Where the data comes from

The sources

Roll calls, the filed text of instruments, committee calendars and rules, floor transcripts, and official publication of enacted law. All public, all citable, all indexed by somebody other than us.

Why this is the firmest ground in the product

Because procedural history is boring and heavily duplicated.

A committee calendar is not a matter of interpretation. It is a document produced by the institution itself, for its own purposes, and mirrored by half a dozen third parties. Nobody has to trust our reading of it, and if we get one wrong, somebody can prove it in a minute.

Everything else the movement publishes about a person is more arguable than this. That is why the wall exists: the strongest evidence gets its own awards and is not diluted by weaker material.

Provenance

Every download carries a provenance record: what was fetched, from where, and when. Those records live in the public repository alongside the data.

So a disagreement about our numbers is resolved by looking at what we actually pulled — not by arguing about whose recollection is right.

Storage

The record lives in the public data repository, per problem: the enacted-law record identifying the instrument, and a roll-call table beside it. Flat files. You can open them without our software, and git log will show you what changed and when.

Gaps, and why they are published

Coverage is never total. Some periods are thin, some stages are poorly recorded, some sources conflict.

Every run emits a gaps list. Those gaps render on the page rather than being quietly smoothed into an average, and confidence in a record is reported, never scored — it is disclosure to the reader, and it does not silently adjust any award.

A named gap is also the most actionable thing on the page. It tells somebody with access to an archive exactly what would close it.

What the engine is never allowed to do

Fill a gap with an inference. If the record does not establish it, the output says so. A plausible interpolation is worse than a hole, because a hole is honest and an interpolation is indistinguishable from evidence.