Methodology
How governed prior-art search runs work.
Serious patent work needs more than a fluent answer. A useful AI-assisted search should show its objective, boundaries, evidence, limitations, cost trail, and review state in one inspectable packet.
Run anatomy
From practitioner intent to review packet
The workflow keeps language models in the parts of the job where they help most: planning, synthesis, and explanation around retrieved evidence. Retrieval controls, cost gates, and source boundaries stay explicit.
Scope the request
Practitioner instructions are normalized into a search objective, target facts, exclusions, jurisdictions, outputs, and review expectations.
Quote before execution
Live work is quoted before retrieval starts, so spend, workflow type, and output expectations are explicit before a matter moves.
Search bounded branches
Server-owned retrieval branches keep search logic controlled while cost gates limit broad corpus work before source evidence is fetched.
Package evidence
Candidate references are paired with source passages, family context, coverage notes, and review status instead of unsupported summaries.
Rerank consistently
Semantic review is used after narrowing, with embedding comparisons kept in the same space and metadata handled as filters, not filler text.
Expose the ledger
The report carries limitations, run events, cost context, candidate dossiers, and version history so reviewers can inspect the path taken.
Evidence before answers
What a reviewer should be able to inspect
The useful question is not whether an AI system can sound confident. It is whether a partner, associate, client, or security reviewer can see what happened and decide what still needs human judgment.
Walk the sample workflow- Search objective
- The precise question the run was meant to support.
- Target dossier
- The public or approved-matter record being searched around.
- Candidate families
- Grouped references with family and self-art risk called out.
- Evidence passages
- Source snippets tied to the candidate, not free-floating assertions.
- Coverage limitations
- What was searched, what was skipped, and what needs attorney follow-up.
- Run and cost ledger
- Quote, execution, regeneration, and export context kept with the work product.
Guardrails
Built for supervision, not blind reliance
ipstrategy.tech is designed for attorney-reviewable patent work: bounded inputs, visible cost controls, source-backed reports, and explicit limitations where a search run should not be overread.
- No confidential client facts in the public sample.
- Live searches require account setup, terms, data-boundary acknowledgement, and quote confirmation.
- Reports are support artifacts for attorney review, not legal opinions.
- Unsupported claims, complete-search guarantees, and black-box citation output are deliberately avoided.
Use the sample to judge the workflow.
A controlled pilot should test whether the packet helps real reviewers move faster with better supervision: useful references, clear limitations, predictable cost, and less mystery in the path from request to report.