kernel(rite-ef): replace unearned/retracted citations with verified ones

Four citation problems, all resolved against the primary scans
(docs/research/*.pdf via pdftotext):

1. temporal_ef.ml's Ember-day date-derivation comment still carried a
   retracted note claiming the scans have no numbered-paragraph statement
   of the rule. Register §3a corrected this in an earlier commit (the
   search used the genitive "Quatuor Temporum"; the rubric heading is the
   nominative "Quatuor Tempora") and the register's own words warn "a
   false 'not in the source' note is worse than no note" -- the code
   still carried the retracted note verbatim. Replaced with the verified
   citation and the primary Latin.

2. temporal_ef.ml cited RG 71 for "Advent I is the Sunday nearest 30
   November". RG 71 is a season-boundary citation only ("a I Vesperis
   dominicae I Adventus..."); it does not say which Sunday opens the
   season. The actual rule is RG 20 (Caput III, "De Dominicis"):
   "Dominica I Adventus ea est, quae cadit die 30 novembris vel est ipsi
   proximior" -- verified against the scan and added to the register.
   Behaviour was always correct; only the citation was borrowed.

3. RG 88 ("de Litaniis minoribus nihil fit in Officio") is cited five
   times across code and tests, load-bearing twice (why Rogation days
   take their season's ordinary ferial class; why privilege_of lets them
   fall through to Ordinary), but appeared nowhere in the register. RG 67
   and RG 69 were likewise cited at expected-divergences-missalemeum.sexp
   (M11) without register backing. All three verified against the scan
   and added to the register's §4 (Octaves / Rogations subsections).

4. Two citations flagged for re-verification, both CONFIRMED accurate
   against the primary scan (word for word):
   - precedence_ef.ml's RG 32 "si impeditur, commemoratur" (the sole
     textual basis for treating a vigil as always-commemorated once RG 33
     doesn't omit it outright) -- the register previously stated RG 32
     only as "same pattern as RG 31", not verbatim; the full sentence is
     now in the register too.
   - precedence_ef.ml's RG 111(b) full sentence (the sole textual basis
     for the shipped Sunday rank-floor fix) -- the register previously
     carried only the fragment "de festo II classis"; the full sentence
     is now there.
   - temporal_ef.ml's RG 91 entry 7 "feria IV cinerum et II, III et IV
     Hebdomadae sanctae" (the sole justification for stopping the I-class
     ferias at Wednesday) was also checked and matches the scan exactly;
     left as-is (already correctly cited), noted in the report.

Bonus finding while re-verifying RG 20: RG 17(d), same chapter, states
"festum D. N. Iesu Christi Regis, celebrandum dominica ultima mensis
octobris" -- primary-source confirmation that Christ the King falls on
the last Sunday of October. This was the register's one
oracle-backed-but-not-primary-verified rule (CLAUDE.md's own carried-item
language); it no longer is. Updated temporal_ef.ml's christ_the_king
comment and closed the item in register §6.

No behaviour change: every edit here is a comment/citation change to
already-correct code. Verified byte-identical `colitur day` output across
1583, 1900, 1902, 2008, 2011, 2026, 2038, 9999. 259/259 tests green.

2 files changed, 60 insertions, 24 deletions

diff --git a/lib/rites/rite_ef/precedence_ef.ml b/lib/rites/rite_ef/precedence_ef.ml
index bfd3b93..3c417bb 100644
--- a/lib/rites/rite_ef/precedence_ef.ml
+++ b/lib/rites/rite_ef/precedence_ef.ml
@@ -596,7 +596,13 @@ let disposition ~(winner : Vocab_ef.rank Precedence.candidate)
        (III class, "si impeditur, commemoratur"), is ALWAYS commemorated
        once RG 33 does not omit it outright, regardless of ferial class --
        a rule Caput IV does not speak to at all (vigils are Caput V, RG
-       28-34, not "feriae"). *)
+       28-34, not "feriae"). RG 32's own full sentence, primary-source-
+       verified (final fix wave): "Vigilia III classis est vigilia S.
+       Laurentii. Haec vigilia praefertur diebus liturgicis IV classis; et,
+       si impeditur, commemoratur, iuxta rubricas" -- confirmed word for
+       word against the scan, not constructed by analogy with RG 31 (the
+       register's own §4 "Vigils" entry states RG 32 only as "same pattern
+       [as RG 31]", not verbatim -- now closed here). *)
     Precedence.Omit
   else
     (* RG 95's other branch: "aut commemorantur aut penitus omittuntur" --
@@ -698,12 +704,17 @@ let admit ~(observed : Vocab_ef.rank Precedence.candidate)
          several are. *)
       (match List.filter is_privileged sorted with [] -> [] | best :: _ -> [ best ])
   | Class2, true ->
-      (* RG 111(b), primary text: "in dominicis II classis, una tantum
-         admittitur commemoratio, SCILICET DE FESTO II CLASSIS, quæ tamen
-         omittitur si commemoratio privilegiata facienda sit" -- "on Sundays
-         of the II class, only ONE commemoration is admitted, NAMELY OF A
-         FEAST OF THE II CLASS, which however is dropped if a privileged
-         commemoration is due." Two clauses, not one: (i) a privileged
+      (* RG 111(b), primary text, RE-VERIFIED word for word against the scan
+         (final fix wave; this sentence is the sole textual basis for the
+         shipped rank-floor fix below, and the register's own §4 "RG 111"
+         entry previously carried only the fragment "de festo II classis",
+         not the full sentence -- now added there too): "in dominicis II
+         classis, una tantum admittitur commemoratio, SCILICET DE FESTO II
+         CLASSIS, quæ tamen omittitur si commemoratio privilegiata facienda
+         sit" -- "on Sundays of the II class, only ONE commemoration is
+         admitted, NAMELY OF A FEAST OF THE II CLASS, which however is
+         dropped if a privileged commemoration is due." Two clauses, not
+         one: (i) a privileged
          commemoration, whenever due, categorically takes the day's one slot
          -- not by comparing its dignity against the ordinary contender's,
          so an ordinary commemoration that would otherwise win on raw
diff --git a/lib/rites/rite_ef/temporal_ef.ml b/lib/rites/rite_ef/temporal_ef.ml
index 1b3d219..c3db15c 100644
--- a/lib/rites/rite_ef/temporal_ef.ml
+++ b/lib/rites/rite_ef/temporal_ef.ml
@@ -17,9 +17,17 @@ let weekday_index d =
 (* The Sunday on or before [d]. *)
 let sunday_on_or_before d = Date.add_days d (-(weekday_index d))
 
