Methodology

A focused collection designed to grow

The current starting roster includes six organizations, four normalized issues, eight public officials, five bills, and the outside firms identified in the retained filings. This roster defines the bounded retrieval used for the current edition. These subjects were chosen for record variety, source availability, and explanatory value. Future editions can add entities, periods, record types, and relationships while applying the same identity, evidence, and review standards.

Normalization, provenance, and deduplication

Raw fields enter typed records while source wording, record ID, URL, reporting period, retrieval date, and evidence label remain attached. For LDA reports, the canonical key uses client, registrant, filing year, and quarter. The latest posting wins; a tied amendment wins; filing ID breaks a final tie. This prevents superseded versions from contributing twice.

Aliases, former names, affiliates, and subsidiaries

Matching compares the complete source-record name with an approved exact alias, case-insensitively. It never uses fragments alone. A former name needs dated evidence and does not automatically absorb an affiliate, subsidiary, committee, or separately named client. For example, BOEING COMPANY may match the reviewed Boeing profile while BOEING EMPLOYEES' CREDIT UNION remains excluded; Meta's former-name note does not pull in every Facebook or subsidiary filing.

Issue classification

Each normalized issue has an explicit allowlist of official LDA codes. A canonical filing counts once per matching code even if the code repeats, but one filing may count under several codes. Thus code totals overlap and cannot be summed as a unique filing count. Original descriptions remain visible because a broad code such as SCI, CPI, or TEC does not by itself establish that the text concerns AI or privacy.

Amounts and nulls

Self-reported lobbying expenses and outside-registrant income are separate series and never become one total. Each applicable field remains reported, reported zero, below threshold, unreported, or absent. A period with both numeric and unavailable inputs is partially reported. Percentage change requires a nonzero numeric starting value and like-for-like measures.

Worked example: the approved Q4 2025 Boeing bundle contains $2,890,000 in self-reported expenses and $450,000 in outside-registrant income across ten matched canonical filings. Those figures are shown separately; $3,340,000 is not published as a lobbying total because the reporting bases may overlap.

Relationships and contextual records

A link requires an approved identity or contextual record with basis, evidence class, confidence, review state, source groups, review date, and a public limitation. Sponsorship, subject overlap, committee responsibility, award-recipient identity, and committee affiliation each mean something different. Similar names, timing, shared keywords, office, or party are not enough. An empty relationship section means no link passed the current review, not that no relationship exists.

Review, refresh, and corrections

Generation and QA reconcile inputs and outputs deterministically. Editorial review checks wording, utility, source freshness, identity, category separation, accessibility, and page-specific limits. Refreshes use approved adapters or curated official-page review; factual changes regenerate outputs. Verified material errors are corrected through the same pipeline and disclosed on existing page or update routes as appropriate.

• Follow the Boeing workflow
• Submit a correction

Publication and indexability

A valid record is not automatically public. A detail page appears in search and the sitemap only after its identity, evidence, usefulness, and publication review is complete. Unpublished, held, and unavailable detail routes stay outside the public directory and use noindex metadata.