The Scope Analysis page supplements AI/ML IP search, trends, and IP overview by providing freedom-to-operate (FTO) and infringement-risk screening. Input a natural language description of subject matter of interest (e.g., product features, invention disclosures, draft claims, etc.) and run a semantic comparison against independent claims across patents and published patent applications in the SynapseIP database. Application claims are shown as published and may change before grant. The closest matches are returned with context-rich analyses.
Traditional search tools require exact keyword matches. Scope Analysis dives into claim language, which is paramount in determining infringement exposure. Each independent claim in the SynapseIP database is embedded and indexed. Features:
This workflow meaningfully reduces the cost and time of a formal infringement assessment.
Note: Start with 10–20 closest claims for rapid results. Increase # of claim comparisons input to broaden coverage.
Draft a description or list that captures subject matter of interest, inventive concepts, implementation details, etc. Including language tied to technical components is recommended for best results.
Use the '# of claim comparisons' input to specify the number independent claims to be returned. Default is 15; expanding to 40-50 can be useful where an initial scope analysis run shows high risk. Check 'Issued patents only' to exclude published applications, whose claims may still change before grant.
Click 'Run scope analysis' to execute embeddings search + KNN graphing operations. Results are returned with calibrated proximity scores, graph positioning, and risk tiles tailored to that query.
The graph displays nodes representing independent claims and graphically presents their distances from the input. Hover nodes to preview claim snippets, click to highlight a specific patent.
In the table, click any claim cell to expand the full text. Patent numbers link to Google Patents to view full documents.
Results table can be exported as a PDF document for offline reference and review.
Radial layout. The input text sits in the center. Nodes closer to the center represent higher proximity (lower cosine distance). Radius is linear in the calibrated proximity score, so a node halfway out is halfway between identical claim language and the corpus baseline.
Tooltip previews. Hover any node to see the patent title and first 200 characters of the matched claim.
Selection sync. Graph, summary tiles, and claim text are synchronized. Clicking a node or claim row highlights both views.
Proximity column. Proximity is calibrated against the corpus: 100% means claim language identical to the input, and 0% means no closer than a claim picked at random from the database. Because every claim in the corpus is an AI/ML claim, all claims share heavy boilerplate, so a raw cosine similarity would read high even for unrelated claims; calibration removes that floor. Scores ≥ 85% indicate near-duplicate claim language (high risk); 65–84% indicates a claim substantially closer than a typical one (moderate); below 65% is near the corpus baseline and generally lower risk, though still worth reading.
Expandable claim text. Click the claim snippet to read the entire independent claim text inline.