deposits
Before a dataset can be used it has to be read — not its data, its documentation. Four datasets were read for this phase. Two were accepted, and the two most interesting findings came from sentences nobody would query.
About: Bingham Canyon · Viburnum Trend
The USGS porphyry copper database defines its sampling unit operationally: all mineralised rock or alteration within two kilometres was combined into one deposit. The compilers explain why — a spatial rule is necessary because deposits have to be classifiable in regions where the geological information is highly variable — and they record in an `includes` column which named bodies were merged. Seventy-six of 690 rows carry such names.[1]
That means a record built from one of these rows may be a larger object than the body a paper or a mine calls by the same name. Lithoen carries the merged names on the mapping so the difference is visible rather than inherited.[1]
The attribute definitions state it plainly: a grade of zero means the value was unavailable. In the porphyry file, 268 of 690 rows carry zero copper grade on that basis and 434 carry zero gold. Any pipeline that read those as measurements would publish hundreds of false numbers, every one of them structurally valid and none of them detectable from the data alone.[1]
This is the single most consequential thing this phase learned by reading metadata instead of data, and it is the reason the ingestion tool records refusing those fields on every record it writes.[1]
The Viburnum Trend appears in the sediment-hosted zinc-lead database as a single row — one deposit, by that dataset’s own definition. Its `includes` column lists separate named orebodies, and the USGS Mississippi Valley-type model states that districts of this class consist of several to as many as 400 deposits.[1]
Lithoen therefore holds the Viburnum Trend as a district rather than a deposit, and records the reinterpretation on the mapping with the confidence marked inferred from documentation rather than stated. This is not a correction of the source: the compilers were building grade-and-tonnage models, and for that purpose a trend is a legitimate sampling unit. It is a different purpose producing a different object.[1]
The USGS Mineral Resources Data System covers the world and would have multiplied this corpus many times over. It was read and refused — not on rights, which are public domain, but on semantics. It merges two databases built over decades by hundreds of contributors; its rows describe occurrences, prospects, mines, shafts and districts with no field that separates them; and it says of itself that positional information is highly variable, that about 4,000 records have no reliable coordinates, and that it is generally sparse.[1]
Taking it would have meant mapping tens of thousands of rows of unknown type to one entity type. The refusal is held as a record rather than as an absence, so that a later phase finds a decision instead of an oversight.[1]
Often said
“A dataset from a major scientific institution can be ingested as it stands.”
What the evidence supports
Rights and authorship are not the same as semantics. Of the four datasets read for this phase, all four were public domain and two were unusable — one because its rows mean different things without saying which.
Supporting records: mrds-refused · porcu-row-is-a-deposit
Bracketed references link to the structured entity records whose claims support each paragraph. Last updated 2026-09-07.