What 38,100 House disclosures taught us about the record

The useful result of building a congressional financial database is not the biggest number on the home page. It is knowing which records the number does and does not represent. On August 24, the production backfill reconciled 38,100 unique House index documents covering calendar years 2013 through 2026. That count is a source-document count, not a trade count.

Most disclosure documents are not transaction reports

Of those documents, 8,350 were periodic transaction reports. The parser could read 5,903 of them and extracted 34,588 transaction rows. Another 2,447 reports were image-only. Those reports remain in the disclosure catalog with their parse limitation visible; they are not silently treated as zero activity.

Nineteen exact duplicate source rows are represented once. Deduplication changes a database count, so it is part of the methodology—not cleanup to hide in an implementation note.

A trade date is not a filing date

The transaction date describes when the filer says an activity occurred. The filing date describes when the report entered the public record. A monitoring product needs both: the first is about reported activity; the second determines when a reader could have known. The trade feed and every briefing keep them separate.

The filer may not be the owner

House forms identify an owner code such as self, spouse, joint, or dependent child. A headline that converts every row into “the member bought” overstates the source. Watchlists and alerts therefore follow the filing while preserving the owner label.

Why this becomes a research product

The public database is the authority layer. A generic brief helps readers discover new filings. Watchlists, saved queries, comparisons, exports, and personalized alerts reduce the work of returning to the record. That is the product boundary: evidence first, monitoring second, occasional analysis third. It does not require turning the site into a general political publication.