Our method

A finding is useful when its evidence holds up to inspection.

The search is directed by an engineer, carried out with AI agents, and reviewed before delivery. The output records both supporting evidence and remaining gaps.

Define the question.

Agree the claims, relevant dates, known references and technical issues. Identify what the team already has and where a deeper search would help.

Follow the disclosure.

Search patents, papers, standards, archived manuals and implementation history as the matter requires. Use the terminology of the relevant era and related engineering fields.

Read and map the sources.

Match passages to individual limitations. Retain original-language materials and label machine translations. Record publication evidence and keep known references distinguishable.

Check the record.

Check quotations against retained source text or page images and verify pinpoints. Human review and a separate fresh-model QA pass examine the mapping and its unresolved points.

Deliver the evidence.

Provide the chart, supporting references, verification record and a clear account of coverage, alternatives and material read but not relied on.

Example search record

Broad discovery. A visible review trail.

See what was found, what was assessed and what remained outside the completed screening.

Raw search hitsRows across 22 search lanes

465,048

Screening pool44,614 patent families + 3,877 non-patent works

48,491

First-round screeningRecorded screening decisions

7,200

JudgedCandidate judgments recorded

ChartedReferences included in the chart

Not yet screenedOutstanding pool recorded in the snapshot

42,075

Logarithmic bar scale. Snapshot supplied by Zeusfyi. Search hits include overlapping records; stages retain different counting units and are not a single conversion cohort. This example is not attributed to a named patent.

Full snapshot context

Found: 45,070, with 4,069 non-patent lane records. Ranked: 44,614, with 3,877 non-patent works ranked with text. First-round decisions: 5,972 standing keep, 268 standing relevant, 960 standing drop. Evidence-round screening: not run / not recorded. Judged: 34, with 33 candidates kept by the reader.

Read the scores correctly

Strength belongs to a specific mapping.

A source can strongly support one element and leave another arguable. A combination can depend on a technical inference, date evidence or a contested reading. The assessment records those dependencies.

Scores are internal research judgments. They do not measure a probability of invalidity, replace counsel’s analysis or turn partial coverage into a complete claim mapping.

Explore the scored example

Start with the technical question

Have a patent and a deadline?

Tell us the patent, claims of interest and timing. We will confirm the scope, fixed fee and delivery date before work begins.

Discuss your matter