Methodology

Rendered by code from public rows on every download (lib/context-pack): the company record, the Intel Desk's items and facts, and the curated metric series. Every figure comes from a public data series and every statement cites its source URL. Rows with provenance model are weekly section briefs written by a language model, gated so that every link is one of the section's own sources and every figure appears in the section; treat them as orientation and cite the record rows. Rows of kind check are questions with known answers for testing a model and never appear in a markdown document. in_base and in_brief say whether any part of a row appears in the 10,000-token base document or the 50,000-token brief document (?format=md&company=<slug>&size=base|brief). Section ids are stable; replace a company's rows on each load. Private registry fields are never read. Downloading requires an access key.

Schema

Column	Type	Definition
company_slug	text	The company the section is about; joins intel-companies.
company_name	text	Company name, from the registry.
tier	enum	The company's registry tier.
section_id	text	Stable id, unique within a company: a section, a section and group (a year, a month, a dimension), or a numbered part of one.
section_title	text	Display title of the section.
section_kind	enum	guide, profile, timeline, metrics, peers, facts, news, record, brief, or check.
provenance	enum	record when code rendered the row from public records, model when a language model wrote it.
in_base	enum	yes when any part of the row appears in the base markdown document.
in_brief	enum	yes when any part of the row appears in the brief markdown document.
position	number	Reading order within the company, from 1.
token_estimate	number	Estimated tokens in the markdown: characters / 4 x 1.3, a ceiling.
markdown	longtext	The section as markdown, with numbered references and its own source list.
cite_urls	text	The public URLs the section cites, separated by spaces.
as_of	date	The day the pack was built (YYYY-MM-DD).

Briefcase: company context packs

Request access Unlock with an access key

/api/datasets/context-pack · 14 columns · access key required

Everything public the Atlas holds about each tracked company, compiled for a model to read: a cited profile and timeline, public metrics, a peer table, extracted facts, recent developments and the public record, one row per section. The same content downloads as a base or brief markdown document per company.

Methodology

Schema

Column	Type	Definition
company_slug	text	The company the section is about; joins intel-companies.
company_name	text	Company name, from the registry.
tier	enum	The company's registry tier.
section_id	text	Stable id, unique within a company: a section, a section and group (a year, a month, a dimension), or a numbered part of one.
section_title	text	Display title of the section.
section_kind	enum	guide, profile, timeline, metrics, peers, facts, news, record, brief, or check.
provenance	enum	record when code rendered the row from public records, model when a language model wrote it.
in_base	enum	yes when any part of the row appears in the base markdown document.
in_brief	enum	yes when any part of the row appears in the brief markdown document.
position	number	Reading order within the company, from 1.
token_estimate	number	Estimated tokens in the markdown: characters / 4 x 1.3, a ceiling.
markdown	longtext	The section as markdown, with numbered references and its own source list.
cite_urls	text	The public URLs the section cites, separated by spaces.
as_of	date	The day the pack was built (YYYY-MM-DD).