Data licensing guides and research

A practical guide for companies and creators: find what you own, check the rights, de-identify, price it, choose a route, license it and get paid.

Data monetization used to mean selling insights. Now it also means licensing the data itself to AI developers. See the models, the prices and the risks.

Understand license scope, ownership evidence, access, privacy, deletion and approval before a data licensing discussion. Preparation, not legal advice.

Compare AI training data sources, provenance, rights, quality and permitted uses without treating a dataset description as a guarantee.

Learn which rights, scope, quality, access and evidence factors affect a data estimate. An indicative range is not a quote or offer.

Use the glossary to distinguish rights, privacy and data-handling terms, including de-identification, anonymization and pseudonymization, before a licensing review.

A practical checklist for ownership, authority, privacy, client rights and evidence before a data licensing review. Not legal clearance.

Review Protobuf presence, oneof cases, defaults, and JSON or wire round trips before a schema handoff.

Review BSON datetime, integer, double, decimal, and timestamp types before interpreting exported documents.

Review ORC schemas, stripes, row indexes, statistics, and footer metadata before relying on file pruning.

Review Zarr shape, data type, chunk grid, key encoding, fill value, and codec metadata before reading arrays.

Review WARC records, capture links, HTTP payload boundaries, dates, digests, and truncation evidence.

Review mbox separators, message headers, MIME boundaries, and attachment parts before extracting mail.

Review Git trees, submodule commits, LFS pointers, tracked objects, and export evidence before handoff.

Review glTF buffers, buffer views, accessors, strides, alignment, counts, and external resource paths.

Review DXF units, WCS coordinates, group codes, entity boundaries, and extrusion metadata before import.

Review FITS HDUs, mandatory cards, BITPIX, NAXIS, padding, and array metadata before reading pixels.

Review PDF file specifications, embedded-file name trees, attachments, streams, and redaction evidence.

Review EPUB package manifests, spine order, href resolution, media types, fallbacks, and linked resources.

Separate missing observations from resampled values with a time-series gap review worksheet.

Review telemetry counter drops for resets, wraparound, overlap, and lost intervals.

Review effective dates, keys, and version evidence before joining reference data to records.

Trace measurement units, conversion factors, precision, and source lineage before comparing values.

Separate skip-pattern non-applicability, item nonresponse, and unusable answers in survey exports.

Review ranking click logs for exposure, position, selection bias, and missing context.

Define impression, click, eligible, and deduplicated denominators before reporting rates.

Review trajectory timestamps, sampling gaps, coordinate order, and interpolation assumptions.

Record subset predicates, source version, exclusions, ordering, and counts without sharing raw rows.

Review PDF page sequence, labels, logical structure, inserts, and missing-page evidence.

Review dataset fingerprint inputs, manifest ordering, mutable sources, and snapshot freeze evidence.

Review BCP 47 language tags, script and region subtags, mixed-language spans, and unknown labels.

Review Arrow IPC stream and file boundaries, schemas, dictionaries, record batches, metadata, and versions before reading batches.

Review SQLite database, WAL, SHM, checkpoint, snapshot, and schema evidence before treating an export as consistent.

Review CSV cells that may become spreadsheet formulas, preserve literal text, and document neutralization before export.

Review table grain, join multiplicity, pre-aggregation, and null behavior before comparing or aggregating tabular exports.

Review parent-child references, orphan handling, snapshot alignment, and bounded samples before relying on linked dataset exports.

Review binary byte order, field widths, alignment, padding, and record strides before interpreting packed exports.

Review multiline log framing, timestamps, sequence fields, and structured log records before counting events or sessions.

Review timezone identifiers, tzdata versions, offsets, ambiguous local times, and conversion evidence before comparing timestamps.

Review raw names, Unicode normalization, separators, dot segments, case rules, and collision evidence before matching files.

Review GeoTIFF CRS keys, raster transforms, units, and NoData handling before interpreting geospatial imagery.

Review LAS and LAZ headers, point formats, scales, offsets, CRS records, and counts before reading point data.

Review netCDF dimensions, coordinates, calendars, units, fill values, and ranges before interpreting variables.

Review HDF5 dataset paths, dimensions, types, chunks, filters, and attributes before reading arrays.

Review SQL dump structure, executable statements, schemas, COPY sections, and sensitive-field boundaries before restore.

Review XML namespace bindings, entity declarations, external references, and parser policy before processing an export.

Review spreadsheet formulas, cached values, hidden sheets, defined names, and links before reading displayed data.

Version label meanings, mappings, deprecations, and provenance without rewriting annotation history.

Separate agreement metrics, disagreement patterns, abstentions, and resolution evidence in annotation review.

Review split units, duplicate relatives, transforms, and group boundaries before interpreting evaluation results.

Reconcile bounding-box pixels, origins, units, transforms, and edge rules before comparing annotations.

