Evidence and Source Verification Policy

Research context and practical verification

Evidence levels depend on the question being asked

No single type of source is authoritative for every proposition. A claim about a patent-office filing rule should begin with the issuing office's current published material. A claim about statutory language requires the relevant law, and an account of a court holding requires the judgment with its applicable jurisdiction and procedural context. A patent document may establish what was disclosed at a publication date, but cannot by itself prove that an invention is commercially adopted or legally enforceable today. For empirical technology claims, original technical papers and well-described evaluations may be more suitable than press releases.

Official classification references: WIPO IPC; WIPO IPC 2026 Guide; CPC portal. These sources are versioned; verify the current scheme before applying a specific symbol.

The evidence register should capture the exact proposition and the reason a source is probative. It should flag whether the material is primary, secondary, self-reported or a research hypothesis. Multiple secondary summaries repeating the same underlying assertion do not constitute independent confirmation. Readers should be able to distinguish a direct observation in a public record from editorial synthesis, and distinguish both from a forecast or an untested commercial opportunity.

Patent records require record-level precision

Patent literature includes applications, publications, grants, family relationships, claims, drawings, classification symbols and legal-status records. These are related but not interchangeable. A family search may group publications from different jurisdictions, while a forward-citation count can change depending on database coverage, duplicate handling, citation source and cutoff date. An examiner-cited reference may have a different evidentiary role from a reference listed by an applicant. A claim of technical influence therefore needs more than an undifferentiated total citation number.

Searches should record databases, query strings, classification filters, assignee normalization choices, date boundaries and language constraints. A document discovered in an index may require verification against an official publication or prosecution file before a legal conclusion is drawn. The existence of an unexamined patent application should not be described as an enforceable patent right. Similarly, an apparent lapse or expiration from a database should be checked using relevant official legal-status data before use in a freedom-to-operate decision.

Classification version and mapping confidence

WIPO publishes the International Patent Classification and describes its hierarchy in an official Guide. The IPC changes through versioned revisions, so a page should identify the scheme version used when presenting a specific classification relationship. CPC is a related but distinct system maintained through its own authorities and revision practices. A technology name on the site's readable taxonomy may map to several classes, subclasses or groups, and a CPC symbol may cover multiple adjacent technologies. An appropriate reference makes both the technical rationale and any uncertainty visible.

Classification confidence should not be stated as an invented numeric percentage. Instead, editors can use descriptive statuses such as verified against the official definition, provisional candidate requiring technical checking, disputed mapping or historical symbol. Search methods should combine conceptual terms, synonyms, functionality and classification coverage rather than assume a class search alone is exhaustive. If a symbol has moved or a definition changed, the older version may still matter when interpreting historical documents; a static mapping must not silently erase that context.

Comparing official databases without confusing their scope

Different patent-office portals expose different collections, metadata fields, query tools and historical coverage. WIPO PATENTSCOPE is particularly relevant to international PCT publication records and connected collections; the USPTO and EPO maintain their own official search and procedural resources. A third-party index may offer convenience or analytics that an official site does not, but its coverage and update cycle must be documented before results are compared. A global patent count is not meaningful without defining the unit: application, publication, grant, patent family, assignee, technology class or another record type.

A transparent comparison uses consistent counting rules where possible and explains where harmonization is not possible. For instance, an applicant name can change through corporate transactions, translations and inconsistent data entry. Counting a merged corporate group as one owner without explaining that aggregation changes the question. The appropriate output is a table describing the fields, coverage assumptions, processing steps and limitations, not an apparently exact ranking detached from source methodology.

Legal interpretation requires a separate reasoning step

An official rule may be clear while its application to a particular technology remains disputed or fact-sensitive. Evidence policy therefore separates quoted statutory or administrative text, the office's published examples, the author's interpretation and a recommendation to seek jurisdiction-qualified advice. A summary of a patentability standard cannot substitute for current claim analysis; a description of FTO search methodology cannot determine infringement across all markets. Technical reference pages must not imply that an IPC/CPC classification decides patent eligibility, novelty or validity.

Material uncertainty should be included close to the claim. If a legal condition differs among jurisdictions, a comparison should identify the controlling office, scope, date and any source language issue. If a public source does not resolve the question, the page should say so instead of filling the gap with an inference. This is especially important for regulatory subjects such as AI, digital assets, data protection and professional responsibility, where rapidly evolving guidance may differ from binding law.

Reproducibility and maintenance record

For significant tables and datasets, a research note should identify the date of collection, origin, available file format, filtering and normalization rules, exclusions, and limitations. Where licensing permits, readers may be offered source links or reproducible public-data steps; the presence of a public source does not automatically grant unrestricted republication rights. Original diagrams and extracted statistics must be presented with enough explanation to avoid misinterpretation. A comparison is stronger when another analyst can retrace the choice of records and produce a similar result.

Every substantive page should receive a review classification: evergreen; periodically changing; fast-changing; or event-driven. Review dates do not license cosmetic content changes. Corrections should target the particular source, proposition, table or procedure affected, and leave a trace of material revisions in the editorial registry. Users may report suspected errors through the contact channel, preferably supplying the exact URL, passage and an authoritative counter-source. The editorial team evaluates the claim independently; a correction request does not guarantee a particular response or conclusion.