# Origenality A bibliographic map of Origen studies: what has been written on Origen of Alexandria, where the scholarship is dense, and where it is thin. Site: https://origenality.com · Source: https://github.com/romain-girardi-eng/origenality ## The problem Anyone starting a piece of work on Origen faces the same question, and no instrument answers it: has this been done? The bibliographies that exist are either closed (subscription indexes), partial (one language, one library), or frozen at their date of publication. Cross-searching five catalogues by hand gives a pile of records with no way of knowing how much of it is the same work listed five times, and no way of seeing what the pile leaves out. Origenality federates the open bibliographic sources on Origen, deduplicates them into one record per work, classifies each record by theme, by work of Origen and by approach, and renders the result as a map you can read at a glance and query in plain language. The measure it gives is a density — how many studies already sit in a given neighbourhood — not a verdict on originality. An empty region may be a gap worth filling or a dead end others have already walked away from; the map says which regions are empty, and that is all it says. ## Where it stands **Data.** Ten open sources harvested: 45 685 raw records, merged into **42 210 clusters** (3 475 duplicates collapsed). 24 802 of those clusters carry an abstract, each one attributed to the database that wrote it and linked back to its record. **Classification.** 21 104 clusters have been read against a controlled vocabulary of themes, works, approaches and relevance. Of those, 4 909 are about Origen and 1 833 give him a section of the argument: **6 742 records form the citable dossier**. 1 100 mention him only; 13 262 are noise the harvest brought in and the classification pushed out. Every published figure counts the first two classes and nothing else. **Site (v0).** The explorer published so far draws the IxTheo/K10plus harvest alone: **1 632 records, of which 1 402 are counted**, 225 mentioned only, 5 held outside the count. It is static — no request leaves the page — and it carries an observatory of the field by decade, language, format, theme and work of Origen. **Citation weight.** Citation counts are measured for 14 772 of the 42 210 clusters (35 %). They are shown as a percentile inside a cohort of the same decade, type and language, so that a 1960s German monograph is not crushed by a 2015 English article. The weight is a visual cue, never a filter: no record is excluded from a search or a count because it has no citation data. ## Limits - **Citation coverage is uneven by language**: 88.9 % for Spanish, 39.8 % for German, 35.9 % for English, 24.4 % for Italian, 9.6 % for French. The cause is the absence of older and non-anglophone records from the bibliometric sources, not a failure of the join. Any reading of citation weight across languages is therefore unsafe, and the site says so on the page that shows it. - **The classification is machine-assisted and its agreement with hand tagging is not currently published.** A reference set of fifty records tagged by hand exists; the measurement against the current prompt has not been run, and no agreement figure is stated anywhere until it has been. - **Subscription indexes are absent** (they cannot be harvested and would not be redistributable), so the coverage of pre-1980 and of some national traditions is thinner than the counts suggest. - **One harvest, one date.** The figures above describe the August 2026 harvest. There is no incremental update yet. - The federated corpus is not what the published site draws: v0 shows the catalogue harvest only, while the classification and the citation work run on all ten sources. ## Reproducing it Python 3.11 or later, standard library only. No key, no service, no build step. Every command below is run from the root of the clone. What a fresh clone runs as it stands, on the files it ships with: ```bash python3 scripts/test_data_gates.py # the test suite python3 site/tools/build_summary_figures.py --check # every published figure recounted python3 site/qa/check_one_population.py # one population on every screen python3 scripts/build_seo_assets.py --check # sitemap and llms.txt still current python3 -m http.server 8020 # then open /site/index.html bash scripts/selftest_public.sh # all of the above, in a throwaway clone ``` What needs a harvest first, because the raw harvests and the merged corpus are not in this repository — they are large, and several sources ask that their dumps not be redistributed: ```bash python3 scripts/harvest_openalex.py # one harvester per source, in scripts/ python3 pipeline/merge_dedup.py --out-dir data/merged # federate and deduplicate python3 pipeline/enrich_citations.py # citation counts and cohorts python3 pipeline/build_site_data.py --tags # the site data layer python3 scripts/qa_checks.py all # the standing checks on the merged corpus ``` The harvesters are rate-limited and resumable, and running them rebuilds the corpus from the sources themselves. `docs/methodology.md` describes the federation, the deduplication rules, the classification and the density measure. `docs/sources.md` lists the sources with their licences. `docs/seo-geo.md` says what the site declares to search engines and to AI crawlers, and how to check it after a deployment. `api-spec/openapi.yaml` is the contract the site consumes; the implementation that serves it is separate from this repository. ## Citing it See `CITATION.cff`. In a footnote: > Romain Girardi, *Origenality: a bibliographic map of Origen studies*, 2026. Cite the underlying databases as well when you cite a record: the attribution template for each is in `ATTRIBUTION.md`, and Semantic Scholar in particular requires attribution under ODC-BY. ## Contact, corrections, takedown romain.girardi@univ-cotedazur.fr Abstracts are published with the name of the database that wrote them and a link to the record it wrote them for. A publisher, a database or an author who asks for a summary to be removed or replaced gets it removed or replaced, without discussion and without having to explain. Write to the address above, name the source or the record, and the change goes into the next build; the takedown is scripted, not manual (`scripts/check_release.py --withdraw`). Corrections to the data are as welcome as takedowns: a merged pair that is not the same work, a work split in two, a misclassified record. Name the record and say what is wrong with it. ## Licence Code: MIT (`LICENSE`). Data: see `DATA_POLICY.md` — they are published under attribution and takedown on request, and the licence of each source is stated in `ATTRIBUTION.md`. The metadata come from ten databases under distinct regimes, so no single licence is asserted over the whole, and `CITATION.cff` therefore declares none.