The shared Dacia GIS family

Three packages under data/dacia/gis/, compiled by make dacia into four Atlas layers. They are shared: no trench owns them, and each is meant to be reused by whichever trench needs a baseline, a boundary or a sheet footprint.

The Roman baseline ships as two layers because the Atlas takes one render hint per layer, not because it is two datasets. Both are compiled from one table.

The rule the family is built on

Geometry is typed by where it came from, and the type is checked. The geometry_provenance vocabulary was frozen at KAN-330 for exactly this: from source_geometry at one end, through georeferenced_source and scholarly_reconstruction, to editorial_reconstruction and display_generalisation at the other.

Nothing in this family is digitised from a source sheet. That is not a caveat in prose - the validator rejects any row here that claims source_geometry or georeferenced_source, and the two drawn GeoJSON files must each carry surveyedGeometry: false and a recorded justification. When the first digitised layer arrives it will have to come with the change that permits it, which is the point: the rule makes an upgrade in confidence a visible edit rather than a quiet one.

Roman Dacia (KAN-341)

roman-dacia.csv holds three kinds of feature, and only one of them has geometry of its own.

• A site is a CND place. The row names a place_id and nothing else spatial; the build reads that place’s reference location and carries the corpus’s own ref_geometry_provenance onto the feature. This is what keeps the baseline from becoming a second, drifting copy of coordinates the corpus already holds, and it is why the validator refuses a site whose place_id does not resolve.
• A road is an ordered list of those same places. The line is what you get by joining them, and it is declared editorial_reconstruction because that is what it is: the route is well attested, the drawn segments are not. Every road note says which one it is - rd-road-danube is straight between crossings while the real road follows the river.
• A limes is the only thing drawn. A frontier system is not a sequence of attested towns, so the three corridors live in roman-dacia-lines.geojson, each a schematic line at provincial scale with low confidence. The dash pattern on the Atlas separates them from roads so the two are never read as the same kind of claim.

The identifications follow the Barrington Atlas and TIR L-34/L-35. Those works are in copyright; this layer cites them and reproduces nothing from them, which is what rights_status: in_copyright records.

Principality phases, 1526-1859 (KAN-342)

Twelve phases across six polities. Change over time is phases, not one polygon, and the validator enforces the arithmetic: a phase must begin in the year of the instrument that opened it, and two phases of one polity may not overlap in time. An overlap means a boundary was moved on one side of a treaty and not the other.

Bounds are inclusive at both ends, because the Atlas filter is valid_from <= year <= valid_to. A phase therefore ends the year before the instrument that replaced it - Wallachia-with-Oltenia runs to 1717 and the reduced phase begins in 1718 - or both would render on the changeover year, which is the overlap the validator refuses.

The phases exist because the territory moved:

Modern national borders are never used as proxies. That is the failure this layer exists to avoid: a 1947 line drawn over 1718 makes Habsburg Oltenia invisible, and makes Moldavia-before-the-Prut invisible too. The rings are drawn from the territorial record instead, declared editorial_reconstruction, given low confidence, and the GeoJSON carries derivedFromModernBorders: false as a checked field. They are envelopes for temporal filtering, not delimitations, and no vertex should be read as a claim about where a boundary ran on the ground.

Two simplifications are recorded as verification debt rather than hidden: the Budjak is not separated from Moldavia before 1775, and the southern Bessarabian districts returned in 1856 are not cut as their own phase.

Josephinian sheet index (KAN-343)

Fifteen footprints over the First Military Survey of Transylvania, 1769-1773, one for each Transylvanian place in the frozen pilot.

covers_place_ids is recomputed from the footprint at build time rather than trusted from the table: a sheet that claims a place outside its own bounds is a mistake the build should catch, whichever of the two is wrong, and the validator refuses it either way.

Two things this index deliberately does not do:

• It does not carry the archive’s sheet numbers. Every row’s archive_sheet_id is pending. Inventing a plausible-looking Coll./Sectio number would be worse than admitting the gap, and the debt that records it blocks Trench A’s research gate - the Iosephina stela’s attestations cannot be reviewed without a sheet locator, and this index is where one would come from.
• It redistributes no scan. The sheets are held by the Kriegsarchiv in Vienna and the layer links to the repository that may show one. scan_redistributed must be no, checked on every row.

The footprints are a regular graticule approximating the survey’s sheet layout, not the archive’s index geometry, which is why they are typed editorial_reconstruction and confidence low.

Treaty frontiers, 1829-1947 (KAN-352, KAN-353)

Eight lines across seven phases, each attributed to the instrument that made it and to the years it held.

No timeless line. A frontier without a start year is not a phase, and the validator refuses one. The 1947 restoration is the only open-ended row, because it is the frontier still in force - and it is recorded as its own phase rather than as a continuation of 1920, since a frontier restored by instrument is a separate legal fact from the frontier it restores.

Three kinds of line, and never one column. line_type separates a proposal from the treaty_line an instrument actually fixed and from a later reconstruction. The Atlas carries that distinction in the dash pattern and the stroke width as well as the colour, so the difference survives for a reader who cannot use colour: a proposal is dotted and thin, a treaty line solid and heavy.

Competing lines are kept, not averaged. Where two sources give different lines for the same moment, each is a feature and each names the other in alternative_of. The validator enforces the disagreement from both sides - a phase carrying two lines where only one declares the conflict fails - and refuses a phase holding two treaty_line rows, because a competing line is a proposal or a reconstruction, not a second instrument claiming the same authority. The case in the data is the Trianon frontier of 1920 beside de Martonne’s ethnographic proposal for the same zone: one an instrument, one an argument. Averaging them would produce a frontier nobody proposed and erase the disagreement the trench exists to show.

The proposal is cited from the Carta Rubra ledger rather than the treaty one, so a row names which ledger its source_id belongs to and the validator resolves it there. That is what lets a frontier phase cite an ethnographic argument without either package absorbing the other.

Nothing here is digitised. The KAN-351 ledger records the geometry position of all eight instruments: no_geometry for Adrianople and the 1944 armistice, commission_geometry_required for Paris 1856 and Bucharest 1913, map_not_georeferenced for Berlin, Trianon and Paris 1947, and annex_map_unacquired for the Second Vienna Award. Tracing a modern boundary in their place would launder a 1947 line into an 1829 claim, so the corridors are drawn, declared editorial, and given low confidence. vd-treaty-frontier-geometry records the debt and blocks Carta Rubra’s data gate, since that trench argues over this frontier.