Review mask geometry, pixel encoding, class maps, background states, and transform history as separate evidence.

Link audio, transcript cues, collection records, and consent evidence without turning gaps into legal conclusions.

Connect OCR text, page geometry, layout boxes, reading order, and processing evidence before interpreting a document export.

Trace synthetic-data sources, generation stages, coverage checks, and limitations without assuming privacy or representation.

Inspect ZIP and tar member names, links, normalized destinations, and special types before extraction.

Separate compressed, declared, observed, and retained bytes before expanding an archive or stream.

Compare filename hints, MIME declarations, file signatures, and decoder stages without a single verdict.

Reconcile API page links, cursors, ordering, duplicates, gaps, and terminal evidence before calling an export complete.

Preserve access, authority, contracts and archive evidence before a company closes or loses control of its systems. Not legal advice.

Preserve history, issues, attachments, checksums and exclusions before an archive or repository moves. A preparation checklist, not a completeness guarantee.

Check ownership, exposure, personal data and exclusions before a private archive enters a licensing review. Not legal clearance.

Move archive access, authority records and evidence together when control of a company changes. A continuity guide, not legal clearance.

Find vendor, platform and customer terms that may restrict a data license before you prepare a sample or ownership pack.

Record source, rights, consent, transformations, exclusions and evidence dates in a reusable data provenance log.

Document dataset contents, coverage, provenance, rights, limitations and permitted uses before a licensing or buyer review.

Assess coverage, quality, provenance, rights, exclusions and buyer-readiness for an AI training dataset. A scorecard, not a guarantee.

Map property records by source authority, geography, personal data, upstream terms and intended use before a property data licensing review.

Map authored, customer, contributor, licensed media, comments and revisions before deciding whether a CMS export fits an AI data review.

Verify an export with a byte-level file manifest, SHA-256 checksums, and a clear boundary between integrity and rights review.

Make every dataset release explainable with stable version identities, prior-release links, change classes, and unresolved review states.

Normalize export timestamps without losing source values, offsets, named time zones, or ambiguous and unknown states.

Catch decoding failures and malformed CSV rows with an explicit encoding contract, strict parsing, and reproducible issue records.

Review JSON Lines exports one physical line at a time, with bounded diagnostics for encoding, syntax, and record-shape failures.

Review Parquet footers, logical types, row groups, statistics, and directory partitions before reading sensitive rows.

Compare Avro writer and reader schemas directionally, recording promotions, defaults, aliases, unions, and open semantic questions.

Review GeoJSON longitude and latitude order, geometry nesting, ring closure, bounding boxes, and spatial edge cases.

Check exported table keys, nulls, join cardinality, and orphan relationships before a sensitive dataset review.

Design incremental exports with explicit watermarks, tie-breakers, overlap rules, and reconciliation evidence.

Preserve deletion events, tombstones, keys, source positions, and unresolved states without treating absence as proof.

Review event order, replay behavior, stable identities, and idempotent loads without confusing delivery order with business time.

Review audio sample rate, channel layout, and bounded clipping observations without turning stream metadata into a broad quality score.

Compare decoded frame timestamps with stream rate summaries to review candidate variable timing, gaps, duplicates, and discontinuities.

Separate image pixels, EXIF orientation, ICC profiles, and metadata exposure before creating a normalized derivative.

Review WebVTT cue order, duration, speaker tokens, overlap policy, and alignment to approved reference anchors.

A practical worksheet for preserving decimal precision, declared units, scale, and rounding rules in numeric exports.

A practical review guide for preserving identifier width, leading zeros, lexical form, and join behavior across exports.

A practical guide to preserving Unicode source text, choosing normalization, and detecting identifier collisions.

A schema-focused guide to keeping null, missing, empty, and default states distinct in JSON exports and patches.

Define table grain, keys, types, missingness, labels, and quality checks before a sensitive export is reviewed.

A staged export sampling checklist using manifests, schemas, synthetic fixtures, and approved redacted artifacts.

Separate exact and near-duplicate records with documented rules, privacy limits, and review decisions.

Distinguish unknown, not applicable, withheld, and empty values, then test labels against a documented codebook.

Compare public code-licensing records by source, rights shape and evidence limits. Public examples do not set a price for your repository.

Read public workplace-archive records with attention to source, rights, privacy and value limits. Public examples do not prove demand.

Compare footage, images and books by rights, consent, provenance and permitted AI uses before discussing a media license.

Compare speech and audio collections by consent, labels, provenance, privacy and permitted use. Do not treat minutes alone as value.

Review public health-data AI program status and rights questions, not a market promise or intake route. Rightmark does not accept health data yet.

Compare robotics and sensor data by episodes, signals, calibration, failure cases, provenance and rights. No scarcity or price claim is implied.

Cite public data-licensing records with source, date, status, rights shape and disclosed or estimated value type.