v2026.09.02 — The Hours, the Readings, and a New Apex

Three releases had gone out without a note. This one covers all of them, and the newest is the largest thing the calendar engine has ever served.

Added

• GET /v1/ordo/officium/{hora} — one hour of the Little Office of the Blessed Virgin Mary, composed for a date. All eight canonical hours, three seasonal sets, Matins keyed by weekday, and the Paschal antiphon resolved by precedence.
• GET /v1/ordo/readings — the citations appointed for a day, with a witnesses count so a consumer can weigh corroboration rather than trust the archive uniformly.
• get_daily_readings — the same schedule as an MCP tool.
• torchandlily.com is now the altar: the day, the readings, the saint, and the office of the hour, in the reader’s own timezone. The developer portal stays at dev.torchandlily.com.

Changed

• GET /v1/ordo now returns week, the day’s standing in its season. It is null where the season numbers no weeks — Christmastide, the Triduum, and the days between Ash Wednesday and the First Sunday of Lent. Treat absence as “this day has no week”, never as zero.
• suggest_endpoint no longer scores a match on terms that appear in most endpoint descriptions. A word that selects everything selects nothing.

Fixed

• Four verses that had been silently absent from the Douay column are restored: Gen 5:32, Jn 11:57, 2 Cor 1:24, 3 Jn 15.
• 1,437 citations that had been answering with the wrong text now decline instead. A refusal is a better answer than a plausible wrong one.

A note on what the office is, and is not

The psalmody is cited, not transcribed. Every psalm, canticle and little chapter is composed at render time from our own Clementine Vulgate, in the office’s own Vulgate numbering — so no modern-to-Vulgate mapping stands between the citation and the text, and none of it carries any risk from the scan it was read against.

What is transcribed — the antiphons, hymns, responsories and orations — comes from The Little Office of the Blessed Virgin Mary, Benziger Brothers, 1915, which is in the public domain. No human has yet read that transcription against the scan, and the archive says so: every provenance row it writes carries probatus: false. Two texts the resolver declined are stored as the book prints them, with no citation attached, because our Clementine and the 1915 setting genuinely differ there.

If you build on this, the kind field on each element tells you which you are holding: CITATUM from our archive, TRANSCRIPTUM from the book.