The Digital Decretals: hyperlinking a 13th-century law code

The Decretals of Gregory IX served as the Church's legal backbone for centuries, and medieval law students learned it through Bernard of Parma's Glossa Ordinaria — the standard commentary keyed, phrase by phrase, to the text. Our client, a professor of medieval canon law, put it best: if the Decretals are the Torah, the gloss is its Talmud. He had spent years hand-transcribing the complete gloss into a Word document — about 4.4 million characters of Latin — and standardizing the medieval citation abbreviations into a searchable form. What he needed was the web edition: free-text search, filters by book and title, and a home for the whole apparatus.

The brief arrived as a detailed email and two files. The first working edition came back the same way the studio builds everything — AI-accelerated, senior-reviewed — and the complete delivery landed as five commits in a single day.

Gloss units

9,872

Citation links

18,257

Chapter pages

1,970

CI checks

47

The transcription is the database

The professor's Word file already had structure: native heading levels for book, title, chapter, and gloss word. The pipeline honors it — a Python parser walks the heading hierarchy into JSON, and zero text loss was confirmed against the Word file. His publishing rule is encoded too: an --books auto mode publishes only books whose title counts match the expected complete-book counts, because he never wants a partially transcribed book online.

Search that respects the apparatus

Medieval legal citations depend on punctuation — qui fil. sint leg., per venerabilemis a precise address, not prose — so search matches exact punctuation by default, with ignore-punctuation, match-case, and whole-word modes on top. An orthographic-tolerance mode collapses ae/oe to e and v to u so medieval spellings match classical ones — and highlight ranges map back to the original string, so a match found under tolerance still highlights the manuscript's own digraphs. Everything runs client-side, and the app measures itself: typical full-corpus queries over all 9,872 units come back in single-digit milliseconds.

Eighteen thousand links, one click each

Because the citations were standardized, they could be precomputed: a build step finds every citation to the Decretals themselves — 18,257 of them, out of roughly 40,000 total allegations across six law codes — and turns each into a hyperlink to the chapter it cites, both in the app and in 1,970 pre-rendered, SEO-indexable chapter pages. The link extraction was validated against the professor's own register of citation counts: 1,970 of 1,971 chapters agree exactly, and the single mismatch is a documented over-count in the register itself.

Validation as a deliverable

Checking every count against the register surfaced things no one had seen in years of transcription: a chapter that appears twice in the Word file, two chapters with no gloss sections, a spreadsheet typo, and that register over-count. Each went into an errata list for the professor — and where the site had to make a call, it makes the call visibly: the duplicated chapter's page renders both blocks with an inline scholarly note, rather than silently merging them.

The most valuable deliverable wasn't the search box — it was the errata list. Systematic validation gave the scholarship something back.

Ships two ways

• ▸A static site — no server, no tracking, zero external requests; system fonts only; CI runs the 47-check corpus suite and a production build on every push to main.
• ▸A single 5 MB HTML file — the same core search and complete text in one self-contained file that runs offline from a double-click. USB-stick scholarship, for colleagues who'd rather not trust a URL.
• ▸The client's files, improved — per-book Word downloads regenerated from his master with formatting preserved, plus his full abbreviations spreadsheet as a live, filterable explorer.

The licensing is split on purpose: the software is MIT, while the transcription and all derived data remain the professor's — his life's work stays his.

What we'd tell a client

• ▸If your data already has structure — even in a Word file — honor it instead of re-inventing it. The heading hierarchy was the schema.
• ▸Validate against an independent source of truth and treat the discrepancies as deliverables, not noise.
• ▸Respect the domain's conventions: exact-punctuation search sounds pedantic until it's the whole point.
• ▸Offline copies still matter. A single file that works from a double-click travels further than any deployment.