Data

Everything, with its provenance

Provisional — not for citation. These records are pipeline output. They have passed automated checks but have not yet been verified by a human against the page images, which is required before release. Figures will move. Fifteen things have already gone wrong, most of them mine.

Downloads

register.csv — every transcribed determination (1,400 rows) with its risk verdict, flags, source citation, page number, and the verbatim source text it was read from.
unscreenable.csv — the subset nobody can screen (362 rows).
tegiszhol_c14bazaar.csv — the same determinations mapped onto the 18 standard c14bazAAR variables, ready for ingestion into the aggregator ecosystem. Records the rule suite held back are excluded; the reservoir-screening note rides along in comment.
tegiszhol-method-preprint.pdf — the method paper, What broke, and why it was almost never the model.

Known gap: no coordinates

The register carries no latitude or longitude, and therefore no country attribution. c14bazAAR derives country from coordinates, so until these sites are georeferenced these determinations will not appear in spatial queries. That is the single most valuable outstanding enhancement.

Coordinates have deliberately not been guessed. Assigning a plausible location to a site is exactly the class of fabrication this project exists to document, and a wrong coordinate is far worse than an absent one — it is undetectable downstream.

Columns worth knowing about

verbatim_row is the source text the record was read from, transcribed as printed. It is the evidence for the row. If the two disagree, trust neither and check the page.

screenable is NO when the material can carry a reservoir offset and no isotopes were published. isotopes_elsewhere_in_document marks records where the data exists in the source but has not yet been joined to the determination — those are excluded from the unscreenable count, because that gap is ours.

flags records every automated check that fired. Several are informational rather than faults: isotopes_absent is the finding itself, lab_code_absent means the source published a date with no laboratory code at all, and error_not_published means the ± term is missing from the page.

age_scale distinguishes uncalibrated BP determinations from calendar dates quoted in prose. Only bp rows are determinations; the rest are excluded from every count.

Citing the sources

Every row carries the citation of the publication that produced the date. Cite that, not this. This compilation is a pointer to other people’s excavation and measurement, and it exists to send readers to them.

Source PDFs are not redistributed. corpus/manifest.csv in the repository records each document’s citation, licence and access route so the corpus is reconstructible.

Code

https://git.sr.ht/~calgacus/tegiszhol

The extraction schema, the deterministic rule suite, the laboratory-code normaliser, and every script that produced these figures. The numbers on this site are generated from the data by scripts/build_site.py; none is typed by hand.