deterministic OCaml engine to compute and validate liturgical calendars for multiple rites, template-driven output to year 9999 https://git.labunix.xyz/colitur.git/atom?h=main 2026-08-22T14:40:24Z 2026-08-22T14:40:24Z Lukasz Kasprzak lukas@labunix.xyz 2026-08-22T14:40:07Z urn:sha1:60e87914718dd2c8e69bb89d8cc48d6bef9bfc74 The Missale Romanum's calendarium footnote for February (LT.txt:5011-5014, scan-corroborated scan2.txt:3050-3058): in a leap year the sixth kalends of March (24 February) is doubled rather than a 29th day appended, so St Matthias moves 24->25 February and St Gabriel of Our Lady of Sorrows 27->28, with 24 February itself carrying no fixed office that year. data/ef/expected-divergences-lms.sexp entry L3 tracked this as an open gap. Implemented as the rubric's own general mechanism (every fixed entry from 24 through 28 February shifts one civil day later in a leap year), not as "move these two named saints": the two readings are indistinguishable on shipped data (nothing else is fixed in that window) and diverge only for a future --overlay entry in the same range, which the mechanism reading covers correctly and a two-saints special case would not. This project has already been bitten once (RG 16(a)) by a rule implemented against shipped data's coincidental shape rather than the rubric itself. Kernel stays rite-agnostic: Rite.t gains a fixed_key field (Date.t -> (int * int) option), the same seam easter already establishes, read only by Layer.on_date's FIXED half -- the movable half (Easter_offset/Nth_weekday) is untouched. Default is the identity mapping, an optional argument on on_date, so every existing caller and every rite that supplies nothing is byte-identical to before this field existed. Date.is_leap is exposed in date.mli (already existed in date.ml) so the rite reads the kernel's own single-sourced leap-year definition. The EF implementation lives in Rite_ef.Temporal_ef.bissextile_fixed_key, fully cited, wired into Rite_ef.context. Blast radius, measured over the full 1583-9999 domain (Calendar.year run twice per civil year, real fixed_key vs identity, every day diffed): all 2,041 leap years in the domain show a visible change for each saint; 6,983 individual liturgical days change total, zero unclassified, across four shapes (Matthias vacates 24 Feb in 1,803 years / occupies 25 Feb in 1,811; Gabriel vacates 27 Feb in 1,699 / occupies 28 Feb in 1,670). L3's own prior estimate (1,650 years, observed-outright only) is confirmed close on the same definition (1,677 measured); the broader observed-or- commemorated count is higher, not lower, showing the narrow estimate undercounted visible impact rather than overcounting it. The lectio differential (2005-2050 fixture) gains a new cited entry, C40 in data/ef/expected-divergences.sexp: lectio implements no such shift, so every leap year in its window now diverges on the two shifted days (21 rows, not the naive 44 -- the comparator never compares commemorations, so a side where the shifted saint has zero comparator-visible footprint on both engines produces no row). The LMS Ordo's 2023-2024 fixture -- the window L3 was originally found in -- now matches on the Creed comparison; its formulary-override bucket count is corrected 182->181 (24 February leaves the counted Proper population, becoming Own_slug-sourced). The missalemeum oracle fixtures (2026-2027, 2038, 2035) cover no leap year and are unaffected. L3 is closed: converted from an active allow-list record to a prose closure paragraph (L1's own precedent), citation preserved verbatim, fix and measured blast radius recorded. The id-list assertion narrows to [L4] alone, and the now-dead "2024-02-24" -> "L3" date mapping is removed. Six golden values pinned across four new test cases: Matthias in a leap year (both the vacated and occupied sides) and a common-year control; Gabriel likewise, deliberately choosing a leap year where he is admitted only as an ordinary commemoration rather than winning outright, a different shape from Matthias's. Mutation-tested: forcing fixed_key back to identity reddens exactly five test cases -- the differential's Layer C count-pin (C40 drops to 0 actual vs 21 expected), the LMS Ordo's Creed and formulary-bucket checks, and both new golden pins -- confirming the tests actually exercise the fix. Reverted; dune test, the exhaustive sweep, and make check are all green. 