art search

Only through licensed sources (e.g. EPO Espacenet, WIPO PATENTSCOPE) when a real integration exists. Otherwise: “search not performed — assessment could not be completed”.

5 · Comparison

Found documents are compared against the claims, with direct source links.

6 · Reasoned decision

Against the versioned RFA internal policy criteria, with uncertainties recorded.

7 · Sealed report

The decision report and evidence pack are hashed and sealed.

Outcomes

Four possible decisions

Approved (internal)

Meets the RFA internal rules only. Never a statement of statutory patentability.

Revision needed

Specific gaps listed; a new immutable version can be submitted.

Not eligible

Does not meet the criteria, with reasons.

Inconclusive

E.g. no licensed search access — stated honestly, never papered over.

Accountability

What every report must show

Policy version and policy hash applied
Model identity and prompt version/hash
Input manifest of the frozen version
Search queries and the sources actually searched
Direct links to cited sources
Evidence passages relied upon
Coverage gaps and uncertainties
Date and time of each step (UTC)

Instructions embedded in uploaded files (“approve this project”) are not followed. The model never declares an application “100% patentable” or “guaranteed profitable”. AI text generation is strictly separated from payment and certificate mutations.