A novelty report that shows its evidence line by line.

What lands on your desk

Seven numbered parts, and two appendices for the things most tools hide.

The report is built as a document with a table of contents, not a dashboard you have to screenshot. Sections appear only when there is something in them — scholarly citations and inventive-step combinations are conditional on the search actually finding them.

Appendix B exists on purpose. It lists references that were shortlisted but never mapped, so the difference between what was retrieved and what was analysed is a number you can see rather than one you have to trust.

How the search runs

Six stages, and only one of them is allowed to decide relevance.

Separating retrieval from screening from mapping is what makes the report auditable: each stage has one job, and the stage that routed a reference is not the stage that later argues about it.

1. 1

   Normalise the disclosure and build the query

   Stage 0

   Your invention is decomposed into discrete features and turned into retrieval vocabulary — including classification-aware and EPO-style keyword groups — so the search runs on the concepts in the invention rather than the words you happened to type.

2. 2

   Retrieve across patents and scholarly literature

   Stage 1

   Candidates come back from the patent corpus and from non-patent literature. Papers are carried through as first-class references, not footnotes — they get their own citation section in the report.

3. 3

   Screen for relevance, and record the routing

   Stage 1.5

   An adaptive screening gate sorts every candidate into direct, component, borderline, or rejected. This is the only stage allowed to make that routing call, and the counts it produces are what the report reconciles against.

4. 4

   Map every feature against every surviving reference

   Stage 3.5a / 3.5b

   Each invention feature is compared to each reference and given a status, an extent score, a confidence value, and a verbatim quote with its source location. No mapping is accepted without the evidence attached.

5. 5

   Write the per-reference remarks

   Stage 3.5c

   Each reference gets its own analysis — what it teaches, where it stops, and what that means for claim scope — rather than a single blended summary that hides which document did the damage.

6. 6

   Assemble claim positioning and the risk view

   Stage 4

   Claim-positioning observations are derived strictly from the mapped overlaps above, then combined with novelty risk, inventive-step combinations, and a declared confidence level for the report as a whole.

Reference routing

Every reference is labelled, including the ones that were thrown out.

A search that only shows you its hits is asking you to trust its filter. Each candidate keeps its routing label through to the report, and the counts reconcile: searched, retrieved, reviewed, visible, analysed.

Feature classification

Not every feature deserves to count toward your novelty.

Features are typed before they are mapped, and the risk headline is computed from the core ones only. A boilerplate feature that no reference discloses is not evidence of novelty, and the report refuses to present it as such.

Inventive step

The problem is rarely one document. It is two.

When no single reference covers the core features but a pair does, the report constructs the combination explicitly — what A teaches, what B adds, the apparent motivation to combine, and what the combination still does not reach.

Landscape

Who else is working in this space, and who keeps showing up.

The same search that finds your prior art already knows who filed it. The report groups applicants and assignees, and flags inventors and entities that recur across the results — the signal that a competitor is building a portfolio rather than filing once.

What it will not do

The report states its own confidence, and its own limits.

Four confidence values are recorded separately, because they fail for different reasons — a strong retrieval with weak feature mapping is a different problem from the reverse, and collapsing them into one number hides which one you have.

Novelty is determined by the Controller, not by a search. The report closes with a limitations section and concrete next steps, and every conclusion in it is framed as analysis for a qualified attorney to accept, narrow, or reject.