2026-08-20T20:11:17Z Lukasz Kasprzak lukas@labunix.xyz 2026-08-20T20:11:17Z urn:sha1:1988d350242b47aa52aa07904c495e7e2c0eba82 Found by auditing the shipped program rather than the diff. The parser accepted OCaml integer-literal syntax, so "Luke 1_1:5" read as chapter ELEVEN and "+5" as 5 -- a typo silently becoming a different chapter, reachable through any user overlay. Numbers are now plain digits and positive, and a descending range is rejected: 1:20-10 is always a transcription error. No shipped citation changed. FOUR PAIRS OF DIFFERENT BOOKS SHARED A FULL TITLE. 1 and 2 Corinthians both rendered "Epistola ad Corinthios", as did Thessalonians, Timothy and Peter -- 108 citations in 2027 alone that a reader cannot resolve to a book. This is the Kings defect fixed earlier and not generalised. The titles now carry their volume numeral, marked CONSTRUCTED, and a test asserts no two books share a name -- while allowing the case where two ids ARE the same book under different numbering, which a tradition relates. Spec section 8.5 is now delivered rather than merely recorded. Shipped styles did not re-parse their own output: 32 of 52 Latin abbreviations and 49 of 52 full titles failed, so a citation copied from colitur's own output into an overlay was passed through untouched and printed in the wrong language, silently. Every shipped name is registered as a spelling and split_book learned multi-word titles by longest-token match. Now 0 of 52 fail beyond the same-book aliases. Overlay errors were written for a compiler author: they named an OCaml source file the reader does not have and buried the useful token. The existing five-path rewriter is replaced by a generic one, applied to every load path rather than one, so "rank: is not one of the allowed values (at Class9)" replaces the raw Of_sexp_error dump. Also: the new-overlay scaffold documented citations and layer without showing them, and its comment implied the wrong nesting -- the single easiest thing to get wrong; error messages echoed whole file lines, copying an unrelated file's contents into stderr when a flag pointed at one; and config --show validated partway down its table, exiting 2 after writing five rows to stdout. 2026-08-17T15:22:43Z Lukasz Kasprzak lukas@labunix.xyz 2026-08-17T15:22:43Z urn:sha1:2b9f8133a22ce6dc309644a98bbe61507f574b38 Date_spec carried one variant, Fixed(month, day), and its own header said Sunday- and Easter-relative forms would arrive with the OF sanctoral. They arrive early because two things needed them at once: a user-supplied overlay carrying a local movable feast had no way to express "the first Sunday of October", and Rogation Wednesday's commemoration has been recorded as architecturally blocked since 2026-08-13. That blockage turns out to have been a premise, not a fact. The register concluded the Rogation "is not Commemoration_only sanctoral data by nature... it needs a genuinely THIRD kind of thing this architecture has no name for" -- but the reasoning rested on there being "no civil (month, day) pair to anchor a Fixed entry to". That is a statement about Date_spec's expressiveness. With Easter_offset the entity is ordinary sanctoral data, the identical shape Add major-litanies already ships, differing only in how the date is written. No third channel was needed. The register is corrected rather than left contradicting the code. Two variants: Easter_offset of int, and Nth_weekday of {month; nth; weekday} with negative nth counting from the end. Fixed's sexp representation is untouched, so all 327 sanctoral entries and the whole overlay parse unchanged. Both are re-validated in t_of_sexp through their smart constructors, extending the idiom Fixed already used -- the failure that guards against is invisible, a spec deserialising into something that silently never resolves and a celebration vanishing with no diagnostic. Easter is supplied by the rite, not computed in the kernel. Computus ships Gregorian and Julian both, and picking one here would hard-code a Roman assumption into rite-agnostic code and be silently wrong for a Julian-reckoning rite. Rite.t gains an easter field; Rite_ef supplies the