-(* RG 71: Advent I is the Sunday nearest 30 November -- equivalently the fourth
-   Sunday before Christmas, i.e. three weeks before the last Sunday on or before
-   24 December. *)
+(* RG 20 (Caput III, "De Dominicis"), primary-source-verified (final fix
+   wave): "Dominica I Adventus ea est, quae cadit die 30 novembris vel est
+   ipsi proximior" -- Advent I Sunday is that which falls on 30 November or
+   is nearest to it. CORRECTED citation: this comment previously cited RG
+   71 for this placement rule -- WRONG, RG 71 (cited on [season] below)
+   states only Advent's own season BOUNDARY ("a I Vesperis dominicae I
+   Adventus..."), not which Sunday opens it; the register's own RG 71 entry
+   is a boundary citation, and the only "nearest 30 November" text there
+   before this fix was UNLYC nn. 39-42, the MODERN form's rule, not this
+   one's. Equivalently the fourth Sunday before Christmas, i.e. three weeks
+   before the last Sunday on or before 24 December. *)
 let advent_start y = Date.add_days (sunday_on_or_before (mk y 12 24)) (-21)
 let year_start = advent_start
 
@@ -47,8 +55,13 @@ let season d =
   else if Date.compare d paschal_end <= 0 then Paschaltide (* RG 76 *)
   else Time_after_pentecost (* RG 77 *)
 
-(* Last Sunday of October, per the 1960 calendar -- NOT the OF's last Sunday
-   before Advent. Register §6 flags this for primary-source confirmation. *)
+(* RG 17(d) (Caput III, "De Dominicis"), PRIMARY-SOURCE-VERIFIED (final fix
+   wave, closing the item register §6 previously carried as "oracle-backed,
+   not yet primary-verified"): "festum D. N. Iesu Christi Regis, celebrandum
+   dominica ultima mensis octobris" -- the feast of Our Lord Jesus Christ
+   the King is to be celebrated on the LAST SUNDAY OF OCTOBER. NOT the OF's
+   last Sunday before Advent -- a genuine EF/OF divergence, not merely a
+   citation gap. *)
 let christ_the_king y = sunday_on_or_before (mk y 10 31)
 
 let same a b = Date.compare a b = 0
@@ -253,18 +266,30 @@ let id = "ef"
 
 (* The third Sunday of September: the Ember week's anchor.
 
-   This specific date-derivation rule is one of the more contested points in
-   the 1962 calendar: pre-1955 practice tied the September Ember days to the
-   week following the Exaltation of the Holy Cross (14 Sept) instead. The two
-   rules only disagree when 1 September is a Monday -- 2025 is such a year --
-   and the primary-source scan available to this project does not contain an
-   explicit numbered-paragraph statement of either rule (searched; see
-   register §3 "Ember days"), so this citation is deliberately left at the
-   rank rules only (RG 91 entries 18/22, cited on [ember] below), not the
-   date-derivation rule itself: a wrong citation is worse than none.
-   Empirically: for 2025 this rule gives 24/26/27 September, confirmed
-   against an independent oracle; the Holy-Cross rule would give 17/19/20
-   September instead. See register §3 for the full note. *)
+   [cited] PRIMARY-SOURCE-VERIFIED (register §3a): MR1962, "De anno et eius
+   partibus", under the heading "Quatuor Tempora" (not a numbered RG
+   paragraph, which is why an earlier paragraph-number search missed it):
+
+     "Quatuor Tempora celebrantur quarta et sexta feria ac sabbato post
+     tertiam dominicam Adventus, post primam dominicam Quadragesimae, post
+     dominicam Pentecostes, post dominicam tertiam septembris."
+
+   -- the Ember Days are kept on the Wednesday, Friday and Saturday after
+   Advent III, after Lent I, after Pentecost, [and] after the third Sunday
+   of September -- confirming all four of this module's anchors, including
+   this specific contested one. This specific date-derivation rule was one
+   of the more contested points in the 1962 calendar: pre-1955 practice
+   tied the September Ember days to the week following the Exaltation of
+   the Holy Cross (14 Sept) instead. The two rules only disagree when 1
+   September is a Monday -- 2025 is such a year, and confirms the
+   third-Sunday reading empirically too (24/26/27 September against an
+   independent oracle, vs the Holy-Cross rule's 17/19/20). An earlier
+   version of this comment said the scan contained no numbered-paragraph
+   statement of either rule and left the citation at the rank rules only
+   (RG 91 entries 18/22) -- WRONG, corrected once the nominative heading
+   "Quatuor Tempora" was found rather than the genitive "Quatuor Temporum"
+   the original search used; register §3a records the correction, because a
+   false "not in the source" note is worse than no note. *)
 let third_sunday_of_september y =
   let sep1 = mk y 9 1 in
   let first_sunday = Date.add_days sep1 ((7 - weekday_index sep1) mod 7) in