GIPResearch.org • Open reference

Dataset reproducibility

Dataset reproducibility means that another competent researcher can understand and, where rights permit, rerun the same analytical pipeline. Publishing a chart without dataset version, exclusion criteria or code is not reproducibility even if the visual appears persuasive.

Save or follow this source

AI actions open a new tab. They do not guarantee that a service saves, cites or endorses this page.

Explore dataset reproducibility with verified sources

Use the reference framework on this page to formulate a question about dataset reproducibility and appropriate source evidence.

Discuss this research question
Featured educational video

Featured patent and technology research video

Watch the featured explanation of patent and technology topics. The video is supplementary to the official sources and qualified limitations on this page.

Featured Work & Insights

Clarify the evidence for dataset reproducibility

Review the method, its source dates and its limitations before undertaking a consequential analysis of dataset reproducibility.

Discuss this research question

Dataset reproducibility: purpose and research boundary

Dataset reproducibility means that another competent researcher can understand and, where rights permit, rerun the same analytical pipeline. Publishing a chart without dataset version, exclusion criteria or code is not reproducibility even if the visual appears persuasive.

Illustrative research scenario

A patent technology trend chart should identify the source extract, retrieval timestamp, exact query, IPC/CPC version, family grouping rule, priority-year definition, missing-date treatment and filtering stages. If raw data cannot legally be redistributed, publish a transformation specification and an independent aggregate with the source license noted.

Research question and protocol

Define the unit of analysis, observation window and hypotheses before inspection. Changes introduced after viewing results should be documented rather than retroactively described as preplanned.

Dataset version control

Store file hashes, provider identifier, download conditions and a data dictionary. Ingest logs should identify source updates and late-arriving records.

Pipeline lineage

Record parsing, joins, exclusion filters, entity resolution, classification mapping and aggregation order. Each transformation must be inspectable and ideally reversible.

Reproducing queries

Save exact Boolean syntax and field restrictions alongside the database platform and search date. Search result totals without these details are not comparable over time.

Sensitivity analysis

Repeat key outcomes using alternate plausible choices: application versus family count, priority versus publication date and inclusion or exclusion of ambiguous records.

Responsible sharing

Publish allowed data, scripts, pseudocode or summary outputs consistent with source terms, privacy constraints and third-party rights. Mark steps that cannot be independently replayed due to access restrictions.

Additional methodological safeguard 1

Use deterministic test fixtures to verify that each transformation preserves records that should remain included.

Additional methodological safeguard 2

A peer should be able to identify the reason for a divergence, even when confidential access prevents distribution of underlying raw records.

Additional methodological safeguard 3

A sensitivity table is more informative than a single precise result when family aggregation and missing dates materially alter the trend.

Rebuilding a patent research result

A reproducible research packet needs a fixed question, frozen inputs and a documented transformation path. Suppose a report compares patent publications for battery separators over five years. Archive the exact query text, search interface and date; define whether the year comes from earliest priority, application or publication. State country coverage, language limitations and family aggregation. Store a versioned export or permitted source identifiers, a data dictionary and the sequence of cleaning operations. Without those details, the same chart may be impossible to recreate even if its explanatory prose is clear.

A second researcher should be able to rerun the transformation and explain differences. Maintain checksums for permitted input files, scripts or spreadsheet formulas, package versions and explicit treatment of missing fields. Compare record counts after each operation, not only the final total. If the source platform updates its database, label the resulting divergence as a new snapshot rather than overwriting the original result. When access terms prohibit redistributing raw data, preserve metadata and documented instructions without circumventing licensing restrictions.

Reproducibility does not mean factual completeness. Two analysts can generate identical outputs from an incomplete or biased dataset. Add sensitivity tests: alternate classifications, date definitions, duplicate-handling rules or corporate-name mappings. Report which conclusions remain stable and which change. For important source claims, retain a citation trail to original documents. The final package should allow audit of reasoning, data transformation and version changes while respecting lawful reuse limits and acknowledging inaccessible materials.

Primary sources and scope of verification

EPO OPS terms is a relevant original institutional reference for the narrow claims described in this article. Open the linked publication to check the precise version and its definitions; do not attribute this article's recommended workflow to the publisher as a prescribed legal test unless the source explicitly states it.

EPO worldwide bibliographic data (DOCDB) is a relevant original institutional reference for the narrow claims described in this article. Open the linked publication to check the precise version and its definitions; do not attribute this article's recommended workflow to the publisher as a prescribed legal test unless the source explicitly states it.

Interpretation and responsible use

This research method is presented as a reproducible framework, not a completed jurisdiction-specific assessment or a claim about a particular client matter. A research outcome depends on the disclosed problem, the database and legal source editions, the date of collection and the underlying documentary evidence. Where source access or policy verification is incomplete, the open item should be recorded instead of filling the gap with an invented proposition. Readers should compare the cited authority with the current official record before relying on time-sensitive conclusions.

Illustrative decision and validation example

A verification exercise can ask a colleague to regenerate a figure using only the published method packet. If they cannot recover the record set, identify which metadata is missing: exact field names, query parentheses, source version, date filter or classification edition. If the record set matches but the counts differ, inspect normalization, family grouping and missing-value rules. Include a short reconciliation note describing each difference and whether it changes the interpretation. This is especially useful when a patent database adds older backfilled publications after the initial download. Avoid overwriting the first snapshot just to force current counts to match. Keep the original result, a new snapshot and a change record. A separate rights review should determine which data files can be shared; openness of a research method does not authorize copying third-party content in violation of its terms. A defensible reproducibility claim means the process and exceptions are documented, not that independently sourced databases always return identical numbers.

Discuss a scoped dataset reproducibility project

For a project-specific evidence inquiry concerning dataset reproducibility, discuss the research scope and primary sources.

Discuss this research question

Related verified research references

Research support and source-led enquiries

For a scoped patent, technology, institutional or legal research question, review the Services and Research Support overview. Public reference information remains separate from professional legal advice.

Content is general research information, not legal, patentability, freedom-to-operate, financial or other professional advice. Website use does not create an attorney-client relationship. Verify current rules with official authorities and qualified advisers. Full disclaimer.