Gregorian one. Layer keeps a split index. Fixed entries stay in the year-independent (month, day) table, which preserves the fast path and the "30 November counted twice in a 371-day span" behaviour validate.mli documents for St Andrew. Movable entries have no year-independent key by construction -- the same spec lands on a different (month, day) each year -- so they resolve per civil year into a rata-die table. index_by_date is removed rather than kept as an alias: two ways to index a layer, one of which silently ignores movable entries, is the trap this change exists to avoid. Two domain-edge defects, both found by the suite rather than reasoning. A liturgical year is Advent-anchored, so resolving civil year y names y +/- 1, and at the edges those are 1582 and 10000, which Computus correctly refuses by raising. The ceiling surfaced through the domain-ceiling test, the floor through `colitur day 1583` failing outright. Fixed once, in Layer.index, the single point that calls the rite's easter -- a clamp repeated per caller would have been two places to get wrong a third time. Rogation Wednesday: RG 87 assigns the Litaniae minores to the Monday, Tuesday AND Wednesday before Ascension (scan1:691); colitur built the first two from the start and never the third. RG 88 keeps it out of the Office, so the Ascension Vigil remains observed. RG 89 routes it through the Major Litanies' rules, but RG 109's closed list names only maiores, so by RG 107 it is an ORDINARY commemoration -- confirmed in the output, not assumed. Name from the Missal's own "feria IV" (scan1:20495-20497), violet to match the Monday and Tuesday offices. Blast radius, full 1583-9999 sweep against the pre-change binary: 2257 days, every one the same shape, zero unclassified, line counts equal both sides. The complement is the informative half -- in the other 6160 years the Rogation does not appear at all, because RG 111 admits one commemoration on a II-class day and an impeded feast takes it under RG 113, a Commemoration_only candidate having no row in RG 91's table. Both shapes are pinned. Allow-lists unmoved: layer 3 compares no commemorations and is blind by construction, and layer 4's 2026-2027 window has a competing feast on both Wednesdays, so the Rogation is capped out in exactly the years the oracle can see. C8 is unchanged, asserted by its own count pin. Not built: Sunday-relative specs, which nothing needs yet, and the --overlay plumbing that would let a user actually supply a local movable feast. Nth_weekday therefore ships with synthetic and property coverage only; inventing a shipped entry to exercise it would be fabricating calendar data. Register: section 6.10. 2026-08-11T14:21:46Z Lukasz Kasprzak lukas@labunix.xyz 2026-08-11T14:21:46Z urn:sha1:745ec8f8edb32ee7ef0a8c12ab84833216296f73 Layer.load narrowed its catch to Sexplib0.Sexp_conv_error.Of_sexp_error, but rank_of_sexp is caller-supplied and may raise anything -- e.g. a hand-written rank parser that calls invalid_arg. Overlay.load already catches every exception from the equivalent call; mirror that here so layer.mli's "never as an exception" promise actually holds. Date_spec.t derived its sexp converters with plain ppx_sexp_conv, unlike Slug and Lang, which hand-write validating parsers specifically so malformed data is rejected at load. (Fixed (month 13) (day 40)) used to deserialise cleanly into a spec that simply never resolves -- a saint quietly vanishing with no diagnostic. t_of_sexp now re-runs the value through the existing fixed validator, the same shape Slug and Lang already use. Covering tests: a Layer.load case where rank_of_sexp raises Invalid_argument instead of Of_sexp_error (would have escaped uncaught before this fix); two Date_spec.t_of_sexp cases (month 13, 31 April) that must raise Of_sexp_error rather than silently constructing an unresolvable spec. 2026-08-11T10:15:44Z Lukasz Kasprzak lukas@labunix.xyz 2026-08-11T10:15:44Z urn:sha1:2e3697b5ce7338568a67531d07a5528f0de1da22 Entries sort by slug so equal layers serialise identically. The by-date index is built once per layer rather than per year, since fixed dates are year-independent; a full-domain sweep would otherwise rescan every entry for every day. load turns parse and validation failures into result.