# Data licensing preparation pilot

Created: 2026-11-17
Updated: 2026-11-17

## Question and decision

What should an organization prepare before its first conversation with a data licensing or supply program? The intended reader owns or manages an organizational dataset and needs to describe it accurately. This pilot tests whether publicly disclosed intake stages support a useful preparation brief. It does not measure provider performance or recommend a provider.

The ten programs in `observations.json` were selected before the main evidence refresh to cover different visible intake types in an earlier supplier-route directory review. Selection is purposive, not random: direct lab interest, marketplaces, an analyst-reviewed directory, content licensing, operational business data and industrial archives. The earlier directory is not a market census. No previous scores, account activity, private data or subjective rankings enter this dataset.

## Collection

Research date: 17 November 2026, Asia/Bangkok. Sources: official English public program pages, directly relevant public explanations and publicly extractable intake labels. The web retrieval tool may use indexed copies; this is a dated reading of accessible source material, not a provider-certified live audit. The OpenAI explanatory article was originally published on 9 November 2023; it is not treated as a new announcement.

Scope stops before entering personal information, creating an account, accepting an agreement, uploading a sample or submitting anything. No such actions were performed. Each record identifies the specific inspected sources and its observation boundary. Additional public explanatory pages were used where the entry page did not expose the application. Depth differs across providers, so disclosure counts would not be comparable scores.

Turing Lazarus returned no usable program body through the extractor. The General Data Company homepage was readable, but the partnership route was not. These two cases stay in the fixed sample as unobservable for this question; neither is classified as closed, unavailable to people, or lacking an intake. Protege's public program and contact route were readable, while its embedded fields were not. Neudata's embedded form was not extracted; its public page described the company/contact request and subsequent stages.

## Coding

One record represents one program. `observation_status` is `substantive` when at least one current source supplies meaningful intake or preparation information; `unobservable` otherwise. A substantive record does not mean a complete application was observed.

`first_action` describes the advertised entry: `interest_form`, `account_creation`, `provider_application`, `listing_request`, `partner_form`, `meeting_request`, `estimate_or_contact`, or `unknown`. It is not a claim that the action was completed. `preparation` contains concise paraphrases of explicit source material; `later_stage` preserves a provider's stated sequence. `limitations` records missing visibility. Field labels are not validation evidence, and provider promises are not independently observed outcomes.

The synthesis groups preparation into company context, dataset characteristics and permission/scope. These are author-created organizing categories. They are a practical interpretation of the observations, not a validated scale. Unknown information is not coded as zero or as an absence of a requirement.

## Reproduce

Run `python3 analyze.py` from this folder. It reads only `observations.json` and produces `results.json` and `evidence-table.md`; it makes no network calls and has no third-party dependencies. The assertions check the fixed sample, dates, permitted action categories, source references and the boundary between substantive and unobservable records. The resulting 8/10 count describes readable evidence in this pilot only; no confidence interval or market percentage is appropriate.

Reproducing the analysis from the supplied observations is deterministic. Reproducing collection requires revisiting the listed URLs, whose contents may change. The download intentionally contains authored factual observations and source pointers, not redistributed full webpages, screenshots, account records or the private research corpus. Copyright in linked source material remains with its owners. Sapien1's own article, worksheet and observation files may be used as the released site's materials; this package grants no rights over source datasets or third-party content.

Run `python3 render-assets.py` to reproduce the two original SVG figures. The entry-point exhibit uses the observation records; the preparation infographic presents the author-created synthesis. Both contain editable text, titles and descriptions.
